JP2005115639A - Creation method and program of occurrence limit line of sediment disaster, evacuation reference line, and warning reference line, caution evacuation support system of sediment disaster - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an objective, highly repetitive and versatile creation method etc. of an occurrence limit line, an evacuation reference line and a warning reference line of sediment disaster, without requiring huge trial and error by the use of an RBF network. <P>SOLUTION: This method includes: a first process S1 in which standardization of real learning data which consists of data sets of short-term rainfall index, long term rainfall index, and sediment disaster occurrence/non-occurrence is performed by a standardization analysis section; a second process S2 in which a distinction border plane is built by splitting a two dimensional former flat surface of the short-term rainfall index and the long-term rainfall index into unit lattice areas formed by grid lines of desired spacing by a distinction border plane analysis section and by setting a radial base function on an lattice point; and a third process S3 which creates at least either of the occurrence limit line, the evacuation reference line and the warning reference line of sediment disaster by an occurrence information line creation section. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to a method of creating a sediment disaster occurrence limit line, an evacuation reference line, and a warning reference line, and in particular, using an RBF network (hereinafter sometimes referred to as “RBFN”), a sediment disaster occurrence limit line and an evacuation line. The present invention relates to a method for creating at least one of a reference line and a warning reference line, a program for executing the method, and a warning and evacuation support system using a generation limit line thereof.

  Sediment disasters (debris flow, landslides, landslides) occur every year across the country, precious lives are lost and valuable assets are destroyed. About 70% of Japan's land is mountainous, with many geologically vulnerable regions, and many geographical features such as steep topography. This is due to social conditions such as population increase to mountain streams, steep slope collapse risk points, landslide risk points), and weather conditions such as storms and rainy seasons that are prone to landslide disasters. Sediment disasters are one of the fatal natural disasters in Japan.

  There are about 520,000 dangerous places for landslide disasters nationwide, and the maintenance rate by hardware measures is as low as about 20%. In addition, it is a budget to implement hard measures for all of these dangerous places. Due to the limitations of time and time, the necessity of covering the delay of hardware measures with software measures has been recognized. The purpose of soft measures is to protect human lives from earth and sand disasters and to minimize the destruction of property. Soft measures include warnings, evacuation instructions, emergency responses according to the damage situation, and secondary actions. A function to accurately and promptly support disaster prevention is necessary, and it is required how quickly various disaster prevention information can be collected, organized and transmitted. In particular, accurate warnings and evacuation instructions are important, and these are usually based on warning and evacuation baselines set using short-term and long-term rainfall indicators, In many cases, one reference line is used in the entire region including a plurality of mountain streams, and most of the reference lines are approximated by straight lines. In other words, many conventional baselines do not consider the risk of landslide disasters (latent risk) due to different terrain factors for each slope or mountain stream, and represent complex natural phenomena by linear approximation. For this reason, there remains a problem regarding its accuracy.

  For this reason, the present inventors have conducted intensive research aiming to construct a warning and evacuation support system that supports warning and evacuation. That is, for example, a linear warning reference line for each mountain stream considering the topographical characteristics of each debris flow dangerous mountain stream, such as the steepest slope and the concentration of rainfall, etc. (Non-Patent Document 1), which can be used to search, combine, and analyze various information on a computer by combining letters, numbers, images, etc. with a map. Using the Geographic Information System (GIS) that has the function to express the results on a map, color display of debris flow risk assessment results for each mountain stream, snake curve display for each mountain stream, evacuation site information display, damage assumption display In addition, a debris flow warning and evacuation support system has been proposed which has functions of inputting and summing damage information and displaying the predicted debris flow occurrence risk determination result for the predicted rainfall (Non-patent Document 2). Moreover, the method of setting the linear breakage generation | occurrence | production limit line for every slope which considered the topographical characteristic for every slope was proposed (nonpatent literature 3).

  The present inventors further use an RBF network excellent in non-linear discrimination for the purpose of setting a high-precision occurrence limit line, etc., without performing a linear approximation of complicated natural phenomena, and using its learning function to optimize the phenomenon. Proposed a method to set the non-linear breakdown collapse limit line for each region by determining the weight of the connection between the intermediate layer and the output layer (Non-Patent Document 4), and then using the RBF network, slope factors or mountain streams A method of setting a non-linear sediment disaster occurrence limit line for each slope or mountain stream considering factors has been proposed (for example, Non-Patent Document 5). This method classifies multiple slopes or mountain streams into multiple groups based on the potential risk so that the impact of the potential risk is significant, and the data within that group determines the average potential risk of the group. A method to determine the weight of the connection between the intermediate layer and the output layer as a function of the potential risk, and to set the occurrence limit line etc. using the obtained functional relationship In addition, it is possible to set non-linear and highly accurate occurrence limit lines for individual sediment disasters.

  An RBF network is a type of neural network that is classified as a computational technique based on a brain or neural network model. A neural network has a hierarchical structure of an input layer, an intermediate layer, and an output layer, and solves computational problems. Is characterized by applying internal weights to external outputs to learn. The intermediate element constituting the intermediate layer of the neural network is also called a basis function, and an arbitrary function can be used. The RBF network uses a radial basis function (RBF) as a basis function. A feature of radial basis functions is that the response of the function decreases monotonically (or increases) with distance from the center point, and the center position, distance scale and exact shape of the radial basis functions are model parameters. is there. In the methods of Non-Patent Document 4 and Non-Patent Document 5 described above, a Gaussian function (bell-shaped function having a constant radius with a height of 1), which is a representative example of the radial basis function, is used. The parameters are the center position and radius of the radial basis function.

  In the method of setting a non-linear generation limit line or the like using such an RBF network, details are shown in Non-Patent Document 4 and Non-Patent Document 5 in order to determine the center position and radius of a radial basis function that are model parameters. As shown in the figure, it is necessary to perform a large amount of trial and error by using as a guide whether the result is consistent with the natural phenomenon. In other words, the generation limit line and the like require reproducibility that can reproduce the learning data with high accuracy and generality that can achieve the prediction that is the original purpose with high accuracy. There is no method for logically determining this parameter, and enormous trial and error is required to determine the optimum parameter. For example, if the center is adjusted to the learning data and the radius is reduced, the reproducibility is improved, but the generalization property is lowered. Even when the center is matched with the learning data, if the radius is increased, the reproducibility decreases but the generalization tends to improve. In addition, for example, when radial basis functions are set for all learning data, this means that the required memory capacity of the computer increases and calculation becomes difficult, but it does not necessarily mean that generalization is improved. Not what you want.

  For this reason, in the conventional techniques disclosed in Non-Patent Document 4 and Non-Patent Document 5, grid lines at predetermined intervals are set on the short-term rainfall index axis and the long-term rainfall index axis, respectively. By dividing the dimensional plane into unit lattice regions formed by lattice lines, and by setting a basis function on the lattice points only when the learning data exists in the four unit lattice regions formed by the lattice points, The idea is to reduce the number of basis functions. In addition, cluster lines are set at half the grid line interval to divide the two-dimensional plane into unit cluster areas that divide the unit grid area into four equal parts, and for each learning data generated / not generated for each unit cluster area. If there is more than one learning data, set one new learning data that is clustered by the centroid method and represent the plurality of learning data to equalize the density of the learning data and generalization ability Improvement of constructing a high discriminant boundary surface. Furthermore, all the basis functions are simplified in that the radius of the short-term rainfall index axis and the radius of the long-term rainfall index axis are the same circular shape and have the same radius, but it still requires a lot of trial and error. ing.

  Also, based on the data of a certain area, as a result of a huge trial and error, I tried to set the occurrence limit line etc. of other areas with different rainfall conditions using an RBF network with a radius set as a suitable parameter, etc. It has been found that a favorable result cannot be obtained, and it is necessary to perform a great deal of trial and error again.

  In other words, the technology for setting a non-linear generation limit line using a conventional RBF network has a problem that it requires enormous trial and error and lacks objectivity because it depends on the experience and intuition of the user to set it, Further, since all the basis functions are circular and have the same radius, there is a problem that sufficient reproducibility is not obtained. Furthermore, the short-term and long-term rainfall index data in the learning data is actual data having a dimension of rainfall, and if an attempt is made to set an occurrence limit line or the like in another area with different rainfall conditions, It is not possible to apply the parameters of basis functions obtained by a huge amount of trial and error, and it is necessary to determine a suitable parameter for that region by performing a huge amount of trial and error using the data of that region. There was a problem of lacking.

  In addition, the present inventors are required to set individual warning reference lines, evacuation reference lines, and occurrence limit lines in consideration of terrain factors for each slope or mountain stream. A method to evaluate the degree was proposed. For example, we propose a method using multiple discriminant analysis, which is a kind of statistical method (Non-patent Document 1, Non-patent Document 3), and calculate the landslide occurrence rate for each slope factor and calculate the occurrence rate. We proposed a method for evaluating the potential risk of individual slopes using a scoring system by adding the scores for each slope factor set in this way as the number of set points for the slope factor / category (Non-Patent Document 5). This method is very simple, does not require empirical judgment, and can automatically adjust the weight between slope factors, and it can be used for actual collapse phenomena (in relation to potential risk). , Collapse probability, or collapse tendency). The present inventors have also proposed a method of using a rough set to find a complex causal relationship between factors and extracting a combination of important factors related to the risk of occurrence of landslide disasters (Non-Patent Document 6). These are techniques that can be used effectively when the present invention is implemented as a mode for setting the occurrence limit line of landslide disasters of individual slopes or individual mountain streams in consideration of the potential risk.

Toru Takahashi and others 5: Debris flow warning evacuation reference rainfall line considering topographic characteristics, Journal of Sabo Society, Vol.53, No.1, p.35-46, 2000 Katsumi Seo and 4 others: Construction of a debris flow warning and evacuation support system using GIS -Development of a model in Oshima-gun, Yamaguchi Prefecture-, Sabo Journal, Vol.53, No.4, p.30-37, 2000 Kazumasa Kuramoto and 5 others: Research on the method of setting the critical rainfall line for landslides in consideration of slope factors on steep slopes, Journal of Japan Society of Civil Engineers, No.658 / VI-48, pp.207-220, 2000.9 Kazumasa Kuramoto et al. 5: Study on the setting of non-linear landslide generation limit rain line using RBF network, Proceedings of Japan Society of Civil Engineers, No.672 / VI-50, pp.117-132, 2001.3 Kazumasa Kuramoto et al. 5: Non-linear slope failure limit rainfall line setting method considering slope factors and its prediction accuracy, JSCE Proceedings, No.707 / VI-55, pp.67-81, 2002.6 Hiroyuki Sugawara and others 5: Research on extraction of factors causing landslides by data mining using rough sets, Journal of Japan Society of Civil Engineers, No.658 / VI-48, pp.221-229, 2000.9

The present invention is objectively set without requiring a huge amount of trial and error, and without requiring user experience and intuition, in view of the above-mentioned situation related to the technology for setting the occurrence limit line of landslide disasters using an RBF network. A method for creating a sediment-related disaster occurrence limit line, evacuation reference line, and warning reference line that is versatile and can be easily applied to other regions with different rainfall conditions, with high reproducibility of learning data, and It is an object of the present invention to provide a program for executing the method and a warning / evacuation support system using the generation limit line, the evacuation reference line, and the warning reference line.
In the present specification, the occurrence information line means a concept including an occurrence limit line, an evacuation reference line, and a warning reference line related to the occurrence of a natural disaster such as a sediment disaster.

  To achieve the above object, the invention of claim 1 comprises a hierarchical structure of an input layer having a plurality of input elements, an intermediate layer having a plurality of intermediate elements, and an output layer having one output element. Is a radial basis function (RBF) and uses an RBF network that learns the weight of the connection between multiple intermediate elements and output elements, and consists of multiple data sets of short-term rainfall indicators, long-term rainfall indicators, and occurrence / non-occurrence of landslide disasters. By learning the learning data, a discriminant boundary surface consisting of the short-term rainfall index, the long-term rainfall index, and the occurrence risk of landslide disasters is constructed. A method for creating a sediment-related disaster occurrence limit line, evacuation reference line, and warning reference line that creates at least one of an occurrence limit line, an evacuation reference line, and a warning reference line. Of the actual learning data consisting of data sets of short-term rainfall index, long-term rainfall index, and occurrence / non-occurrence of landslide disasters, using actual data with the respective rainfall dimensions as the data, the actual values of the short-term rainfall index and the long-term rainfall index A process for obtaining standardized learning data consisting of a data set of short-term rainfall index, long-term rainfall index, and occurrence and non-occurrence of landslide disasters, which are converted into standardized data by applying predetermined conversions to the data, and the standardized learning data as RBF network learning data A step of constructing a discrimination boundary surface, and a step of creating at least one of a landslide disaster occurrence limit line, an evacuation reference line, and a warning reference line standardized as contour lines corresponding to a predetermined occurrence risk of the discrimination boundary surface Is.

  The invention described in claim 2 is characterized in that, in the method of creating the sediment disaster occurrence limit line, the evacuation reference line, and the warning reference line according to claim 1, the radial basis function is a Gaussian function.

  According to a third aspect of the present invention, in the method for creating the sediment-related disaster occurrence limit line, the evacuation reference line, and the warning reference line according to the first or second aspect, the step of obtaining the standardized learning data is a short-term operation from the actual learning data. The maximum value of the actual data of the rainfall index and the long-term rainfall index is obtained, and the actual data reference value is calculated by multiplying the maximum value by a first constant that is set in advance at 1 or more. By calculating the ratio of each of the short-term rainfall index and the long-term rainfall index, which are preset by the value, by dividing the second constant, and multiplying the actual learning data for each of the short-term rainfall index and the long-term rainfall index. This is a process of performing conversion to dimension and obtaining standardized learning data using them as standardized data.

  The invention according to claim 4 is the method for creating a judgment boundary surface in the method for creating the occurrence limit line of landslide disaster, the evacuation reference line, and the warning reference line according to any one of claims 1 to 3, A step of setting a center position of the radial basis function, which includes setting grid lines at predetermined intervals on the short-term rainfall index axis and the long-term rainfall index axis, respectively, Divide the dimensional plane into unit grid areas formed by grid lines, and for each unit grid area, there are learning data for landslide disasters in the unit grid area, and radial basis at the upper right grid point forming the unit grid area Only when no function is set, a radial basis function is set at this grid point, and there is non-generated learning data in the unit grid area, and a radial basis function is set at the lower left grid point that forms the unit grid area. If not Are those wherein the step of setting the radial basis function in this grid point only.

  According to a fifth aspect of the present invention, in the method of creating the sediment disaster occurrence limit line, the evacuation reference line, and the warning reference line according to the fourth aspect, the step of constructing the discrimination boundary surface includes the occurrence of a sediment disaster for each unit lattice area. There is a step of clustering each non-occurrence learning data, and the step is preset with the upper limit number of learning data set in advance in the learning data in the unit grid area for each occurrence / non-occurrence of landslide disaster. The upper and lower limits of the connection weight suppression parameter are read from the database, and the suppression parameter for each learning data generated and not generated for each unit grid area is determined based on the number of learning data in the unit grid area. For the learning data of the unit cell area having the above learning data, a lower limit value of the suppression parameter is set, and the learning data of the unit cell area having only one learning data is set. The upper limit value of the suppression parameter is set in the data, and the upper limit value and the lower limit value of the suppression parameter are proportionally set in the learning data of the unit lattice area having the number of learning data in between. If there is multiple learning data for each occurrence and non-occurrence of learning data, the new learning data is combined with the learning data by the centroid method in the coordinate space consisting of the short-term rainfall index axis and the long-term rainfall index axis. And deleting a plurality of original learning data.

  The invention according to claim 6 is the method of creating the determination boundary surface in the method for creating the occurrence limit line of landslide disaster, the evacuation reference line, and the warning reference line according to any one of claims 1 to 5, The step of classifying the radial basis functions into a plurality of groups according to preset conditions, and assuming that the classified groups are groups of radial basis functions each having the same radius, the objective function of the RBF network is optimized And a step of optimizing the radius of the radial basis function for each group.

  The invention described in claim 7 is classified as a group consisting of radial basis functions each having the same radius to be optimized in the method of creating the sediment disaster occurrence limit line, the evacuation reference line, and the warning reference line according to claim 6. And the radius of the short-term rainfall index axis and the radius of the long-term rainfall index axis, respectively.

  The invention according to claim 8 is the method for creating the occurrence limit line of landslide disaster, the evacuation reference line and the warning reference line according to claim 6 or claim 7, wherein the step of constructing the discriminant boundary surface is a radial basis function. A step of setting the center position, and the step of setting the center position of the radial basis function is a step of classifying the radial basis functions into a plurality of groups according to preset conditions, It has identification information to identify any of the disaster occurrence area, non-occurrence area, or mixed area, and this process sets the grid lines at predetermined intervals on the short-term rainfall index axis and the long-term rainfall index axis respectively. Divide the two-dimensional plane of the rainfall index and the long-term rainfall index into unit grid areas formed by grid lines, and only if there is learning data for landslide occurrence in the unit grid area for each unit grid area Lattice area When the radial basis function is not set at the upper right grid point to be formed, the radial basis function of the generation region is identified and set at the upper right grid point, and the radial basis function is already set at the upper right grid point If this is changed to a radial basis function of the mixed region, and no non-occurrence learning data exists in the unit cell region, the radial basis function is not set at the lower left lattice point forming the unit cell region Is a process of identifying and setting the radial basis function of the non-occurrence area at the lower left grid point, and changing it to the radial basis function of the mixed area when the radial basis function has already been set at the lower left grid point. The group classification is a classification based on identification information.

  The invention described in claim 9 optimizes the radius of the radial basis function in the method for creating the sediment-related disaster occurrence limit line, the evacuation reference line, and the warning reference line according to any one of claims 6 to 8. The process includes obtaining an initial value of a radius that is a variable for optimization, setting an initial point defined by the value of the variable, a current point defined by the given value of the variable, a variable Select at least one of the following and give a step width to the selected variable and compare the tunnel destination point skipped with an objective function. If the relationship satisfies a preset condition, the tunnel destination point is In the tunnel processing step that performs the update as a point, the steepest descent processing step that calculates the sensitivity for the variable of the objective function and performs the convergence calculation to find the local point that minimizes the objective function according to the sensitivity, and the steepest descent processing step Result obtained Determined whether to end the process of optimizing the local point as the optimum point, to update the optimization calculation times and has a determining step of determining whether to continue processing the optimization back to the tunnel process.

  The invention according to claim 10 is the method for creating the sediment-related disaster occurrence limit line, the evacuation reference line, and the warning reference line according to claim 9, wherein the initial point setting step is performed by setting the radius within a preset upper and lower limit values. First random number generating means for generating an initial value is generated, a plurality of initial points are generated by the first random number generating means, and objective functions of the generated initial points are respectively obtained. This is a step of setting the smallest point as an initial point for optimization.

  The invention according to claim 11 is the method for creating the occurrence limit line of landslide disaster, the evacuation reference line and the warning reference line according to claim 9 or claim 10, wherein the initial point is a short-term rainfall index axis for each of a plurality of groups. And the radius of the long-term rainfall index axis are the initial points set to be the same.

  A twelfth aspect of the present invention is the method for creating a landslide disaster occurrence limit line, an evacuation reference line, and a warning reference line according to any one of the ninth to eleventh aspects, wherein the tunnel processing step is one of variables. The second random number generating means for selecting one variable is selected, and the step width coefficient is generated using the third random number generating means for generating a step width coefficient on the condition of preset upper and lower limit values. And the step of setting the tunnel destination point using the product of the sensitivity of the objective function at the current point for the selected variable and the coefficient of the step width as a step width.

  According to a thirteenth aspect of the present invention, in the method for generating the sediment-related disaster occurrence limit line, the evacuation reference line, and the warning reference line according to any one of the ninth to twelfth aspects, the tunnel processing step is set in advance. Using a fourth random number generating means for generating an application coefficient of an objective function that compares within the upper and lower limit conditions, a value that is inversely proportional to the number of optimization calculations is set as a condition within the preset upper and lower limit values. An upper limit is set in the fourth random number generation means, an application coefficient is generated by the fourth random number generation means, the sum of the application coefficient and 1 is set as an application constant, and the value of the objective function at the current point and the application constant are The product and the value of the objective function of the tunnel destination point are compared, and if the tunnel destination point is smaller, an annealing process step is performed for updating the tunnel destination point as the current point.

  The invention described in claim 14 is the method for creating the sediment-related disaster occurrence limit line, the evacuation reference line, and the warning reference line according to any one of claims 9 to 13, wherein The determination is based on the absolute value of the amount of change in the objective function between the optimization calculation and the previous optimization calculation, and the number of optimization calculations.

  According to a fifteenth aspect of the present invention, in the method for creating a landslide disaster occurrence limit line, an evacuation reference line, and a warning reference line according to any one of the sixth to fourteenth aspects, the radius is optimized in advance. The optimization is performed under the condition that the value is equal to or more than the set minimum value.

  According to a sixteenth aspect of the present invention, in the method for creating a sediment disaster occurrence limit line, an evacuation reference line, and a warning reference line according to any one of the eighth to fifteenth aspects, the optimization of the radius is performed in the occurrence region. In the radial basis function of the non-occurrence region, the optimization is performed under the condition that the ratio between the radius of the short-term rainfall index axis and the radius of the long-term rainfall index axis is not more than a predetermined value, respectively. Is.

  The invention described in claim 17 is the method of creating a sediment disaster occurrence limit line, an evacuation reference line, and a warning reference line according to any one of claims 1 to 16, wherein a sediment disaster occurrence limit line is set, The evacuation reference line and the warning reference line are the slopes of the entire region or the generation limit line of a mountain stream including a plurality of slopes or a plurality of mountain streams having different potential risks of occurrence of sediment disasters due to topographic factors.

  The invention described in claim 18 is the method for creating the sediment-related disaster occurrence limit line, the evacuation reference line, and the warning reference line according to any one of claims 1 to 16, wherein each slope or each mountain stream is provided. The process of calculating the risk of occurrence of landslides due to topographic factors, the process of classifying slopes or streams into multiple groups based on the magnitude of each potential risk, and the average of the slopes or streams included in each group A step of calculating a latent risk, a step of learning a plurality of connection weights using an RBF network for each group, a step of approximating the learned weights as a function of the average potential risk, and an average Calculating a connection weight corresponding to each potential risk by using an approximated function by regarding the potential risk of each as a potential risk, and using the weight to calculate each potential risk The process of constructing a discriminant boundary surface corresponding to each and the occurrence limit line of landslide disasters of individual slopes or individual mountain streams having individual potential risks as contour lines corresponding to the predetermined risk of occurrence of each individual discriminant boundary surface And a step of setting at least one of an evacuation reference line and a warning reference line.

  The invention described in claim 19 is the method of creating the earth and sand disaster occurrence limit line, the evacuation reference line, and the warning reference line according to claim 18, wherein the potential risk is a potential risk evaluated using a scoring system. Calculate the sediment disaster rate by category for each slope factor or mountain stream factor, and use that rate as the number of points set for the category, and add the points for each slope factor or mountain stream factor to the potential risk. To do.

  The invention according to claim 20 is the short-term rainfall index having a rainfall amount dimension in the method for creating the occurrence limit line of landslide disaster, the evacuation reference line, and the warning reference line according to any one of claims 1 to 19. And the long-term rainfall index data are the maximum rainfall within 3 hours from the estimated time of occurrence or the effective rainfall with a half-life of 1.5 hours and the effective rainfall with a half-life at that time of 72 hours. It is what.

  The invention according to claim 21 comprises a hierarchical structure of an input layer having a plurality of input elements, an intermediate layer having a plurality of intermediate elements, and an output layer having one output element by a computer, wherein the intermediate elements have radial basis functions (RBF). Using the RBF network that learns the connection weights of multiple intermediate elements and output elements, learn multiple learning data consisting of a dataset of short-term rainfall indicators, long-term rainfall indicators, and occurrence / non-occurrence of landslide disasters. By constructing a discriminant boundary surface consisting of short-term rainfall indicators, long-term rainfall indicators, and landslide disaster occurrence risk, the sedimentary disaster occurrence limit line and evacuation standard are contour lines corresponding to the predetermined occurrence risk of the discriminant boundary surface. A program for creating a landslide disaster occurrence limit line, an evacuation reference line, and a warning reference line for creating at least one of a line and a warning reference line, Of the actual learning data consisting of the data set of short-term rainfall index, long-term rainfall index, and landslide disaster occurrence / non-occurrence using real data with the same amount of rainfall as the data of the long-term rainfall index and long-term rainfall index, the short-term rainfall Standardized learning, a process for obtaining standardized learning data consisting of a data set of short-term rainfall indicators, long-term rainfall indicators, and occurrence / non-occurrence of sediment-related disasters, which are converted into standardized data by applying predetermined conversion to the actual data of the indicators and long-term rainfall indicators. At least one of the process of constructing the discrimination boundary surface using the data as learning data of the RBF network and the occurrence limit line of landslide disaster, evacuation reference line, and warning reference line standardized as contour lines corresponding to the predetermined occurrence risk of the discrimination boundary surface And a step of creating the above.

  The invention according to claim 22 is an input unit for reading actual learning data comprising a data set of short-term rainfall indices, long-term rainfall indices, and landslide disaster occurrence / non-occurrence data of actual data having a rainfall amount dimension from an external or actual database. The maximum value is obtained from the short-term rainfall index and the long-term rainfall index of the actual data read by the input unit, and each actual data is multiplied by a first constant preset at 1 or more. Calculate the reference values, calculate the respective ratios of the short-term rainfall index and the long-term rainfall index set in advance by the respective actual data reference values, and calculate the respective ratios for the short-term rainfall index and the long-term rainfall. A standardization analysis unit that performs transformation to be dimensionless by multiplying actual learning data of each index and calculates each standardized learning data, a first constant and a second A standardized analysis database that stores numbers, a standardized database that stores calculated standardized learning data, a discriminant boundary surface analysis unit that constructs a discriminant boundary surface using the standardized learning data as learning data for the RBF network, and a discriminant boundary surface An earth and sand disaster warning and evacuation support system having an output unit that displays, prints, or outputs information, and a discrimination boundary surface analysis unit sets grid lines at predetermined intervals on a short-term rainfall index axis and a long-term rainfall index axis, respectively. Then, the two-dimensional plane of the short-term rainfall index and the long-term rainfall index is divided into unit grid areas formed by grid lines.For each unit grid area, there is learning data on occurrence of sediment disasters in the unit grid area and A radial basis function is set at this lattice point only if the upper right lattice point that forms the unit lattice area is not set, and the unit lattice area A radial basis function center position setting unit that sets a radial basis function at this lattice point only when non-generated learning data exists and a radial basis function is not set at the lower left lattice point forming the unit lattice region; The radial basis functions are classified into a plurality of groups according to preset conditions, and the classified groups are groups of radial basis functions each having the same radius, and the objective function of the RBF network is grouped as an objective function for optimization. And a radial basis function radius optimizing unit that optimizes the radius of each radial basis function.

  The present invention uses RBFN, does not require enormous trial and error, can be set objectively without requiring user experience and intuition, has high reproducibility of learning data, and other regions with different rainfall conditions It is possible to create a landslide disaster occurrence limit line, evacuation reference line, and warning reference line with versatility that can be easily applied to the landslide disaster database. It is possible to respond to disaster countermeasures, etc., and it can show great benefits for national land conservation and lifesaving.

  Embodiments of the present invention will be described below. As described above, the present invention is a vast amount of technology that relies on the experience and intuition of the user to determine suitable parameters for the RBF network, even after various improvements and improvements as described above. The present inventor has conducted intensive research as a basic problem to be solved for the problem that RBFN with the parameters obtained as a result of trial and error is not applicable to other regions with different rainfall conditions. As a result of overlapping, a suitable method for solving the problem has been invented.

  In other words, as a result of detailed examination, depending on the data size of the short-term rainfall index and the long-term rainfall index of the learning data, suitable values such as the radius of the basis function, the center position of the basis function, the upper and lower limits of the connection weight suppression parameter, etc. It has been found that the value can be changed, and it has been found that these basic problems can be solved by using standardized learning data obtained by standardizing short-term and long-term rainfall index data as learning data. In the first embodiment, the data set of short-term rainfall index, long-term rainfall index, and occurrence / non-occurrence of landslide disasters, using actual data with the dimension of rainfall as the short-term rainfall index and long-term rainfall index data, respectively. Among the actual learning data, the short-term rainfall index and the long-term rainfall index and the long-term rainfall index are converted into standardized data by performing predetermined conversion on the actual data. A step of obtaining standardized learning data comprising a data set of rainfall indicators and occurrence / non-occurrence of sediment disaster, a step of constructing a discriminant boundary surface using the standardized learning data as learning data of the RBF network, and a predetermined occurrence of the discriminant boundary surface And a step of creating at least one of a landslide disaster occurrence limit line, an evacuation reference line, and a warning reference line standardized as contour lines corresponding to the degree of danger.

  This embodiment will be described with reference to FIG. FIG. 1 is a flowchart showing a method for creating a sediment disaster occurrence information line according to an embodiment of the present invention. Step S1 is a process of obtaining the above-described standardized learning data, Step S2 is a process of constructing a discrimination boundary surface, and Step S3 is an occurrence information line of landslide disasters, that is, an occurrence limit line, an evacuation reference line, and a warning reference line. It is a step of creating at least one of them. In addition, the process of outputting those analysis results is provided as step S4 after that.

The actual learning data 2 provided to the standardized learning data 7 is a collection of actual data, but may be input during standardization or stored in the actual database 1 as shown in FIG. May be read out. Further, without including the step S1 for obtaining the standardized learning data 7, the standardized learning data 7 may be input as the learning data of the RBF network in the step S2 for constructing the discriminant boundary surface, or separately calculated. The standardized learning data 7 may be stored in the standardized database 6 and read out.
In addition, as a program executed on a computer, measured rainfall data and data on occurrence / non-occurrence of sediment disasters are used as input data for a system including a computer, and a short-term rainfall index having a rainfall dimension from the input data. Preferably, the method includes a step of calculating actual data of a long-term rainfall index, storing the actual data as actual learning data in a database, and reading the actual learning data from the database. Alternatively, when actual learning data is already obtained, it is preferable that the input can be performed as a step of collecting or reading out actual learning data of a system including a computer. That is, since rain data may exist in various forms, it is desirable that a system including a computer that implements the present invention be configured so that the various forms of rain data can be processed as input data. .
In the claims and specification of the present application, the actual data means the raw data of the phenomenon that actually occurred, and the actual learning data means the actual data used as the learning data of the RBF network. There is no general difference. Also, the standardized data and the standardized learning data have the same relationship and there is no substantial difference.

  Although the present invention is not limited as the radial basis function of RBFN, it is preferable to use a Gaussian function that is also a representative example of the radial basis function, and when other functions are used, the obtained discrimination boundary surface is Degraded more than Gaussian function. For example, when a quadratic function is used, a good accuracy can be obtained in the average part (inner center part) of the data set, but the base part is adversely affected due to the shape of the function, Deterioration that a good approximation cannot be performed as a whole occurs.

  The present invention is preferably implemented in a form further comprising a step of reversely converting the standardized sediment disaster occurrence limit line obtained as described above. In other words, the occurrence limit line expressed as a function of the standardized short-term rainfall index and long-term rainfall index is subjected to the inverse transformation of the predetermined transformation applied to the actual data having the rainfall dimension, and the same dimension as the actual data It is preferable to implement as a mode having a step of obtaining a sediment-related disaster occurrence limit line expressed as a function of a short-term rainfall index and a long-term rainfall index. The conventional debris disaster occurrence limit lines, etc. are occurrence limit lines expressed as a function of the short-term and long-term rainfall indices, all of which have the same dimensions as the actual data. Continuity with the limit line can be secured, and it can be used with the same feeling as before. It should be noted that this step is preferably configured to be performed in a system including a computer that implements the present invention.

Next, a preferred embodiment of step S1 for obtaining standardized learning data will be described with reference to FIG. In step S1, actual data is first selected (step S11). This selection includes two concepts: when actual data is input as described above, and when actual learning data 2 is stored in advance in the actual database 1 and read out.
Next, in step S12, a standardized data reference value is selected. The standardized data reference value is used for standardization and may be any value, and is not intended to limit the present invention, but a good value such as 1, 10, 100 is desirable. Is easy to handle and display. In this embodiment, the standardized data reference values are SSmmax and SLmmax for the short-term rainfall index and the long-term rainfall index, respectively.
These standardized data reference values may be input, or may be stored as a standardized data reference value table 4 in the standardization analysis database 3 as shown in FIG. The standardized data reference value table 4 stores several numbers selected in advance as a set.
Step S13 is a step of calculating conversion coefficients for standardization. For example, (1) Find the maximum value (RSmax, RLmax) of the actual data of the short-term rainfall index and the long-term rainfall index from all the actual learning data, and (2) Each actual data reference value (RSmmax, RLmmax) is calculated by multiplying by a predetermined constant of 1 or more, and an operation for dividing the standardized data reference value selected in the previous step S12 by the actual data reference value is performed. The coefficient obtained in this way is the standardized conversion coefficient 5.

That is, standardized conversion coefficients for converting actual data are SSmmax / RSmmax and SLmmax / RLmmax, respectively. The predetermined constant is a constant that is multiplied by the maximum value of the actual data in order to obtain a reference value of the actual data, but is not limited to the learning data used for analysis, It is preferable to decide to exceed approximately 10%. It can be determined based on the past maximum actual data rounded, but when using the present invention for many areas, it is determined based on 1.1 times the past maximum actual data. It is preferable to carry out. If this constant is too large, the learning data concentrates only near the origin of the coordinate axes, and the area where no learning data exists increases, causing problems such as an increase in computer capacity and processing burden. This predetermined constant may also be input at the time of analysis, or may be set in advance as 1.1, 1.2, etc., stored in the standardized analysis database 3, etc., and read out.
The standardized conversion coefficient 5 is stored in the standardized analysis database 3 in the present embodiment, but may be stored separately by providing a database.
Step S14 is a process in which standardized data is calculated. In this process, the standardized conversion coefficient 5 calculated in step S13 is read from the standardization analysis database 3, and is made dimensionless by obtaining the product of this and the short-term rainfall index data or long-term rainfall index data of the actual learning data. Perform conversion.
In step S15, the standardized learning data 7 obtained in step S14 is stored in the standardized database 6.

Next, step S2 for constructing a discrimination boundary surface as learning data of the RBF network using the standardized learning data 7 analyzed in step S1 will be described with reference to FIG.
First, a preferred embodiment of the step of setting the center position of the radial basis function in step S21 will be described.
In the implementation of the present invention, the standardized learning data 7 related to the short-term rainfall index and the long-term rainfall index is read from the standardized database 6, and grid lines at predetermined intervals are set on the short-term rainfall index axis and the long-term rainfall index axis, respectively. The two-dimensional plane of the short-term rainfall index and the long-term rainfall index is divided into unit grid areas formed by the grid lines, and for each unit grid area, (1) learning data on occurrence of sediment disasters (hereinafter simply “ The basis function is set at the lattice point only when the upper right grid point forming the region is present and the basis function is not set. A basis function is set at a lattice point only when generation learning data (hereinafter, sometimes simply referred to as “non-occurrence data”) exists and a basis function is not set at the lower left lattice point forming the region. Preferably carried out by a procedure such.

FIG. 2 is a conceptual diagram showing a method for setting the center position of a radial basis function based on learning data generated / not generated in such a unit cell region.
In FIG. 3, the vertical axis 11 represents the hourly rainfall as the short-term rainfall index, the horizontal axis 12 represents the effective rainfall as the long-term rainfall index, and the non-occurrence rain data 14 as white circles are generated as black squares. The rainfall data 14 is plotted.
FIG. 4 is a conceptual diagram showing how the basis function 13 is set by plotting learning data (rainfall data 14) generated / not generated by dividing this into unit grid regions.

  The predetermined interval related to the setting of the grid line means a predetermined constant interval that is set in advance. For example, the predetermined interval is determined by setting the number of divisions with the reference value of the standardized data as an upper limit. This is desirable because it is easy to understand in terms of image. The number of divisions in this case is not intended to limit the present invention, but is preferably about 20 to 60. If the number of divisions is too small, the reproducibility of the learning data is reduced. Although the processing burden increases, generalization is not necessarily improved.

  Although the method for setting the center position of the basis function is different from the above-described prior art (Non-Patent Document 4 or the like), similarly, the number of basis functions can be reduced. Furthermore, in this setting method, non-occurrence data can be evaluated on the inside, that is, on the safe side, and generated data can be evaluated on the outside, that is, on the dangerous side. In the present invention, it goes without saying that the center position of the basis function may be set by a method similar to the above-described conventional technique.

  Next, a preferred embodiment of the clustering analysis step in step S22 will be described. As described in the above prior art (Non-Patent Document 4, etc.), the learning data clustering is performed by equalizing the density of the learning data, that is, the dense learning data is obtained by averaging the new learning data. Instead, it is performed in order to construct a discriminating boundary surface having a high generalization ability, and it is desirable to perform this also in the implementation of the present invention. In the clustering, it is necessary to calculate a connection weight suppression parameter, which will be described first.

  In general, the RBFN is a self-suppression term consisting of the product of the square of the difference between the output value of the RBFN and the teacher value (that is, occurrence or non-occurrence of landslide disasters, for example, 0 and 1 respectively), the connection weight, and the suppression parameter. One of the features of RBFN is that it has a self-suppression term, with learning that minimizes the objective function. There are several meanings of providing a self-suppression term, one of which is that the self-suppression term has the effect of making the generalized inverse matrix square, and the number of learning data is the number of basis functions (ie It is possible to give an answer even when the weight is less than the number of connection weights to be learned, and the second is that the superiority of each basis function can be adjusted. The above prior art and the clustering of the present invention utilize the second property that the superiority of each basis function can be adjusted.

Specifically, this will be described with reference to FIG. FIG. 5 plots the rainfall data 14a to 14d, with the vertical axis 11 representing the hourly rainfall as the short-term rainfall index and the horizontal axis 12 representing the effective rainfall as the long-term rainfall index.
Further, the unit grid area divided when setting the center position of the basis function 13 is used, and the upper limit number of learning data set in advance for each of the generated / non-generated learning data (rainfall data 14a to 14d) in the area Using the upper limit value and lower limit value of the suppression parameter, for each learning data generated / non-occurring for each region, (1) based on the number of learning data in the region, learning data equal to or greater than the upper limit number The lower limit value of the suppression parameter is set for the learning data of the area having the upper limit value of the suppression parameter for the learning data of the area that is only one, and the upper limit value of the suppression parameter is set for the learning data of the area in between. The lower limit value is set by proportional distribution. (2) When there are a plurality of learning data, clusters 16a to 16d are set as shown in FIG. The new learning data (cluster representative points 15a to 15d) obtained by clustering the learning data (rainfall data 14a to 14d) by the center of gravity method is set, and a plurality of the original learning data is deleted. be able to.
Note that the upper limit number of learning data set in advance may be different between generated data and non-generated data, and may be set to the same value.

  For generated data, the number of data is usually significantly smaller than non-generated data, so the target generation limit line may be set even if the suppression parameter is a fixed value that does not depend on the number of data. In some cases, fixing the generated data suppression parameter is preferable because the parameter can be reduced. Note that the upper limit number of learning data may be obtained, for example, by obtaining the maximum number of learning data existing in the region, but the number of non-occurrence data is usually very wide, from 0 to tens of thousands. Yes, using the maximum number is not preferable in terms of the distribution of self-suppression terms. Considering the appropriate distribution of self-suppression terms, it is set to about 50 to 200, especially about 100, regardless of the actual maximum number of data in the region. It is preferable to do this.

  By this clustering, there is at most one occurrence data and no occurrence data in each unit lattice area, the density of learning data is equalized, and a discrimination boundary surface with high generalization ability can be constructed. It is possible to set the occurrence limit line of a highly generalized sediment disaster using this. In the present invention, clustering may be performed by a method similar to the above-described conventional technique.

  The setting of the center position of the basis function and the clustering of the learning data may be performed at the actual learning data stage or the standardized learning data stage. In particular, the standardization is performed first, and then the basis It is preferable to set the center position of the function and cluster the learning data. As a result, except for standardization processing, it is possible to perform unified processing including other regions, and it becomes possible to more easily confirm and examine the relationship and consistency.

As described above, according to the first embodiment using the standardized learning data, the RBFN parameters such as the radius and center position of the basis function and the upper and lower limit values of the connection weight suppression parameter are initially tested as in the prior art. It is necessary to obtain a suitable value in advance by error, but the value obtained by trial and error is based on the standardized learning data, and the predetermined procedure of the present invention is applied even in regions with different rainfall conditions. By using the learning data standardized by the above, it can be used as a generally suitable parameter. That is, according to the present embodiment, since the RBFN parameter know-how obtained previously can be used, it does not require enormous trial and error, does not require user experience and intuition, and is objective. It is possible to set the occurrence limit of sediment disasters with high reproducibility. Therefore, it also has versatility that can be easily applied to other areas with different rainfall conditions.
It should be noted that RBNF parameters such as the basis function radius, center position, and connection weight suppression parameter upper and lower limit values obtained in advance by trial and error may be stored in or obtained from the standardization analysis database 3. The value may be input at the time of analysis.

  Next, returning to FIG. 1, the radial basis function radius optimization step of step S23 according to the present embodiment will be described. Although not shown, this step S23 uses the above-described standardized learning data, further classifies the basis functions into a plurality of groups according to a predetermined condition, and the classified groups are composed of basis functions each having the same radius. As a group, the objective function of RBFN is used as an optimization objective function, and the radius of the basis function for each group is optimized. The optimization objective function is an RBFN objective function as described above, which is an objective function for minimizing it. Hereinafter, it may be referred to as “energy”.

  Although it is theoretically possible to set all the radiuses of the basis functions as independent optimization variables, the number of variables becomes enormous and it is practically impossible to find a solution. Even if the solution is obtained, there is no theoretical specificity that each basis function has independently as a natural phenomenon, and the generalization ability does not necessarily improve.

  Therefore, in the present invention, the basis functions are classified into a plurality of groups according to a predetermined condition set in advance, and the classified groups are assumed to be groups composed of basis functions having the same radius. The classification according to the predetermined condition preferably includes a basis function in a generated data area, that is, a very dangerous area, and a non-generated data area, that is, 3 groups of basis functions in extremely safe areas and areas in which generated data and non-generated data are mixed, that is, areas that require careful handling. It is preferable that the radius of the short-term rainfall index axis and the radius of the long-term rainfall index axis, that is, the radius optimization variable is set to 6, respectively. It goes without saying that the optimization using these six variables can greatly improve the reproducibility of the learning data as compared with the conventional technique in which trial and error is performed using the above-described one variable.

  The classification of the basis function according to the preset condition is not limited to this. For example, the classification condition of the generation area is set to an area where generation data of a predetermined ratio or more exists, and the classification condition of the non-generation area is set to the predetermined condition. This may be implemented by a classification method in which non-occurrence data exceeding the ratio is present and the others are mixed regions. In the case of this classification, for example, in the case of so-called unpredictable data that suddenly occurs in a highly safe area due to, for example, accumulated rain conditions in the past, the influence is excluded, and the data area is It can be classified as a non-occurrence area.

  The grouping of the basis functions is preferably performed in the step S21 for setting the center position of the basis functions described in the present embodiment. That is, the above-described step S21 for setting the center position of the basis function is also a step of classifying the basis functions into a plurality of groups according to a predetermined condition, and the basis functions are respectively a sediment disaster occurrence region, a non-occurrence region, or a mixed region. (1) For each unit grid area described above, (1) A basis function is set at the upper right grid point that forms the area only if the generated data exists in the area. If not, identify and set the basis function of the generation region at the lattice point. If a basis function has already been set, change it to a mixed region basis function, and (2) do not occur in the region Only when there is data, if no basis function is set at the lower left grid point that forms the area, the basis function of the non-occurrence area is identified and set at that grid point. If this is mixed Changing the basis function area is preferably carried out by a procedure such.

  Next, the preferred embodiment of the radial basis function radius optimization step S23 will be described more specifically. FIG. 6 is an overall flowchart showing the overall processing flow. (1) An initial point search for obtaining an initial value of a radius as an optimization variable and setting an initial point defined by the value of the variable. Optimize the step S231, (2) the current point defined by the value of the given variable, and the tunnel destination point that skipped the value by selecting at least one of the variables and giving the step width to the selected variable If the relationship satisfies a predetermined condition set in advance, the tunnel processing step S232 for updating the tunnel destination point as the current point, and (3) sensitivity to the objective function variable (objective) A function that is partially differentiated with respect to the variable), and a result obtained in the steepest descent process step S233 for performing a convergence calculation for obtaining a local point that minimizes the objective function according to the sensitivity, and (4) the result obtained in the steepest descent process step Judgment The determination process S234 for deciding whether to end the optimization process with the local point as the optimal point or to update the number of optimization calculations and return to the tunnel process and continue the optimization process, and four sub-processes Have.

  In general, the optimization problem is that if the value of the objective function forms a monotonous peak or valley in relation to the point specified by the value of the variable, the optimal point depends on the sensitivity of the objective function to each variable. That is, the optimum variable combination can be obtained quickly and efficiently by the steepest descent method. On the other hand, the optimization problem of the present invention, as a result of detailed examination of the present inventors, it is clear that the multi-modality, that is, the multi-modality that there are a plurality of local points where the objective function shows a minimum value, It became clear that the simple steepest descent method cannot be optimized. That is, the optimization process of the present invention is characterized by having the above-described tunnel processing step S232 in order to solve this problem.

Below, it is classified into 3 groups of occurrence area, non-occurrence area and mixed area, and the optimization of 6 variables, with the radius of short-term rainfall index axis and the radius of long-term rainfall index axis of each group as optimization variables As an example, preferred specific examples of the sub-steps S231 to S234 of the radial basis function radius optimization step S23 will be further described.
In the following, in FIGS. 7 to 10, X means a radius that is a variable for optimization, X ₁ and X ₂ are radiuses of basis functions in non-occurrence regions, and X ₃ and X ₄ are radii of basis functions in generation regions. , X ₅ and X ₆ indicate the radius of the basis function of the mixed region, the former (the suffix is an odd number) indicates the radius of the long-term rainfall index axis, and the latter (the suffix is an even number) indicates the radius of the short-term rainfall index axis.

FIG. 7 is a flowchart showing a specific example of the initial point search step S231, in which a first initial radius value is generated on the condition of a predetermined lower limit value / upper limit value (X _min , X _{max in the} figure). Random number generation means (random in the figure), a plurality of initial points (k in the figure) are generated by the random number generation means, and the energy of the generated initial points (E in the figure) is obtained respectively. In this embodiment, the point at which the energy value is the smallest is selected (min in the figure), and that point is set as the initial point for optimization. In this step, as shown in the figure, the optimization calculation counter (j in the figure) is also initialized.

The initial point search step S231 in FIG. 7 includes the short-term rainfall index axis radius (X ₂ , X ₄ , X _{6 in the} figure) and the long-term rainfall index axis radius (X ₁ , X _{3 in the} figure) for each of the three groups. , X ₅ ), that is, as a feature that is set as a circular basis function, and is not intended to limit the present invention, it is easy to escape from the local solution by this setting. Thus, optimization can be performed efficiently. Generating random numbers within the predetermined upper and lower limits of the initial value is also effective for facilitating escape from the local solution and performing optimization efficiently.

The upper and lower limit values and the number of initial points of the initial value can be used in other areas where the rain conditions are different by using the learning data standardized by the predetermined procedure of the present invention. The upper and lower limit values and the number of initial points of the initial value may be obtained and set as appropriate by trial and error in advance and stored in the standardized database 6 or the like.
FIG. 7 shows an example in which the initial point of optimization is selected by comparing the energy values at the set initial point, but after each optimization calculation is performed a predetermined number of times. The multi-start point method can be selected by comparing the results. Alternatively, the initial point may be set by setting a fixed initial point without generating a random number.

Next, the tunnel processing step S232 will be described. FIG. 8 is a flowchart showing a specific example thereof, and shows an embodiment in which the processing is started by obtaining the energy of the current point and the sensitivity to each variable of the energy. That is, the second random number generation means selects one variable to be subjected to tunnel processing, and the third random number generation means generates a condition within a lower limit value 0 and a predetermined upper limit value (α in the figure). The tunnel destination point is set with the step width as the product of the step width coefficient and the sensitivity of the energy at the current point for the selected variable. However, the first constraint is that the tunnel destination value of the selected variable is greater than or equal to a predetermined minimum value (ζ in the figure). If the selected variable is outside the mixed area, its short-term rainfall index The ratio between the radius of the axis (X ₂ and X _{4 in the} figure) and the radius of the long-term rainfall index axis (X ₁ and X _{3 in the} figure) is less than the predetermined value (ξ in the figure). The tunnel destination point is set as a constraint condition for. These first constraint condition and second constraint condition are pre-set conditions.

Subsequently, the energy of the tunnel destination point is obtained, and the upper limit value (β in the figure) of the fourth random number generating means is optimized using predetermined lower limit / upper limit values (β _min and β _{max in the} figure). The sum of the application coefficient generated by the fourth random number generation means having the upper limit value and 0 as the lower limit value and the sum of 1 is used as an application constant. The embodiment includes an annealing process in which the product is compared with the energy of the tunnel destination point, and the tunnel destination point is updated when the tunnel destination point is smaller.

  That is, in this step S232, (1) only one variable to be tunneled is selected with a random number, and (2) the tunnel method is applied, and the product of the step width coefficient and sensitivity generated by the random number is used as the step width. By setting the tunnel destination point by skipping the value of the variable, (3) applying the annealing method, the application coefficient generated by a random number having a lower limit value 0 and an upper limit value inversely proportional to the number of optimization calculations is applied. This is a process of using as the amount of getter, comparing the product of the energy of the tunnel destination point, the energy of the current point and (1 + getter amount), and performing the process of updating the current point only when the tunnel destination point is small, Is to search for local points over a wide range, compare them, and gradually narrow down the search range along with the number of optimization calculations to find an optimal solution. The optimization problem of the present invention having multimodality Solves extremely effectively Rukoto is a method that can.

  However, the present invention is not limited to this, and can be implemented using other methods that can solve the optimization problem having multimodality. For example, an annealing process may not be included, a wide range of local points may be searched to rank several higher ranks, and an optimal point may be searched on condition that the order does not change a predetermined number of times. It is also possible to search a wide range of local points and rank several higher ranks, and then search for an optimum point only in the vicinity of the several higher ranks. Alternatively, a well-known GA method may be used as an effective method for the optimization problem having multimodality, but there is a problem that an enormous amount of time is required for analysis.

It should be noted that the upper limit value of the step width coefficient of this process, the condition for limiting the tunnel destination value (radius) of the variable, the upper and lower limit values of the application coefficient, and the like need to be initially determined by trial and error. The result can be used in other areas where the rain conditions are different by using the learning data standardized by the predetermined procedure of the present invention.
Conditions for restricting the upper and lower limit values of the step width coefficient, the tunnel destination value (radius) of the variable, and the upper and lower limit values of the application coefficient are appropriately obtained and set in advance by trial and error, and stored in the standardized database 6 or the like. Good.

  Next, the steepest descent process step S233 will be described. As described above, the steepest descent process step S233 is an effective method when the value of the objective function forms a monotonous peak or valley in relation to the point defined by the value of the variable. A specific example of the preferred embodiment is shown in the flowchart of FIG. This method uses the given current point as a base point to find the sensitivity to the objective function variable, and finds a local point that minimizes the objective function according to the sensitivity, that is, finds a local point near the given current point. This is a well-known method for performing the convergence calculation, and does not particularly limit the present invention, and a detailed description thereof will be omitted.

In this step, as shown in FIG. 9, the first constraint is that all the radii are equal to or greater than the minimum value (ζ in the figure). That the ratio between the radius of X (X ₂ and X _{4 in the} figure) and the radius of the long-term rainfall index axis (X ₁ and X _{3 in the} figure) is less than the predetermined value (ξ in the figure), respectively. As a constraint, it is preferable to perform a convergence operation.

Next, the determination step S234 for optimization processing will be described. A specific example of the preferred embodiment of the present invention is shown in the flowchart of FIG. The determination performed in this determination step is not intended to limit the present invention, but as shown in FIG. 10, the absolute value of the amount of change in the objective function between the current optimization calculation and the previous optimization calculation (in the figure) , Abs) is equal to or greater than a predetermined value (ε in the figure) and the number of optimization calculations is equal to or less than a predetermined allowable number (j _max ). it can.

  As described above, according to the present embodiment, which includes the step S23 for using the standardized learning data and further optimizing the radius of the basis function, the occurrence limit of the high-precision landslide disaster in which the reproducibility of the learning data is greatly improved. Lines etc. can be set. In addition, the process of optimizing the radius reduces the number of parameters that must be trial and error based on the user's experience and intuition. The RBFN parameters such as the upper and lower limit values of the weight suppression parameter can be obtained extremely easily, and can be set more suitably.

  In the present embodiment, the step of optimizing the radius of the radial basis function using the standardized learning data has been described. However, the step of optimizing the radius of the radial basis function using the actual learning data may be used. .

  In such a form, except that actual learning data is used, the step of classifying the basis functions into a plurality of groups and the step of optimizing the radius of the basis function for each of the classified groups are as described above. Since this can be carried out in the same manner as in the second embodiment, the description thereof will be omitted, but it is possible to set a high-precision landslide disaster occurrence limit line etc. that greatly improves the reproducibility of the learning data. In addition, since it has a process of optimizing the radius, the number of parameters that must be trial and error dependent on the user's experience and intuition is reduced, and the center position of the basis function to be determined by trial and error and the connection The RBFN parameters such as the upper and lower limit values of the weight suppression parameter can be obtained extremely easily, and further, it can be set more suitably.

Next, returning to FIG. 1, the basis function synthesis step shown in step S24 will be described. After the parameters of the respective basis functions are determined up to step S23, these basis functions are synthesized.
An example of the radial basis function is a Gaussian function as described above, and this Gaussian function has a shape as shown in FIG. This Gaussian function is arranged at a lattice point, and the non-occurrence height is set to 1, and the occurrence height is set to −1, and these basis functions are synthesized to form a surface while learning. This is the discrimination boundary surface. An example of this discrimination boundary surface is shown in FIG.

In FIG. 12, the Z-axis 18 shows the height of the Gaussian function calculated as 1 for the non-occurrence case and -1 for the occurrence, and the X-axis 19 shows the effective rainfall as a long-term rainfall index. The Y-axis 20 shows the hourly rainfall as a short-term rainfall index.
A discrimination boundary surface 17 is drawn in the three-dimensional coordinates represented by these axes. The discrimination boundary surface 17 is obtained by combining the respective Gauss functions existing at the lattice points as surfaces as described above. Since non-occurrence is 1 and occurrence is -1, for example, the probability of non-occurrence of landslide disasters is high at locations where this surface is high, and the probability of occurrence is high at locations where it is low.
Step S24 is a step of obtaining such a discriminant boundary surface 17 by synthesizing basis functions.
Next, in FIG. 1, the discriminant boundary surface analysis data 9 obtained by analyzing the discriminant boundary surface including the discriminant boundary surface 17 obtained in step S24 is stored in the discriminant boundary surface analysis database 8. S25.
Further, step S3 is a step of creating a generation information line using the obtained discrimination boundary surface 17 or the discrimination boundary surface analysis data 9. This embodiment of the generation process S3 of the occurrence information line such as earth and sand disaster will be described in detail below.

  In the present invention, it may be implemented as a form in which generation information lines such as slopes of the entire region or mountain stream generation limit lines are set, including a plurality of slopes or mountain streams having different risks of occurrence of sediment disasters due to topographic factors. Moreover, it can also be implemented as a form in which occurrence information lines such as a sediment disaster occurrence limit line for each individual slope or each mountain stream are set.

  Neither the setting of the occurrence limit line or the like of this region nor the setting of the individual occurrence limit line or the like is limited to the detailed embodiment. Reference 4 and individual occurrence limit lines, for example, can be set by the method described in the above-mentioned Non-Patent Document 5, etc., so detailed description will be omitted. Similar to the prior art disclosed in Non-Patent Document 4, an intermediate element and an output element are learned by learning a plurality of learning data composed of a data set of a short-term rainfall index, a long-term rainfall index, and occurrence / non-occurrence of landslide disasters. The discriminant boundary surface consisting of the short-term rainfall index, the long-term rainfall index, and the occurrence risk of sediment-related disasters is constructed, and the sediment as the contour line corresponding to the predetermined risk of occurrence of the discriminant boundary surface is learned. disaster It can be set or generated limit line, evacuation baseline vigilance reference line.

  On the other hand, the individual occurrence limit lines and the like are classified into a plurality of groups based on the magnitude of the potential risk, as in (1) individual slopes or individual mountain streams, as in the prior art disclosed in Non-Patent Document 5. ) For each group, the weight of the connection between the intermediate element and the output element is learned by learning the learning data of the group, and (3) the connection weight learned for each group has a group average potential risk. Consider the weight learned for each individual slope or mountain stream and approximate it as a function of potential risk, for example, straight line approximation. (4) Based on the approximate function relationship, the potential of individual slope or individual mountain stream The connection weight corresponding to the risk level is obtained, (5) the individual judgment boundary surface of the individual slope or individual stream is constructed using the weight, and (6) the predetermined occurrence risk level of the individual judgment boundary surface is corresponded. And generating a limit line of landslides individual slope or individually streams as high line, evacuation reference line, it is possible to set the warning reference line.

  When the radius is optimized when setting the individual generation limit lines and the like, the optimization can be performed in various forms. That is, for example, for all slopes or mountain streams, that is, all learning data is set as one RBFN learning data, and the energy is optimized as an objective function, and the optimized radius is applied to all groups. It can be implemented in the form.

  In addition, for each group classified based on the degree of potential risk, the learning data for each group is used as RBFN learning data for each group, and the energy for each group is optimized independently as an objective function, and the optimized radius is set for each group. It is also possible to implement in a form that applies only to the group. In this case, it is necessary to ensure the uniformity between groups and to construct an individual discrimination boundary surface corresponding to the potential risk. For example, a basis function with a center position set for all learning data is used. Used in all groups, approximates the weight of the connection learned for each group as a function of the group average potential risk, and approximates the optimized basis function radius for each group as a function of the group average potential risk. Based on the approximate functional relationship, the weight and radius of the connection corresponding to the potential risk of the individual slope or individual stream can be obtained, and the individual judgment boundary surface of the individual slope or individual stream can be constructed using the weight and radius. Consideration such as making it necessary is necessary.

  In addition, in the form in which the radius is optimized independently for each group, the inside of the group may be optimized as the same radius, or a plurality of regions having the same radius, for example, as described above Although it is possible to classify and optimize in the generation / mixture area, it is preferable to divide the area because the accuracy can be further improved.

  The optimization of the radius when setting individual occurrence limit lines, etc. is also performed by using the learning data for each group as the learning data for RBFN for each group and the sum of the energy for each group as the objective function. It can also be implemented. In other words, the RBFN for each group is used, and the optimization objective function is set as the sum of the energy of RBFN for each group, so that optimization can be performed for all groups. Even in this case, as described above, it is necessary to consider the uniformity of the groups so that an individual discrimination boundary surface corresponding to the potential risk level can be constructed. It may be optimized, and it is also possible to classify and optimize into a plurality of regions having the same radius.

  Sediment disaster occurrence limit line, evacuation reference line, and warning reference line can be set as contour lines of the discriminant boundary surface corresponding to each specified occurrence risk level. , Occurrence limit line> Evacuation reference line> Warning reference line. This order means the order of physical risk, and does not necessarily mean the magnitude of the value of the risk of occurrence on the discrimination boundary surface. For example, as described above, when 0 and 1 are used for the teacher value used in the learning performed in RBFN, that is, for the occurrence and non-occurrence of earth and sand disaster, respectively, the smaller the value, the higher the physical occurrence risk Means big. The method of using 0 for non-occurrence and 1 for non-occurrence is, as described in Non-Patent Document 5 described above, the region that greatly exceeds the rainfall conditions of sediment-related disasters that occurred in the past, It is clear that it is a very dangerous area even if there is no learning data), but unless there is learning data, the basis function cannot be set, and the output value of RBFN must be zero It is an effective way to avoid. The obtained occurrence limit line or the like is usually expressed as a curve on a two-dimensional plane with the short-term rainfall index and the long-term rainfall index as orthogonal axes.

  Next, an embodiment relating to the evaluation of the latent risk will be described using a slope as an example. As is well known, slope factors can be broadly classified into topographic factors, geological / soil factors, environmental factors, or earthquake factors. For each major category, for example, the topographic factors include slope, slope height, It is subdivided into slope orientation, slope shape, crossing shape, and emergency line, and the potential risk is obtained as an evaluation based on these factors. As a specific evaluation method, for example, well-known multivariate analysis, fuzzy theory, or a score system can be used, and the present invention is not limited. The evaluation used is preferred. For example, calculate the sediment disaster occurrence rate by category for each factor, use that rate as the number of points set for the factor / category, and add the scores for each factor set in this way to create the potential hazards for individual slopes. It is preferable to evaluate the degree. This method is very simple, does not require empirical judgment, and can automatically adjust the weight between factors. The actual collapse phenomenon (the probability of collapse in relation to the potential risk) Or the tendency to collapse).

  Although the present invention is not limited in evaluating the potential risk, as disclosed in the above-mentioned Non-Patent Document 6, using rough sets, the record of occurrence / non-occurrence of landslide disasters as slope factors. It is preferable to analyze the relationship, find a complex causal relationship between factors, extract a combination of important factors related to the risk of occurrence, and evaluate the potential risk based on the extracted combination of important factors.

Generation information lines such as the generation limit line, the evacuation reference line, and the warning reference line set as described above are output to the output unit 10 in step S4 shown in FIG. The output unit 10 is a concept including a component that transmits a data signal to another arithmetic device such as a display device such as a liquid crystal display, a printer, or a computer.
The output unit 10 may output not only occurrence information lines such as an occurrence limit line, an evacuation reference line, and a warning reference line, but also data used for analysis as well as a discrimination boundary surface as an analysis result.

  As a combination of actual data of a short-term rainfall index and a long-term rainfall index having dimensions that can be suitably used in the present invention, according to the inventors' energetic research, as already shown in the drawings, the short-term rainfall index A combination of a maximum hourly rainfall within 3 hours from the estimated occurrence time and an effective rainfall with a half-life at that time of 72 hours as a long-term rainfall index is particularly preferable.

Next, a sediment disaster occurrence information line creation program will be described as a second embodiment of the present invention. This program is a computer-readable program for executing the above-described method for creating a sediment disaster occurrence information line according to the present invention, and at least the amount of rainfall as data of a short-term rainfall index and a long-term rainfall index is stored in the computer. The actual data of the short-term rainfall index and the long-term rainfall index among the actual learning data consisting of the data set of the short-term rainfall index, the long-term rainfall index, and the occurrence of landslide disaster Standardized learning data consisting of a data set of short-term and long-term rainfall indices and the occurrence and non-occurrence of landslide disasters, and standardized learning data as RBF network learning data are constructed. Occurrence of landslide disasters standardized as contour lines corresponding to the predetermined risk of occurrence on the process and discrimination boundary surface Boundary lines, can be implemented as a program for executing the steps of creating at least one evacuation reference line and the warning reference line.
The contents, actions, and effects of each process are as already described in the embodiment of the method for creating the sediment disaster occurrence information line. In addition, the description which overlapped with the description which concerns on embodiment of the creation method of the occurrence information line of earth and sand disaster was abbreviate | omitted.

Next, as a third embodiment of the present invention, an earth and sand disaster warning and evacuation support system will be described with reference to FIGS. FIG. 13 is a configuration diagram of the earth and sand disaster warning and evacuation support system according to the present embodiment. In FIG. 13, the earth and sand disaster warning and evacuation support system includes a earth and sand disaster occurrence information management system 21, and four databases of an actual database 1, a standardization analysis database 3, a standardization database 6, and a discriminant boundary surface analysis database 8.
The earth and sand disaster occurrence information management system 21 receives input of data (actual learning data) 25 and analysis conditions 26 at the input unit 22 or stores actual learning data 2 stored in the actual database 1 through the input unit 22 in advance. You may read from the input part 22.
The actual learning data 2 input by the input unit 22 or read from the actual database 1 by the input unit 22 is standardized by the standardization analysis unit 23. The contents of the standardization analysis in the standardization analysis unit 23 will be described with reference to FIG. 14, but are the same as step S <b> 1 described in the embodiment relating to the method for creating the landslide disaster occurrence information line.

The standardization analysis unit 23 includes a standardized data reference value selection unit 28, a conversion coefficient calculation unit 29, and a standardized data calculation unit 30.
In the embodiment of the method of creating the sediment disaster occurrence information line, there was an actual data selection step. However, in the earth and sand disaster warning and evacuation support system, an input unit 22 is provided. Since the data is input or the actual learning data 2 is stored in advance in the actual database 1 and is read out, the standardization analysis unit 23 does not include a component to be selected.
The standardized data reference value selection unit 28 selects and reads one of several numbers selected in advance from the standardized data reference value table 4 stored in the standardization analysis database 3. An example of this standardized data reference value has already been described in the description of step S12, but there are good numbers such as 1,10,100. For this selection, a recommended value as the standardization analysis unit 23 may be determined in advance and automatically read out, or the standardized data reference value table 4 is output to the system user. 10, and a user's desired value may be input via the input unit 22. Alternatively, the user may input a desired value via the input unit 22 without referring to the standardized data reference value table 4.
The conversion coefficient calculation unit 29 calculates conversion coefficients for standardization. Since the calculation of the standardized conversion coefficient has already been performed in the description of step S13, it is omitted here. The standardized conversion coefficient 5 is stored in the standardized analysis database 3.
The standardized data calculation unit 30 reads the standardized conversion coefficient 5 stored in the standardization analysis database 3 and applies the standardized learning data 7 by non-dimensional conversion by multiplying it by the real learning data 2 read from the real database 1. Calculate. The calculated standardized learning data 7 is stored in the standardized database 6.
In addition, the description which overlapped with the description which concerns on embodiment of the preparation method of the occurrence information line of earth and sand disaster was abbreviate | omitted.

Next, the discrimination boundary surface analysis unit 24 will be described with reference to FIG.
FIG. 15 is a conceptual diagram showing a detailed configuration of the discriminating boundary surface analysis unit 24. In FIG. 15, the discriminant boundary surface analysis unit 24 includes a radial basis function center position setting unit 32, a clustering analysis unit 33, a radial basis function radius optimization unit 34, and a basis function synthesis unit 39. The radial basis function radius optimization unit 34 is further configured by an initial point setting unit 35, a tunnel processing unit 36, a steepest descent processing unit 37, and a determination unit 38.
The discriminant boundary surface analysis unit 24 reads the standardized learning data 7 stored in the standardization database 6 and executes analysis on each component. The discriminant boundary surface analysis unit 24 operates on each component. Since the same effects as those described in step S2 in the method for creating the sediment disaster occurrence information line are described here, the description thereof is omitted here.
However, in the embodiment of the method of creating the sediment disaster occurrence information line, as step S25, the discriminant boundary for the discriminant boundary surface analyzed by combining the basis functions in step S24 and other discriminant boundary surface analysis data 9 is used. Although a process of storing in the surface analysis database 8 is provided, in the system according to the present embodiment, each component of the radial basis function center position setting unit 32 to the determination unit 38 including the basis function synthesis unit 39 is provided. The data as the analysis result is transmitted to the discriminant boundary surface analysis database 8 and stored as the discriminant boundary surface analysis data 9.

The generation information line creation unit 27 creates a generation information line by using the discrimination boundary surface obtained by the analysis by the discrimination boundary surface analysis unit 24 or the discrimination boundary surface analysis data 9. And the effect is the same as that described in the explanation of Step S3 of the method for creating the occurrence information line of earth and sand disaster.
The output unit 10 is a display device such as a liquid crystal display, a printer, or outputs a data signal to another arithmetic device such as a computer. Outputs occurrence information lines such as occurrence limit lines, evacuation reference lines, and warning reference lines.
The warning and evacuation support system of the present invention is characterized by using at least one of a sediment disaster occurrence limit line, an evacuation reference line, and a warning reference line set by the method for creating a sediment disaster occurrence information line of the present invention. The detailed configuration is not intended to limit the present invention. For example, the same configuration as the debris flow warning and evacuation support system (Non-Patent Document 2) proposed by the present inventors described above. It is clear that the present invention can be implemented.

  For example, there is a sediment disaster occurrence limit line, an evacuation reference line, and a warning reference line set by the method of the present invention, and a geographic information system (GIS) is used, and a sediment disaster disaster risk based on measured rainfall data is used. Warning judgment evacuation support with the function of displaying the color judgment result, snake curve display, evacuation site information display, damage assumption display, damage information input and counting, and predicted sediment disaster risk determination result for the predicted rainfall It can be implemented as a system.

  In addition, the program includes the creation program for information on occurrence of earth and sand disasters according to the present invention. For example, using the additional learning / forgetting function of RBFN, the occurrence limit line reflecting the latest information such as hardware measures, evacuation reference line, warning It is also possible to update the reference line and maintain or improve its reliability.

  According to the embodiment as described above, the present invention uses RBFN, does not require enormous trial and error, can be set objectively without requiring user experience and intuition, and the reproducibility of learning data is high. A method to set the occurrence limit line of landslide disaster that has high versatility and can be easily applied to other regions with different rainfall conditions, a program for executing the method, and warning evacuation using the occurrence limit line A support system can be provided.

  Hereinafter, the present invention will be described more specifically with reference to examples. First, target data used for analysis in an embodiment related to a method for setting a fracture occurrence limit line, which will be described below, will be described. The target area is the southern part of Shimonoseki City, Yamaguchi Prefecture, where the risk of landslides is high throughout the country, and the target slope is a natural slope set as a dangerous spot in the target area. The slope where the occurrence of landslide was confirmed from 1975 to 1998 (hereinafter referred to as “target period”) was used. In addition, the non-occurrence slope has not been crushed within the target period, and the slope has not yet been taken countermeasures. AMeDAS data observed at Shimonoseki Meteorological Observatory during the target period was used as the target rainfall data. The generated rainfall is a series of rainfalls including the landslide occurrence time, and the generated rainfall factor used in the analysis is the maximum hourly rainfall within 3 hours from the time of occurrence of the short-term rainfall index (hereinafter sometimes referred to as “time rainfall”). The long-term rainfall index is an effective rainfall with a half-life at that time of 72 hours (hereinafter sometimes referred to as “effective rainfall”). On the other hand, non-occurrence rainfall was calculated by subtracting out-of-occurrence rainfall from all rainfalls in the target period, and the non-occurrence rainfall factors used in the analysis were hourly rainfall and effective rainfall at all times during non-occurrence rainfall.

  In this actual learning data, the maximum values (RLmax, RSmax) of effective rainfall and hourly rainfall are 290.6 mm / hr and 57 mm / hr, respectively, and the actual data reference values (RLmmax, RSmmax) set for standardization are respectively 330mm / hr and 60mm / hr. The standardized data reference values (SLmmax, SSmmax) set corresponding to the actual data reference values are 10 and 10, respectively. That is, in this embodiment, the conversion coefficients to be applied to the actual data in order to obtain standardized data are effective rainfall and hourly rainfall SLmmax / RLmmax = 0.0303hr / mm and SSmmax / RSmmax = 0.167hr / mm, respectively. In the example, standardized data converted using this conversion coefficient is used.

  Next, the division of the two-dimensional plane between the long-term rainfall index (effective rainfall) and the short-term rainfall index (time rainfall) performed in this embodiment will be described. In this embodiment, the rectangular area defined by the standardized data reference values (10 and 10) for each of the effective rainfall and the hourly rainfall is divided into unit grid areas each formed by grid lines set at predetermined intervals. The number of divisions was 33 and 40, respectively. That is, in this embodiment, a rectangular area defined by each standardized data reference value is divided into 1320 unit cell areas.

  Clustering of learning data is performed based on this division. For each generated data and non-generated data for each unit grid area, if there are a plurality of such data in the area, the plurality of data are centroided. By setting one new learning data clustered by the method and deleting the original multiple data, the density of the learning data used in RBFN was equalized. As a result, the number of learning data finally used for RBFN learning is 37 for generated data and 275 for non-generated data.

The suppression parameter λ related to this clustering is the upper limit number N _{max of the} actual data in the region, the lower limit value λ _min of the suppression parameter corresponding to the region data greater than the upper limit number, and the suppression parameter corresponding to only one region data. using the upper limit value lambda _max, set by _{λ = (λ max N max -λ} min + N (λ min -λ max)) / (N max -1). In the case where N exceeds N _max was treated as N = N _max. The upper limit number N _max of real data in the area was set to 100 for non-occurrence data and 10 for occurrence data. The lower limit value λ _min and the upper limit value λ _max of the suppression parameter are set to 0.01 and 10 which are the same for both non-occurrence data and generated data. The setting condition of the suppression parameter is set as a preferable condition in the present embodiment as a result of trial and error.

  The center position of the basis function is set as follows: (1) Only when the generation data exists in the region, if the basis function is not set at the upper right lattice point forming the region, the basis of the generation region is set at that lattice point. The function is identified and set. If a basis function is already set, change it to a mixed area basis function. (2) Lower left corner that forms the area only if non-occurrence data exists in the area. If no basis function is set for a grid point, a non-occurrence area basis function is identified and set for that grid point, and if a basis function is already set, it is changed to a mixed area basis function According to the above procedures, the classification and identification were carried out as either a sediment disaster occurrence area, a non-occurrence area, or a mixed area. As a result, the number of basis functions of the set generation region, non-occurrence region, and mixed region is 22, 260, and 15, respectively.

  Hereinafter, the results of comparison and examination under various conditions will be described. First, FIG. 16 is a two-sided view comparing the influence of the radius of the basis function when setting the landslide occurrence limit line using RBFN. That is, all the basis functions have the same radius and are fixed to a predetermined value, and the weight of the connection between the intermediate element and the output element is learned using RBFN. The discriminant boundary surface is constructed, and the contour lines corresponding to the predetermined occurrence risk of the discriminant boundary surface are shown in two-dimensional coordinates of effective rainfall (long-term rainfall index) and hourly rainfall (short-term rainfall index). The left figure shows the long-term rainfall index 40 and the short-term rainfall index 10, and the right figure shows the long-term rainfall index 30 and the short-term rainfall index 6. The radius value is a magnification in units of the scale length of each grid line constituting the unit grid area, and the radius value described below is also the same.

  Learning is performed by using 0 and 1 for teacher values, that is, occurrence and non-occurrence of landslide, and a smaller value means a higher physical risk. Although five lines are shown in FIG. 16, the discrimination corresponding to the occurrence risk levels of 0.0, 0.2, 0.4, 0.6, and 0.8 in descending order of the physical occurrence risk level (in order of distance from the origin), respectively. It is a contour line of the boundary surface. The coordinate axes in Fig. 16 are full scales of actual data reference values (330mm / hr and 60mm / hr, respectively) of effective rainfall and hourly rainfall, and the displayed contour lines are those of the actual database. is there. The display of the teacher value and the figure is the same in other embodiments described below.

  FIG. 16 shows that the white and gray areas indicate the safe and dangerous areas, respectively, and means that the risk of occurrence of collapse gradually increases as it approaches the gray area from white, but the smaller the set occurrence risk, the more This shows that the inner area decreases, and that the constructed boundary surface is reasonable and consistent with natural phenomena. The left diagram and the right diagram in FIG. 16 have energy of 22.52 and 22.54, respectively, and it can be said that almost the same optimization is performed, but there is a great difference between the left diagram and the right diagram. Indicates that the radius has a great influence on the generation limit line and the like.

FIG. 17 is a two-sided view comparing the influences of the suppression parameters when setting the landslide occurrence limit line using RBFN. That is, all the basis functions have the same radius, a predetermined initial value is set, and the radius is optimized by a simple steepest descent method. The suppression parameters are all set to 1 for generated data, and non-generated data Λ _{min is set} to 0.1, λ _max is set to 10 in the left figure, and 1 in the right figure. The initial values of the set radius are 30 for the long-term rainfall index and 6 for the short-term rainfall index, and the optimized radius results are 20.40 and 23.94 on the left and 21.36 and 13.90 on the right, respectively.

FIG. 17 is an example showing that the setting condition of the suppression parameter has a significant effect on the result, and the setting condition of the suppression parameter, such as setting λ _{max of} non-occurrence data to 10 described above, is such trial and error. As a result of the examination by the above, it is set as suitable in the present embodiment. This result can be suitably used in other regions with different rainfall conditions by using learning data standardized by the predetermined procedure of the present invention.

  FIG. 18 is a three-sided view comparing the influence of the initial value of the radius when setting the landslide occurrence limit line using RBFN. That is, as described above, the areas are classified into the three areas of occurrence / non-occurrence / mixture, and the respective radii are optimized by a simple steepest descent method. The initial values of the radii are the same for the three areas. The left figure is 1.0, the middle figure is 2.0, and the right figure is 3.0. The radius was optimized by setting a lower limit of 0.01. The radius of the optimized result is the long-term rainfall index and the short-term rainfall index, respectively, the left figure is 0.667 and 0.846 for the occurrence area, 2.27 and 1.18 for the non-occurrence area, and 0.01 and 1.20 for the mixed area. Are 0.443 and 0.01, 1.89 and 4.60, 0.01 and 0.01, and the right figure is 4.29 and 0.01, 0.783 and 3.01, and 1.10 and 2.31.

  FIG. 18 shows that optimization by the simple steepest descent method is not sufficient as an optimization method for setting a collapse occurrence limit line or the like having versatility due to the significant influence of the initial radius value. This is due to the fact that the optimization problem of the present invention has multimodality, that is, multimodality in which there are a plurality of local points at which the objective function exhibits a minimum value. The results in FIG. 18 are all about energy 20.5, which is about 10.5 lower than the energy optimized by the simple steepest descent method without dividing the region. This means that a high reproducibility can be obtained by dividing, and the effect of dividing the area is shown.

  FIG. 19 is a two-sided view showing a comparison of the influence of the step width used in the convergence calculation of the steepest descent method when setting the fracture occurrence limit line or the like using RBFN. That is, it is optimized by changing only the step width (γ in FIG. 9) under the same conditions as in the right diagram of FIG. 18. The step width is set to 0.5 in FIG. 0.72. The left figure of FIG. 19 is the same figure as the right figure of FIG. 18, and FIG. 19 shows that the optimization problem of the present invention having multi-modality affects the result even by changing the step width. This indicates that it is necessary to select an appropriate step width.

  Next, the optimization method used in the present embodiment will be described based on the various examination results. That is, the overall processing flow is the overall flowchart of FIG. 6 described in the embodiment, in which (1) an initial point search step for obtaining an initial value of a radius and setting an initial point; and (2) a current point. And select one of the variables, give a step width to the selected variable, compare the tunnel destination point with the skipped value by energy, and if the relationship satisfies the predetermined condition, set the tunnel destination point to the current point. And (3) a steepest descent process step for performing convergence calculation to obtain a local point that minimizes energy according to the sensitivity, and (4) a steepest descent process step. Judgment to determine the result obtained in the processing step and decide whether to end the optimization process with the local point as the optimal point or to update the number of optimization calculations and return to the tunnel processing step to continue the optimization process Process and 4 And a sub-process.

In detail, these four sub-processes are the flowcharts shown in FIGS. 7 to 10, respectively, and the parameters and conditions used in this embodiment will be described below. Regarding the initial point search process of FIG. 7, the lower limit value and upper limit value (X _min and X _{max in the} figure) of the random number generator (random in the figure) are non-occurrence areas (X ₁ and X _{2 in the} figure). The generation areas (X ₃ and X _{4 in the} figure) are the same 1 and 3, and the mixed areas (X ₅ and X _{6 in the} figure) are 0.5 and 1. The radius of the long-term rainfall index axis (the radius with an odd suffix in the figure) and the radius of the short-term rainfall index axis (the radius with an even suffix in the figure) are the same, that is, set as a circular basis function. The number of initial points to be generated (k in the figure) is 30. In the figure, E is the energy of RBFN, min is a minimum value selector, and j is a counter for the number of optimization calculations.

Regarding the tunnel processing step of FIG. 8, the lower limit value of the random number generator (random in the figure) for generating the step width coefficient is 0, and the upper limit value (α in the figure) is 5. The minimum value (ζ in the figure), which is the first constraint condition of the variable X (radius), is 0.01, and the second constraint condition, the radius of the long-term rainfall index axis to be applied outside the mixed area (in the figure, The maximum ratio (ξ in the figure) of X ₁ and X ₃ ) and the radius of the short-term rainfall index axis (X ₂ and X _{4 in the} figure) is 2. This restriction condition is the same in the steepest descent process step of FIG. The lower limit and upper limit values (β in the figure) for determining the upper limit value (β in the figure) of the random number generator (random in the figure) that generates an application coefficient that is the inverse of the number of optimization calculations. _min and β _max ) are 0 and 0.02. In the figure, j _max is the allowable number of calculations, and 40 was used.

Regarding the steepest descent process step in FIG. 9, the step width (γ in the figure) for the convergence calculation is 0.72, and the convergence determination condition (ε in the figure) is 0.0001. The constraint condition of variable X (in the figure, ζ and ξ) is the same as the tunnel processing step described above. Regarding the determination step in FIG. 10, the determination condition for the amount of change in the objective function (ε in the figure) and the determination condition for the number of optimization calculations (j _{max in the} figure) are 0.0001 and 40, respectively, as described above.

  First, among the above parameters / conditions, the condition for setting the initial point as a circular basis function and the maximum ratio between the radius of the long-term rainfall index axis and the radius of the short-term rainfall index axis to be applied outside the mixed region are set. The result of optimization without providing the second constraint will be described. FIG. 20 is a two-sided view showing the results. The radius of the optimized result is the long-term rainfall index and the short-term rainfall index, respectively, and the left figure shows the occurrence areas 0.233 and 1.63, and the non-occurrence area 1.19. 3.98, mixed area is 3.61 and 0.01, and the right figure is 3.43 and 0.20, 1.16 and 3.54, 5.02 and 0.01. The energy is 19.45 on the left and 19.21 on the right.

  That is, under this parameter / condition, both the energy can be greatly reduced, but there are still two local solutions, and it is difficult to obtain an optimal solution stably, and the radius of the long-term rainfall index axis And the radius of the short-term rainfall index axis are large, and it has been found that there is an aspect that is not consistent with natural phenomena and is not appropriate.

  Therefore, except for the condition to set the initial point as a circular basis function and the mixed area that requires careful handling, the radius of the long-term rainfall index axis and the radius of the short-term rainfall index axis applied to the non-occurrence area and the occurrence area Including the second constraint that sets the maximum ratio, optimization was performed according to the above parameters and conditions. FIG. 21 is a diagram showing the results, and the optimized radii are the long-term and short-term rainfall indices, 0.839 and 0.419 for the occurrence area, 2.02 and 3.66 for the non-occurrence area, and 4.02 for the mixed area, respectively. And 0.01, and the energy is 19.74.

  According to the optimization using these parameters and conditions, the same result could be obtained stably without falling into a local solution. This is because by setting the initial point as a circular basis function, if one of the variables starts from an extremely small point, it becomes smaller and covers it with the other variables, and from the local solution It is thought that it is possible to prevent the escape from becoming impossible.

  As described above, according to the present embodiment, the RBFN parameters such as the radius and center position of the basis function and the upper and lower limit values of the connection weight suppression parameters are initially determined by trial and error, as in the prior art. Although it is necessary, the value obtained by trial and error is based on the standardized learning data, and even in regions with different rainfall conditions, by using the standardized learning data according to the predetermined procedure of this embodiment Can be used as generally suitable parameters. That is, according to the present embodiment, since the RBFN parameter know-how obtained previously can be used, it does not require enormous trial and error, does not require user experience and intuition, and is objectively reproduced. It is possible to set the occurrence limit line of landslide disasters with high characteristics. Therefore, it also has versatility that can be easily applied to other areas with different rainfall conditions.

  In addition, it is possible to set the limit line of occurrence for high-precision earth and sand disasters with greatly improved reproducibility of learning data, and there are parameters that must be trial and error based on the experience and intuition of users. RBFN parameters such as the center position of the basis function to be determined initially by trial and error and the upper and lower limits of the connection weight suppression parameter can be obtained extremely easily, and more preferably Can be set.

  As mentioned above, although the Example of this invention was described, it is clear that various changes may be made to the form and detail, without deviating from the spirit and scope of this invention prescribed | regulated by the claim. .

  For example, the various parameters described in the examples were set as a result of examination by trial and error based on the regions used in the examples, particularly the learning data, but it could be said that the optimum settings were not necessarily made. In addition, there is a possibility that there is no possibility, and further, including the results of examinations targeting other areas with different rainfall conditions, further optimal settings should be made, and the present invention is not limited at all.

  In the embodiment, an example in which the initial point is set as a circular basis function has been described. For example, the ratio of the radius of the long-term rainfall index axis to the radius of the short-term rainfall index axis is limited to a predetermined range. The maximum ratio between the radius of the long-term rainfall index axis and the radius of the short-term rainfall index axis applied to areas other than the mixed area is set to 2 in the embodiment, but the present invention is not limited to this. It is not something.

  It has a wide range of uses, such as drafting disaster prevention plans and creating hazard maps in public institutions such as local governments and disaster prevention centers, and issuing evacuation advisories and warnings due to heavy rains. It is also expected to be used as a teaching material for disaster prevention and evacuation training in educational institutions. In addition, private companies that operate construction and civil engineering projects can also be used as tools for unearthing needs for disaster prevention projects and project proposals, or for sharing disaster information for collaboration with public institutions. It can also be applied to applications such as research and development and design projects related to disaster prevention technology.

It is a flowchart which shows the creation method of the occurrence information line of the earth and sand disaster which concerns on embodiment of this invention. It is the conceptual diagram which showed the setting method of the center position of the radial basis function based on the learning data of the generation | occurrence | production / non-generation in a unit cell area | region. It is the graph which plotted rainfall data. FIG. 5 is a conceptual diagram in which basis functions are set by plotting learning data generated / not generated by dividing into unit lattice regions. It is the graph which plotted rainfall data. It is the whole flowchart which showed the flow of the whole process about preferable embodiment of an optimization process. It is the flowchart which showed the preferable specific example of the initial point search process for an optimization process. It is the flowchart which showed the preferable specific example of the tunnel processing process for an optimization process. It is the flowchart which showed the preferable specific example of the steepest descent process process for an optimization process. It is the flowchart which showed the preferable specific example of the determination process for an optimization process. It is a conceptual diagram expressing the shape of a Gaussian function. It is a conceptual diagram which shows the specific example of a discrimination | determination boundary surface. It is a block diagram of a warning and evacuation support system for earth and sand disasters according to the present embodiment. It is a conceptual diagram which shows the detailed structure of a standardization analysis part. It is a conceptual diagram which shows the detailed structure of a discrimination | determination boundary surface analysis part. FIG. 5 is a two-sided view comparing the influence of the radius of the basis function when setting the landslide occurrence limit line using RBFN. FIG. 5 is a two-sided view comparing the influences of the suppression parameters when setting the landslide occurrence limit line using RBFN. FIG. 5 is a three-sided view comparing the influence of the initial value of the radius when setting the fracture occurrence limit line using RBFN. FIG. 6 is a two-sided view comparing the influence of the step width used in the convergence calculation of the steepest descent method when setting the landslide occurrence limit line using RBFN. Optimize without setting the second constraint to set the maximum ratio between the radius of the long-term rainfall index axis and the condition of setting the initial point as a circular basis function and the radius of the long-term rainfall index axis applied to the mixed area It is the two views which showed the result. Optimized, including conditions for setting the initial point as a circular basis function and a second constraint that sets the maximum ratio between the radius of the long-term rainfall index axis and the short-term rainfall index axis applied to the mixed region It is the figure which showed the result.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 ... Real database 2 ... Actual learning data 3 ... Standardization analysis database 4 ... Standardization data reference value table 5 ... Standardization conversion coefficient 6 ... Standardization database 7 ... Standardization learning data 8 ... Discrimination boundary surface analysis database 9 ... Discrimination boundary surface analysis data DESCRIPTION OF SYMBOLS 10 ... Output part 11 ... Vertical axis 12 ... Horizontal axis 13 ... Basis function 14a-14d ... Rainfall data 15a-15d ... Cluster representative point 16a-16d ... Cluster 17 ... Discrimination boundary surface 18 ... Z axis 19 ... X axis 20 ... Y Axis 21 ... Sediment disaster occurrence information management system 22 ... Input unit 23 ... Standardization analysis unit 24 ... Discrimination boundary surface analysis unit 25 ... Data 26 ... Analysis condition 27 ... Generation information line creation unit 28 ... Standardized data reference value selection unit 29 ... Conversion Coefficient calculation unit 30 ... Standardized data calculation unit 32 ... Radial basis function center position setting unit 33 Clustering analyzer 34 ... Radial Basis Function radius optimizing unit 35 ... initial point setting unit 36 ... tunneling processor 37 ... steepest descent processing unit 38 ... judging unit 39 ... basis function combining unit

In order to achieve the above object, the invention according to claim 1 is the first aspect in which the standardization analysis unit performs standardization of actual learning data including a data set of short-term rainfall index, long-term rainfall index, and occurrence / non-occurrence of landslide disaster. A radial basis function is set on the grid points by dividing the two-dimensional plane of the short-term rainfall index and the long-term rainfall index into unit grid areas formed by grid lines with a desired interval by the process and the discriminant interface analysis unit. Occurrence of landslide disaster having a second step of constructing a discrimination boundary surface and a third step of creating at least one of a landslide disaster occurrence limit line, an evacuation reference line, and a warning reference line by an occurrence information line creation unit A method of creating a limit line, an evacuation reference line, and a warning reference line,
In the first step, the standardization analysis unit generates standardized learning data by standardizing the actual data of the short-term rainfall index and the long-term rainfall index among the actual learning data stored in advance in the actual database or input from the input unit. And storing it in a standardized database.
In the second step, the discriminant boundary surface analysis unit reads standardized learning data from the standardized database, sets grid lines at predetermined intervals on the axes indicating the standardized short-term rainfall index and the long-term rainfall index, respectively. A two-dimensional plane of the index and the long-term rainfall index is divided into unit grid areas formed by the grid lines, and for each unit grid area, learning data on occurrence of sediment disasters exists in the unit grid area, and the unit grid A radial basis function is set only when a radial basis function is not set at the upper right grid point forming the region, and there is learning data that does not cause sediment disaster in the unit grid region, and the unit grid region is Analyzing discriminant interface by constructing discriminant interface with means to set non-generated radial basis function at the lattice point only when the lower left lattice point is not set Is a process to be stored in the database,
In the third step, the occurrence information line creation unit reads out the discrimination boundary surface from the discrimination boundary surface analysis database, and standardizes it as a contour line corresponding to a predetermined occurrence risk, the landslide occurrence limit line, the evacuation reference line, and the warning standard This is a process of creating at least one of the lines .

The invention according to claim 2 is a first step of executing standardization of actual learning data comprising a data set of short-term rainfall index, long-term rainfall index, and occurrence / non-occurrence of sediment disaster by the standardization analysis unit, and discriminant boundary surface analysis The two-dimensional plane of the short-term rainfall index and the long-term rainfall index is divided into unit grid areas formed by grid lines with a desired interval by the unit, and a radial basis function is set on the grid points to construct a discrimination boundary surface. Sediment disaster occurrence limit line, evacuation reference line, and the third process of creating at least one of a sediment disaster occurrence limit line, an evacuation reference line, and a warning reference line by the occurrence information line creation unit A method for creating a warning baseline,
In the first step, the standardization analysis unit generates standardized learning data by standardizing the actual data of the short-term rainfall index and the long-term rainfall index among the actual learning data stored in advance in the actual database or input from the input unit. And storing it in a standardized database.
In the second step, the discriminant boundary surface analysis unit reads standardized learning data from the standardized database, sets grid lines at predetermined intervals on the axes indicating the standardized short-term rainfall index and the long-term rainfall index, respectively. A two-dimensional plane of the index and the long-term rainfall index is divided into unit grid areas formed by the grid lines, and for each unit grid area, learning data on occurrence of sediment disasters exists in the unit grid area, and the unit grid A radial basis function is set only when a radial basis function is not set at the upper right grid point forming the region, and there is learning data that does not cause sediment disaster in the unit grid region, and the unit grid region is Means for setting a non-generated radial basis function at a lattice point only when a radial basis function is not set at the lower left lattice point to be formed;
In the means for setting the generated radial basis function and the non-occurring radial basis function, the learning data and the earth and sand disaster occurrence data included corresponding to the short-term rainfall index and the long-term rainfall index standardized for each unit grid area to be set A means of clustering non-disaster learning data;
Means for optimizing the radius of the generated radial basis function and the non-generated radial basis function set in the means for setting the generated radial basis function and the non-generating radial basis function;
Optimized by means of learning data clustered by means of clustering learning data on occurrence of landslide disaster and learning data on non-occurrence of landslide disaster, and means for optimizing the radius of radial basis functions of occurrence and non-occurrence of radial basis functions Means for combining generated and non-generated radial basis functions with a radius
And constructing a discrimination boundary surface and storing it in the discrimination boundary surface analysis database,
In the third step, the occurrence information line creation unit reads out the discrimination boundary surface from the discrimination boundary surface analysis database, and standardizes it as a contour line corresponding to a predetermined occurrence risk, the landslide occurrence limit line, the evacuation reference line, and the warning standard Create at least one of the lines .

According to a third aspect of the present invention, the radial basis function is a Gaussian function in the method of creating the sediment disaster occurrence limit line, the evacuation reference line, and the warning reference line according to the first or second aspect.

According to a fourth aspect of the present invention, there is provided a method for creating the occurrence limit line, the evacuation reference line, and the warning reference line of a sediment disaster according to the second aspect , wherein the radius of the generated radial basis function and the non-generated radial basis function are optimized. The means to achieve
First random number generating means for generating an initial value of a radius within a preset upper and lower limit value is provided, and a plurality of initial points are generated by the first random number generating means, and the generated initial values An initial point setting means for obtaining an objective function of a radial basis function of a point and setting a point where the value of the objective function is the smallest as an initial point of optimization;
A second random number generating means for selecting one variable from a plurality of variables included in the objective function; and a third random number generating means for generating a step width coefficient on the condition of a lower limit value of 0 and a predetermined upper limit value. A tunnel processing means for setting the tunnel destination point using the product of the sensitivity of the objective function at the current point defined by the variable of the optimization initial point and the step width coefficient as a step width and updating the tunnel destination point as the current point;
A steepest descent processing means for performing a convergence calculation for obtaining a local point that minimizes the objective function according to this sensitivity by obtaining a sensitivity to the variable of the objective function of the current point updated by the tunnel processing means;
If the value of the objective function of the local point obtained by the steepest descent processing means satisfies a predetermined condition, the local point is set as the optimal point, and if the predetermined condition is not satisfied, the number of optimization calculations is updated. Determination means for returning to the tunnel processing means and continuing the optimization process,
The predetermined condition in the steepest descent processing means is that the absolute value of the amount of change in the value of the objective function is smaller than a predetermined value or the number of optimization calculations is larger than a predetermined allowable number of times. Is.

According to a fifth aspect of the present invention, in the method for generating the sediment disaster occurrence limit line, the evacuation reference line, and the warning reference line according to the fourth aspect, the initial point setting means is a two-dimensional plane of a short-term rainfall index and a long-term rainfall index. Are classified into three groups of landslide disaster occurrence areas, non-occurrence areas, and mixed areas based on landslide disaster occurrence data and landslide disaster non-occurrence data existing in unit grid areas formed by grid lines of predetermined intervals, For each group, an initial point in which the radius of the axis of the short-term rainfall index and the radius of the axis of the long-term rainfall index are set to be the same is set .

The invention according to claim 6 is the method for creating the occurrence limit line of landslide disaster, the evacuation reference line, and the warning reference line according to claim 4 , wherein the tunnel processing means is set within a preset upper and lower limit value. A fourth random number generating means for generating an application coefficient of the objective function to be compared, wherein a value inversely proportional to the number of optimization calculations is set as an upper limit value of the application coefficient on the condition of the preset upper and lower limit values; A fourth random number generation device for generating, wherein the sum of the application coefficient and 1 is an application constant, the product of the objective function value of the current point and the application constant, and the value of the objective function of the tunnel destination point And when the value of the objective function of the tunnel destination point is small, the tunnel destination point is updated to the current point .

Invention of claim 7, claim 1 to landslides occurrence limit line according to any one of claims 6, in the creation process of evacuation reference line and alert the reference line, occurrence limit of landslides to the set Lines, evacuation reference lines, and warning reference lines are the slopes of the entire region or the occurrence limit of the stream, including multiple slopes or multiple streams with different potential risks of landslides due to topographic factors Is.

The invention according to claim 8 is a first step of performing standardization of actual learning data comprising a data set of a short-term rainfall index, a long-term rainfall index, and occurrence / non-occurrence of landslide disaster by a standardization analysis unit on a computer ; The boundary surface analysis unit divides the two-dimensional plane of the short-term rainfall index and the long-term rainfall index into unit cell areas formed by grid lines with a desired interval, and sets a radial basis function on the lattice points to set the discriminant boundary surface. Sediment disaster occurrence limit line for executing the second process to be constructed and the third process of creating at least one of the sediment disaster occurrence limit line, the evacuation reference line, and the warning reference line by the occurrence information line creation unit, An evacuation reference line and warning reference line creation program,
In the first step, the standardization analysis unit generates standardized learning data by standardizing the actual data of the short-term rainfall index and the long-term rainfall index among the actual learning data stored in advance in the actual database or input from the input unit. And storing it in a standardized database.
In the second step, the discriminant boundary surface analysis unit reads standardized learning data from the standardized database, sets grid lines at predetermined intervals on the axes indicating the standardized short-term rainfall index and the long-term rainfall index, respectively. A two-dimensional plane of the index and the long-term rainfall index is divided into unit grid areas formed by the grid lines, and for each unit grid area, learning data on occurrence of sediment disasters exists in the unit grid area, and the unit grid A radial basis function is set only when a radial basis function is not set at the upper right grid point forming the region, and there is learning data that does not cause sediment disaster in the unit grid region, and the unit grid region is Analyzing discriminant interface by constructing discriminant interface with means to set non-generated radial basis function at the lattice point only when the lower left lattice point is not set Is a process to be stored in the database,
In the third step, the occurrence information line creation unit reads out the discrimination boundary surface from the discrimination boundary surface analysis database, and standardizes it as a contour line corresponding to a predetermined occurrence risk, the landslide occurrence limit line, the evacuation reference line, and the warning standard It is a step of creating at least one of the lines .

The invention according to claim 9 includes a standardization analysis unit that performs standardization of actual learning data including a data set of a short-term rainfall index, a long-term rainfall index, and occurrence / non-occurrence of landslide disaster, and a short-term rainfall index and a long-term rainfall index. A discriminant boundary surface analysis unit that constructs a discriminant boundary surface by dividing a two-dimensional plane into unit lattice regions formed by grid lines with a desired interval and setting radial basis functions on the lattice points, and the occurrence limit of sediment disasters In a warning and evacuation support system for earth and sand disasters having an occurrence information line creation unit that creates at least one of a line, an evacuation reference line, and a warning reference line,
The standardization analysis unit generates the standardized learning data by standardizing the actual data of the short-term rainfall index and the long-term rainfall index from the actual learning data stored in advance in the actual database or input from the input unit. Store and
The discriminant interface analysis unit reads the standardized learning data from the standardized database, sets grid lines at predetermined intervals on the axes indicating the standardized short-term rainfall index and the long-term rainfall index, and sets the short-term rainfall index and the long-term rainfall index. Are divided into unit grid areas formed by the grid lines, and for each unit grid area, there are learning data on occurrence of earth and sand disasters in the unit grid area, and the upper right corner of the unit grid area is formed. Set the radial basis function to be generated only when no radial basis function is set for the grid point, and there is no sediment disaster occurrence data in the unit grid area, and the lower left grid point that forms the unit grid area Only when the radial basis function is not set in, the discriminant boundary surface is constructed with means for setting a non-generated radial basis function at the lattice point, and the discriminant boundary surface analysis database is stored. To pay,
The occurrence information line creation unit reads out the discrimination boundary surface from the discrimination boundary surface analysis database, and at least one of the occurrence limit line of landslide disaster, the evacuation reference line, and the warning reference line standardized as a contour line corresponding to a predetermined occurrence risk level. To create .

The invention described in claim 10 includes a standardization analysis unit that performs standardization of actual learning data including a data set of a short-term rainfall index, a long-term rainfall index, and occurrence / non-occurrence of landslide disaster, and a short-term rainfall index and a long-term rainfall index. A discriminant boundary surface analysis unit that constructs a discriminant boundary surface by dividing a two-dimensional plane into unit lattice regions formed by grid lines with a desired interval and setting radial basis functions on the lattice points, and the occurrence limit of sediment disasters In a warning and evacuation support system for earth and sand disasters having an occurrence information line creation unit that creates at least one of a line, an evacuation reference line, and a warning reference line,
The standardization analysis unit generates the standardized learning data by standardizing the actual data of the short-term rainfall index and the long-term rainfall index from the actual learning data stored in advance in the actual database or input from the input unit. Store and
The discriminant interface analysis unit reads the standardized learning data from the standardized database, sets grid lines at predetermined intervals on the axes indicating the standardized short-term rainfall index and the long-term rainfall index, and sets the short-term rainfall index and the long-term rainfall index. Are divided into unit grid areas formed by the grid lines, and for each unit grid area, there are learning data on occurrence of earth and sand disasters in the unit grid area, and the upper right corner of the unit grid area is formed. Set the radial basis function to be generated only when no radial basis function is set for the grid point, and there is no sediment disaster occurrence data in the unit grid area, and the lower left grid point that forms the unit grid area Means for setting a non-generated radial basis function at the lattice point only when a radial basis function is not set in
In the means for setting the generated radial basis function and the non-occurring radial basis function, the learning data and the earth and sand disaster occurrence data included corresponding to the short-term rainfall index and the long-term rainfall index standardized for each unit grid area to be set A means of clustering non-disaster learning data;
Means for optimizing the radius of the generated radial basis function and the non-generated radial basis function set in the means for setting the generated radial basis function and the non-generating radial basis function;
Optimized by means of learning data clustered by means of clustering learning data on occurrence of landslide disaster and learning data on non-occurrence of landslide disaster, and means for optimizing the radius of radial basis functions of occurrence and non-occurrence of radial basis functions Means for combining generated and non-generated radial basis functions with a radius
To construct a discriminant boundary surface, store it in the discriminant boundary surface analysis database,
The occurrence information line creation unit reads out the discrimination boundary surface from the discrimination boundary surface analysis database and creates at least one of the occurrence limit line of landslide disaster, evacuation reference line, and warning reference line standardized as contour lines corresponding to the predetermined occurrence risk To do .

In order to achieve the above object, the invention according to claim 1 is the first aspect in which the standardization analysis unit performs standardization of actual learning data including a data set of short-term rainfall index, long-term rainfall index, and occurrence / non-occurrence of landslide disaster. By dividing the two-dimensional plane of the short-term rainfall index and the long-term rainfall index into unit grid areas formed by grid lines of desired intervals by the process and the discriminant boundary surface analysis unit, a radial basis function is set on the grid points , A second step of constructing a discriminant boundary surface formed by synthesizing the radial basis functions on the lattice points, and at least one of the occurrence limit line of the earth and sand disaster, the evacuation reference line, and the warning reference line by the generation information line creation unit A method for creating a sediment-related disaster occurrence limit line, an evacuation reference line, and a warning reference line having a third step of creating any one of the following steps:
In the first step, the standardization analysis unit generates standardized learning data by standardizing the actual data of the short-term rainfall index and the long-term rainfall index among the actual learning data stored in advance in the actual database or input from the input unit. And storing it in a standardized database.
In the second step, the discriminant boundary surface analysis unit reads standardized learning data from the standardized database, sets grid lines at predetermined intervals on the axes indicating the standardized short-term rainfall index and the long-term rainfall index, respectively. The two-dimensional plane of the index and the long-term rainfall index is divided into unit grid areas formed by the grid lines, and for each unit grid area, standardized learning data for occurrence of sediment disasters exists in the unit grid area and the unit A radial basis function is set only when no radial basis function is set at the upper right grid point forming the grid area, and there is standardized learning data that does not cause sediment disasters in the unit grid area, and the unit grid Means for setting a non-generating radial basis function at a lattice point only when the radial basis function is not set at the lower left lattice point forming the region ;
Standardized learning data for occurrence of landslide disasters corresponding to the short-term rainfall index and long-term rainfall index standardized for each set unit grid area in the means for setting the generated radial basis function and the non-generating radial basis function, and A means of clustering standardized learning data that does not cause sediment disasters,
Means for optimizing the radius of the generated radial basis function and the non-generated radial basis function set in the means for setting the generated radial basis function and the non-generating radial basis function;
By standardized learning data clustered by means of clustering standardized learning data of landslide occurrence and non-occurrence of landslide disasters, and by means of means to optimize the radius of generated radial basis functions and non-generated radial basis functions Means for combining generated and non-generated radial basis functions with optimized radii;
And constructing a discrimination boundary surface and storing it in the discrimination boundary surface analysis database,
The means for optimizing the radius of the generated radial basis function and the non-generated radial basis function exists in the unit cell area formed by a grid line with a predetermined interval between the two-dimensional plane of the short-term rainfall index and the long-term rainfall index. Categorized into 3 groups of landslide disaster occurrence area, non-occurrence area and mixed area based on the occurrence data of landslide disaster and non-occurrence data of landslide disaster, and the initial value of radius is generated within the preset upper and lower limit values First random number generating means for generating a plurality of initial values of a plurality of radii by the first random number generating means, obtaining an objective function of a radial basis function of the generated initial values of the plurality of radii, The point where the value of the objective function is the smallest is set as the initial value of the optimization radius, and the radial basis function radius in the short-term rainfall index axis direction and the radial basis function in the long-term rainfall index axis direction are set for each of the three groups. An initial point setting means and the radius of the are set to the same,
A second random number generating means for selecting one variable from a plurality of variables included in the objective function; and a third random number generating means for generating a step width coefficient on the condition of a lower limit value of 0 and a predetermined upper limit value. The tunnel destination value is set under a predetermined condition with the product of the sensitivity of the objective function of the current value defined by the variable of the initial value of the optimization radius and the step width coefficient as the step width, and this tunnel destination value is set as the current value. Tunnel processing means to be updated as
A steepest descent processing means for calculating a sensitivity for a variable of the objective function of the current value updated by the tunnel processing means and performing a convergence calculation for obtaining a local value that minimizes the value of the objective function according to the sensitivity;
When the value of the objective function of the local value obtained by the steepest descent processing means satisfies a predetermined condition, the local value is set as an optimum value, and when the predetermined condition is not satisfied, the number of optimization calculations is updated. Determination means for returning to the tunnel processing means and continuing the optimization process,
The predetermined condition in the tunnel processing means is that the radius of the radial basis function of the tunnel destination value is not less than the preset minimum value, and the radial basis function in the long-term rainfall index axis direction in the occurrence and non-occurrence areas of sediment disasters Characterized in that the maximum ratio of the radius and the radial basis function radius in the direction of the short-term rainfall index axis does not exceed the set value,
The predetermined condition in the steepest descent processing means is that the absolute value of the amount of change in the value of the objective function is smaller than a predetermined value or the number of optimization calculations is larger than a predetermined allowable number of times, and The maximum ratio between the radial basis function radius in the long-term rainfall index axis direction and the radial basis function radius in the short-term rainfall index axis direction in the occurrence and non-occurrence areas of sediment disasters is characterized by not exceeding the set value. ,
In the third step, the occurrence information line creation unit reads out the discrimination boundary surface from the discrimination boundary surface analysis database, and standardizes it as a contour line corresponding to a predetermined occurrence risk, the landslide occurrence limit line, the evacuation reference line, and the warning standard This is a process of creating at least one of the lines.

The invention according to claim 2 is the method for creating the sediment disaster occurrence limit line, the evacuation reference line and the warning reference line according to claim 1 , wherein the set sediment disaster occurrence limit line, the evacuation reference line and the warning reference line are set. Is characterized by the occurrence limit of slopes or streams in the entire region, including multiple slopes or mountain streams with different potential risks of landslides due to topographic factors.

According to a third aspect of the present invention, the computer includes a first step of performing standardization of actual learning data including a data set of a short-term rainfall index, a long-term rainfall index, and a sediment disaster occurrence / non-occurrence by a standardization analysis unit, and a discrimination The boundary plane analysis unit divides the two-dimensional plane of the short-term rainfall index and the long-term rainfall index into unit grid regions formed by grid lines with a desired interval, and sets radial basis functions on the grid points. A second step of constructing a discriminant boundary surface formed by synthesizing the radial basis functions of and a generation information line creation unit create at least one of the sediment disaster occurrence limit line, the evacuation reference line, and the warning reference line A program for creating a landslide disaster occurrence limit line, an evacuation reference line, and a warning reference line for executing the third step,
In the first step, the standardization analysis unit generates standardized learning data by standardizing the actual data of the short-term rainfall index and the long-term rainfall index among the actual learning data stored in advance in the actual database or input from the input unit. And storing it in a standardized database.
In the second step, the discriminant boundary surface analysis unit reads standardized learning data from the standardized database, sets grid lines at predetermined intervals on the axes indicating the standardized short-term rainfall index and the long-term rainfall index, respectively. The two-dimensional plane of the index and the long-term rainfall index is divided into unit grid areas formed by the grid lines, and for each unit grid area, standardized learning data for occurrence of sediment disasters exists in the unit grid area and the unit A radial basis function is set only when no radial basis function is set at the upper right grid point forming the grid area, and there is standardized learning data that does not cause sediment disasters in the unit grid area, and the unit grid Means for setting a non-generating radial basis function at a lattice point only when the radial basis function is not set at the lower left lattice point forming the region ;
Standardized learning data for occurrence of landslide disasters corresponding to the short-term rainfall index and long-term rainfall index standardized for each set unit grid area in the means for setting the generated radial basis function and the non-generating radial basis function, and A means of clustering standardized learning data that does not cause sediment disasters,
Means for optimizing the radius of the generated radial basis function and the non-generated radial basis function set in the means for setting the generated radial basis function and the non-generating radial basis function;
By standardized learning data clustered by means of clustering standardized learning data of landslide occurrence and non-occurrence of landslide disasters, and by means of means to optimize the radius of generated radial basis functions and non-generated radial basis functions Means for combining generated and non-generated radial basis functions with optimized radii;
And constructing a discrimination boundary surface and storing it in the discrimination boundary surface analysis database,
The means for optimizing the radius of the generated radial basis function and the non-generated radial basis function exists in the unit cell area formed by a grid line with a predetermined interval between the two-dimensional plane of the short-term rainfall index and the long-term rainfall index. Categorized into 3 groups of landslide disaster occurrence area, non-occurrence area and mixed area based on the occurrence data of landslide disaster and non-occurrence data of landslide disaster, and the initial value of radius is generated within the preset upper and lower limit values First random number generating means for generating a plurality of initial values of a plurality of radii by the first random number generating means, obtaining an objective function of a radial basis function of the generated initial values of the plurality of radii, The point where the value of the objective function is the smallest is set as the initial value of the optimization radius, and the radial basis function radius in the short-term rainfall index axis direction and the radial basis function in the long-term rainfall index axis direction are set for each of the three groups. An initial point setting means and the radius of the are set to the same,
A second random number generating means for selecting one variable from a plurality of variables included in the objective function; and a third random number generating means for generating a step width coefficient on the condition of a lower limit value of 0 and a predetermined upper limit value. The tunnel destination value is set under a predetermined condition with the product of the sensitivity of the objective function of the current value defined by the variable of the initial value of the optimization radius and the step width coefficient as the step width, and this tunnel destination value is set as the current value. Tunnel processing means to be updated as
A steepest descent processing means for calculating a sensitivity for a variable of the objective function of the current value updated by the tunnel processing means and performing a convergence calculation for obtaining a local value that minimizes the value of the objective function according to the sensitivity;
When the value of the objective function of the local value obtained by the steepest descent processing means satisfies a predetermined condition, the local value is set as an optimum value, and when the predetermined condition is not satisfied, the number of optimization calculations is updated. Determination means for returning to the tunnel processing means and continuing the optimization process,
The predetermined condition in the tunnel processing means is that the radius of the radial basis function of the tunnel destination value is not less than the preset minimum value, and the radial basis function in the long-term rainfall index axis direction in the occurrence and non-occurrence areas of sediment disasters Characterized in that the maximum ratio of the radius and the radial basis function radius in the direction of the short-term rainfall index axis does not exceed the set value,
The predetermined condition in the steepest descent processing means is that the absolute value of the amount of change in the value of the objective function is smaller than a predetermined value or the number of optimization calculations is larger than a predetermined allowable number of times, and The maximum ratio between the radial basis function radius in the long-term rainfall index axis direction and the radial basis function radius in the short-term rainfall index axis direction in the occurrence and non-occurrence areas of sediment disasters is characterized by not exceeding the set value. ,
In the third step, the occurrence information line creation unit reads out the discrimination boundary surface from the discrimination boundary surface analysis database, and standardizes it as a contour line corresponding to a predetermined occurrence risk, the landslide occurrence limit line, the evacuation reference line, and the warning standard It is a step of creating at least one of the lines.

The invention described in claim 4 includes a standardization analysis unit that performs standardization of actual learning data including a data set of a short-term rainfall index, a long-term rainfall index, and occurrence / non-occurrence of landslide disaster, and a short-term rainfall index and a long-term rainfall index. A discriminant boundary formed by dividing a two-dimensional plane into unit lattice areas formed by lattice lines with a desired interval, setting a radial basis function on the lattice point, and combining the radial basis functions on the lattice point In a landslide disaster warning and evacuation support system having a discriminant boundary surface analysis unit that constructs a surface, and an occurrence information line creation unit that creates at least one of the occurrence limit line, evacuation reference line, and warning reference line of a landslide disaster,
The standardization analysis unit generates the standardized learning data by standardizing the actual data of the short-term rainfall index and the long-term rainfall index from the actual learning data stored in advance in the actual database or input from the input unit. Store and
The discriminant interface analysis unit reads the standardized learning data from the standardized database, sets grid lines at predetermined intervals on the axes indicating the standardized short-term rainfall index and the long-term rainfall index, and sets the short-term rainfall index and the long-term rainfall index. Is divided into unit grid areas formed by the grid lines, and for each unit grid area, standardized learning data for occurrence of earth and sand disasters exists in the unit grid area and forms the unit grid area. A radial basis function is set only when no radial basis function is set at the grid point of the grid, and there is standardized learning data that does not cause sediment disaster in the unit grid area, and the unit grid area is formed at the bottom left Means for setting a non-generated radial basis function at a lattice point only when a radial basis function is not set at the lattice point ;
Standardized learning data for occurrence of landslide disasters corresponding to the short-term rainfall index and long-term rainfall index standardized for each set unit grid area in the means for setting the generated radial basis function and the non-generating radial basis function, and A means of clustering standardized learning data that does not cause sediment disasters,
Means for optimizing the radius of the generated radial basis function and the non-generated radial basis function set in the means for setting the generated radial basis function and the non-generating radial basis function;
By standardized learning data clustered by means of clustering standardized learning data of landslide occurrence and non-occurrence of landslide disasters, and by means of means to optimize the radius of generated radial basis functions and non-generated radial basis functions Means for synthesizing generated and non-generated radial basis functions with an optimized radius;
To construct a discriminant boundary surface, store it in the discriminant boundary surface analysis database,
The means for optimizing the radius of the generated radial basis function and the non-generated radial basis function exists in the unit cell area formed by a grid line with a predetermined interval between the two-dimensional plane of the short-term rainfall index and the long-term rainfall index. Categorized into 3 groups of landslide disaster occurrence area, non-occurrence area and mixed area based on the occurrence data of landslide disaster and non-occurrence data of landslide disaster, and the initial value of radius is generated within the preset upper and lower limit values First random number generating means for generating a plurality of initial values of a plurality of radii by the first random number generating means, obtaining an objective function of a radial basis function of the generated initial values of the plurality of radii, The point where the value of the objective function is the smallest is set as the initial value of the optimization radius, and the radial basis function radius in the short-term rainfall index axis direction and the radial basis function in the long-term rainfall index axis direction are set for each of the three groups. An initial point setting means and the radius of the are set to the same,
A second random number generating means for selecting one variable from a plurality of variables included in the objective function; and a third random number generating means for generating a step width coefficient on the condition of a lower limit value of 0 and a predetermined upper limit value. The tunnel destination value is set under a predetermined condition with the product of the sensitivity of the objective function of the current value defined by the variable of the initial value of the optimization radius and the step width coefficient as the step width, and this tunnel destination value is set as the current value. Tunnel processing means to be updated as
A steepest descent processing means for calculating a sensitivity for a variable of the objective function of the current value updated by the tunnel processing means and performing a convergence calculation for obtaining a local value that minimizes the value of the objective function according to the sensitivity;
When the value of the objective function of the local value obtained by the steepest descent processing means satisfies a predetermined condition, the local value is set as an optimum value, and when the predetermined condition is not satisfied, the number of optimization calculations is updated. Determination means for returning to the tunnel processing means and continuing the optimization process,
The predetermined condition in the tunnel processing means is that the radius of the radial basis function of the tunnel destination value is not less than the preset minimum value, and the radial basis function in the long-term rainfall index axis direction in the occurrence and non-occurrence areas of sediment disasters Characterized in that the maximum ratio of the radius and the radial basis function radius in the direction of the short-term rainfall index axis does not exceed the set value,
The predetermined condition in the steepest descent processing means is that the absolute value of the amount of change in the value of the objective function is smaller than a predetermined value or the number of optimization calculations is larger than a predetermined allowable number of times, and The maximum ratio between the radial basis function radius in the long-term rainfall index axis direction and the radial basis function radius in the short-term rainfall index axis direction in the occurrence and non-occurrence areas of sediment disasters is characterized by not exceeding the set value. ,
The occurrence information line creation unit reads out the discrimination boundary surface from the discrimination boundary surface analysis database, and at least one of the occurrence limit line of landslide disaster, the evacuation reference line, and the warning reference line standardized as a contour line corresponding to a predetermined occurrence risk level. To create.

Claims (22)

  A hierarchical structure of an input layer having a plurality of input elements, an intermediate layer having a plurality of intermediate elements, and an output layer having one output element, wherein the intermediate element is a radial basis function (RBF) and the plurality of intermediate elements Using the RBF network that learns the weight of the connection with the output element, by learning a plurality of learning data consisting of a data set of short-term rainfall index, long-term rainfall index, and sediment disaster occurrence / non-occurrence, the short-term rainfall index and A discriminant boundary surface consisting of a long-term rainfall index and the risk of occurrence of landslide disasters is constructed, and as the contour lines corresponding to the predetermined risk of occurrence of the discriminant boundary surface, the occurrence limits of landslide disasters, evacuation reference lines and warning reference lines A method for creating a sediment disaster occurrence limit line, an evacuation reference line, and a warning reference line for creating at least one of the following: Among the actual learning data consisting of the data set of short-term rainfall index, long-term rainfall index and landslide disaster occurrence / non-occurrence using real data to perform the predetermined conversion respectively to the actual data of the short-term rainfall index and long-term rainfall index A step of obtaining standardized learning data comprising a data set of short-term rainfall index, long-term rainfall index and sediment disaster occurrence / non-occurrence as standardized data, and using the standardized learning data as learning data of the RBF network A step of constructing a surface, and a step of creating at least one of a sediment disaster occurrence limit line, an evacuation reference line, and a warning reference line standardized as contour lines corresponding to a predetermined occurrence risk of the discrimination boundary surface A method for creating the occurrence limit of landslide disasters, evacuation reference lines, and warning reference lines.   The method for creating a sediment disaster occurrence limit line, an evacuation reference line, and a warning reference line according to claim 1, wherein the radial basis function is a Gaussian function.   The step of obtaining the standardized learning data obtains the maximum values of the actual data of the short-term rainfall index and the long-term rainfall index from the actual learning data, respectively, and multiplies the maximum value by a first constant preset at 1 or more. Each actual data reference value is calculated, and each ratio obtained by dividing the second constant of each of the short-term rainfall index and the long-term rainfall index set in advance by each actual data reference value is calculated. 2. The step of performing non-dimensional conversion by multiplying actual learning data of each of the short-term rainfall index and the long-term rainfall index, and obtaining standardized learning data using them as standardized data. The method for creating the sediment disaster occurrence limit line, evacuation reference line, and warning reference line described in Item 2.   The step of constructing the discriminant boundary surface includes a step of setting a center position of the radial basis function, which includes an axis indicating the short-term rainfall index (hereinafter referred to as a short-term rainfall index axis) and a long-term rainfall index. A grid line with a predetermined interval is set on each of the axes (hereinafter referred to as the long-term rainfall index axis), and a two-dimensional plane of the short-term rainfall index and the long-term rainfall index is defined as a unit grid area formed by the grid lines. The grid is divided for each unit grid area only when there is learning data on occurrence of sediment disaster in the unit grid area and no radial basis function is set at the upper right grid point forming the unit grid area. A radial basis function is set at a point, and non-occurrence learning data exists in the unit cell region, and the radial basis function is not set at the lower left point forming the unit cell region. Set radial basis functions Claims 1 to landslides occurrence limit line according to any one of claims 3, create evacuation reference line and the warning reference line method, characterized in that a step of.   The step of constructing the discriminant boundary surface includes a step of clustering learning data on occurrence / non-occurrence of landslide disaster for each unit grid area, and the step includes the unit for each occurrence / non-occurrence of landslide disaster. Read out the upper limit value and lower limit value of the learning data preset in the learning data in the lattice area and the preset weight suppression parameter of the connection weight from the database, and generate / not occur for each unit lattice area Based on the number of learning data in the unit cell region, the suppression parameter for each learning data is set to the learning data in the unit cell region having learning data equal to or greater than the upper limit number, and the lower limit value of the suppression parameter is set. The upper limit value of the suppression parameter is set in the learning data of the unit cell region where the learning data is only one, and the number of learning in the meantime is set. Learning data of the unit grid area having data, the learning data for each of the generated and non-generated learning data for each unit grid area, the upper limit value and the lower limit value of the suppression parameter is proportionally distributed and set When there are a plurality of learning data, one new learning data is set by the centroid method in the coordinate space composed of the short-term rainfall index axis and the long-term rainfall index axis, and the original learning data is deleted. A method for creating a sediment-related disaster occurrence limit line, an evacuation reference line, and a warning reference line according to claim 4.   The step of constructing the discriminant boundary surface includes the step of classifying the radial basis functions into a plurality of groups according to a preset condition, and the classified group is a group composed of radial basis functions each having the same radius. And a step of optimizing a radius of a radial basis function for each group using an objective function of the RBF network as an optimization objective function. To create the landslide disaster occurrence limit line, evacuation reference line and warning reference line.   The radius to be optimized is a radius of each group classified as a group of radial basis functions each having the same radius, and a radius of a short-term rainfall index axis and a radius of a long-term rainfall index axis, respectively. The method for creating a sediment disaster limit line, an evacuation reference line, and a warning reference line according to claim 6 characterized by the above-mentioned.   The step of constructing the discriminant boundary surface includes the step of setting the center position of the radial basis function, and the step of setting the center position of the radial basis function includes a plurality of the radial basis functions according to preset conditions. The radial basis function includes identification information for identifying any of the occurrence area of landslide disaster, the non-occurrence area, or the mixed area, and the process includes the short-term rainfall index axis. And the long-term rainfall index axis are set at predetermined intervals to divide the two-dimensional plane of the short-term rainfall index and the long-term rainfall index into unit grid areas formed by the grid lines, and for each unit grid area Only when there is learning data on the occurrence of sediment disaster in the unit grid area, if no radial basis function is set at the upper right grid point forming the unit grid area, A radial basis function is identified and set, and if a radial basis function is already set at the upper right grid point, it is changed to a radial basis function of a mixed area, and learning that does not occur in the unit grid area Only when data is present, if a radial basis function is not set at the lower left lattice point forming the unit lattice region, the radial basis function of the non-generating region is identified and set at the lower left lattice point, When a radial basis function has already been set at the lower left grid point, this is a step of changing this to a radial basis function of a mixed region, and the classification of the group is a classification based on the identification information, The creation method of the occurrence limit line of the earth and sand disaster of Claim 6 or Claim 7, the evacuation reference line, and the warning reference line.   The step of optimizing the radius of the radial basis function includes an initial point setting step of obtaining an initial value of the radius which is a variable of the optimization and setting an initial point defined by the value of the variable, The current point defined by the value of the variable is compared with the tunnel destination point selected by selecting at least one of the variables, giving a step width to the selected variable, and skipping the value. When the set condition is satisfied, the tunnel processing step for updating the tunnel destination point as the current point, and the sensitivity of the objective function to the variable are obtained, and the local point that minimizes the objective function according to the sensitivity is determined. Determine the steepest descent process step for performing convergence calculation for obtaining, and the result obtained in the steepest descent process step, and terminate the optimization process with the local point as the optimal point, or update the number of optimization calculations Tunnel A determination step for determining whether to return to the physical process and continue the optimization process; and a landslide disaster occurrence limit line and an evacuation reference line according to any one of claims 6 to 8 And how to create a warning baseline.   The initial point setting step includes first random number generation means for generating an initial value of the radius on the condition of preset upper and lower limit values, and a plurality of the initial points are generated by the first random number generation means. 10. The step of obtaining each of the generated objective functions at a plurality of initial points and setting a point at which the value of the objective function is the smallest as an initial point for optimization. To create the landslide disaster occurrence limit line, evacuation reference line and warning reference line.   The initial point is an initial point in which a radius of a short-term rainfall index axis and a radius of a long-term rainfall index axis are set to be the same for each of the plurality of groups. How to create earth and sand disaster occurrence limit lines, evacuation reference lines, and warning reference lines.   The tunnel processing step selects a variable using a second random number generating means for selecting one of the variables, and generates a coefficient of the step width on the condition of a preset upper and lower limit value. 3 is used to generate a step width coefficient, the product of the sensitivity of the objective function at the current point with respect to the selected variable and the step width coefficient as the step width, and the tunnel destination point 12. The method for creating a landslide disaster occurrence limit line, an evacuation reference line, and a warning reference line according to any one of claims 9 to 11, further comprising: a set step.   The tunnel processing step uses the fourth random number generating means for generating the application coefficient of the objective function to be compared on the condition of a preset upper and lower limit value, and uses the preset upper and lower limit value as a condition. A value inversely proportional to the number of optimization calculations is set as the upper limit value of the application coefficient in the fourth random number generation means, the application coefficient is generated by the fourth random number generation means, and the sum of the application coefficient and 1 is calculated. An application constant, the product of the objective function value of the current point and the application constant is compared with the value of the objective function of the tunnel destination point, and if the tunnel destination point is smaller, the tunnel destination point The method according to any one of claims 9 to 12, further comprising an annealing process step of updating the current point as a current point. .   The determination performed in the determination step is a determination performed based on an absolute value of a change amount of the objective function between a current optimization calculation and a previous optimization calculation, and the number of optimization calculations. The method for creating a sediment disaster occurrence limit line, an evacuation reference line, and a warning reference line according to any one of claims 9 to 13.   The landslide disaster according to any one of claims 6 to 14, wherein the radius optimization is an optimization performed on the condition that the radius is equal to or greater than a preset minimum value. Of creating the occurrence limit line, evacuation reference line, and warning reference line.   The optimization of the radius is performed on the condition that the ratio of the radius of the short-term rainfall index axis and the radius of the long-term rainfall index axis is not more than a predetermined value in the radial basis function of the occurrence area and the non-occurrence area. The method for creating a sediment-related disaster occurrence limit line, an evacuation reference line, and a warning reference line according to any one of claims 8 to 15, characterized in that the optimization is performed.   The sedimentary disaster occurrence limit line, evacuation reference line, and warning reference line set above include the slopes or streamflows of the entire region, including multiple slopes or multiple mountain streams with different potential risks of sediment disasters due to topographic factors. The method according to any one of claims 1 to 16, wherein the occurrence limit line, the evacuation reference line, and the warning reference line are prepared.   A step of calculating a risk of occurrence of a sediment-related disaster due to topographic factors for each slope or individual mountain stream, a step of classifying the slope or mountain stream into a plurality of groups based on the magnitude of the individual potential risk, and the group A step of calculating an average potential risk of a slope or a stream included in each, a step of learning the weights of the plurality of connections using the RBF network for each group, and a plurality of the learned weights A step of approximating each of the average potential risks as a function of the average potential risk, and a connection corresponding to the individual potential risks by using the average function as the individual potential risks. A step of calculating a weight of the discriminant, a step of constructing the discriminant boundary surface corresponding to each potential risk using the weight, and a contour line corresponding to a predetermined risk of occurrence of the individual discriminant boundary surface And a step of setting at least one of an occurrence limit line, an evacuation reference line, and a warning reference line of landslides of individual slopes or individual mountain streams having the individual potential risks. A method for creating a sediment disaster occurrence limit line, an evacuation reference line, and a warning reference line according to any one of claims 1 to 16.   The potential risk is a potential risk evaluated by using a scoring system, and calculates a sediment disaster occurrence rate by category for each slope factor or mountain stream factor, and sets the occurrence rate as the set score of the category. 19. The method for creating a landslide disaster occurrence limit line, an evacuation reference line, and a warning reference line according to claim 18, wherein the potential risk level is obtained by adding a score for each factor or each mountain stream factor.   The actual data of the short-term rainfall index and the long-term rainfall index having the above-mentioned rainfall dimension are 72 times the maximum hourly rainfall within 3 hours from the occurrence estimated time or the effective rainfall with a half-life of 1.5 hours and the half-life at that time, respectively. 20. The method for creating a sediment-related disaster occurrence limit line, an evacuation reference line, and a warning reference line according to any one of claims 1 to 19, wherein the effective rainfall is time.   A computer has a hierarchical structure of an input layer having a plurality of input elements, an intermediate layer having a plurality of intermediate elements, and an output layer having one output element, wherein the intermediate element is a radial basis function (RBF) and the plurality of intermediate By using the RBF network that learns the weight of the connection between the element and the output element, the short-term rainfall is learned by learning a plurality of learning data consisting of a short-term rainfall index, a long-term rainfall index, and a sediment disaster occurrence / non-occurrence data set. Build a discriminant boundary surface composed of indicators, long-term rainfall indicators, and the risk of landslide disasters, and create contour lines for landslide disasters, evacuation reference lines, and warning standards as contour lines corresponding to the predetermined risk of occurrence of the discriminant boundary surface A program for creating a sediment-related disaster occurrence limit line, an evacuation reference line, and a warning reference line for creating at least one of the lines, the computer storing the short-term rainfall index Among the actual learning data consisting of the short-term rainfall index, the long-term rainfall index, and the occurrence / non-occurrence of the landslide disaster, using the actual data having the dimension of rainfall as the long-term rainfall index data, the short-term rainfall index and Obtaining standardized learning data consisting of a data set of short-term rainfall index, long-term rainfall index, and occurrence / non-occurrence of landslide disaster as standardized data by applying predetermined conversion to actual data of long-term rainfall index, respectively, and the standardized learning The step of constructing the discriminant boundary surface using the data as learning data of the RBF network, and the occurrence limit line of landslide disaster, the evacuation reference line and the warning reference line standardized as contour lines corresponding to a predetermined occurrence risk of the discriminant boundary surface A process of creating at least one of the following: a sediment-related disaster occurrence limit line, an evacuation reference line, and a warning base Line of creation program.   An input unit that reads actual learning data consisting of a data set of short-term and long-term rainfall indexes and data on the occurrence and non-occurrence of landslides from external or real databases. The respective maximum values are calculated from the short-term rainfall index and the long-term rainfall index of the actual data, and each actual data reference value is calculated by multiplying the maximum value by a first constant set in advance at 1 or more. The respective ratios obtained by dividing the second constant of each of the short-term rainfall index and the long-term rainfall index set in advance by the actual data reference value are calculated, and the respective ratios are actually learned for each of the short-term rainfall index and the long-term rainfall index. A standardization analysis unit that performs transformation to be dimensionless by multiplying the data and calculates the respective standardized learning data, and stores the first constant and the second constant A database for standardization analysis, a standardization database for storing the calculated standardized learning data, a discriminant boundary surface analysis unit for constructing a discriminant boundary surface using the standardized learning data as learning data for the RBF network, and displaying the discriminant boundary surface Or a landslide disaster warning and evacuation support system having an output unit for printing or outputting information, wherein the discrimination boundary surface analysis unit sets grid lines at predetermined intervals on a short-term rainfall index axis and a long-term rainfall index axis, respectively. Then, the two-dimensional plane of the short-term rainfall index and the long-term rainfall index is divided into unit grid areas formed by the grid lines, and there is learning data on occurrence of sediment disaster in the unit grid area for each unit grid area. And a radial basis function is set at the lattice point only when the upper right lattice point forming the region is not set, A radial basis function center position setting unit that sets a radial basis function at a lattice point only when non-generated learning data exists and a radial basis function is not set at the lower left lattice point forming the region; The radial basis functions are classified into a plurality of groups according to preset conditions, and the classified groups are groups of radial basis functions each having the same radius, and the objective function of the RBF network is optimized. A evacuation support system for earth and sand disasters, comprising: a radial basis function radius optimization unit that optimizes a radius of a radial basis function for each group.