JP6609391B1 - Prediction model creation device, prediction model creation method, and prediction model creation program - Google Patents

Abstract

A parameter of a prediction model is locally learned by specifying a period for which prediction accuracy is to be improved and a range of actually measured values that cannot be ignored. A prediction model creation apparatus according to the present invention obtains an actual measurement value for obtaining an actual measurement value indicating a property of raw water acquired by a water purification plant and an actual measurement value indicating a quantity of a medicine injected into the raw water. And a parameter optimization unit that optimizes the parameters of the prediction model using the actual measured value as the explanatory variable and the value indicating the amount of the drug as the objective variable. A range in which the difference between the predicted value indicating the amount of drug output by the prediction model and the actual measured value indicating the amount of drug is significantly large, and / or the range of actual values excluding outliers. It is characterized by comprising a receiving period / range receiving unit, and a parameter correcting unit for correcting a parameter using an actual measurement value corresponding to the received period and / or range. [Selection] Figure 11

Description

  The present invention relates to a prediction model creation device, a prediction model creation method, and a prediction model creation program.

  In the water purification plant, the flocculant is injected into the raw water obtained from rivers and the like, and the pollutants contained in the raw water are aggregated. The agglomerated contaminants are precipitated or filtered. The amount of the flocculant to be injected varies depending on the values indicating the properties of the raw water (turbidity, temperature, hydrogen ion concentration, etc.). Therefore, it is necessary to predict the amount of the flocculant to be injected based on the properties of the raw water. Recently, not only water purification plants, but also a technology that uses prediction models to predict various numerical values necessary for creating social infrastructure operation plans.

  The prediction device of Patent Literature 1 predicts future power demand. The daily maximum temperature, minimum temperature, etc. fluctuate with a relatively long cycle (for example, several months), and also with a relatively short cycle (for example, one day). The prediction device learns the parameters of the prediction model with a long-period moving average value. When the difference between the prediction value of the prediction model and the measured value of the moving average with a short period reaches a certain threshold, the prediction device corrects the parameter of the prediction model.

  The river water level prediction apparatus of Patent Document 2 predicts the water level of a pumping station near the estuary from the water level / precipitation amount in the upstream area and the tide level in the downstream area. In general, the fluctuation cycle of water level and precipitation in the upstream region is longer than the fluctuation cycle of the tide level in the downstream region. Therefore, the river water level prediction apparatus repeats learning the parameters of the prediction model based on the latest tide level in a short cycle. Only when the error of the prediction model (difference between the predicted value and the actual measurement value) increases to a certain extent, the latest actual measurement values of the water level and precipitation in the upstream area are used to learn the parameters of the prediction model retrospectively. cure.

JP2015-90691A JP-A-9-95917

  In the prediction of the injection amount of the flocculant at the water purification plant, there is a high need to make the current injection amount coincide with the mode value of the most recent months, for example. However, the prediction device of Patent Document 1 cannot learn a prediction model as such.

In exceptionally hot days in midsummer, immediately after heavy rain, the turbidity of raw water obtained by the water purification plant becomes larger than expected, and the flocculant becomes ineffective in the first place. However, when the prediction model is learned by uniformly using the actual measurement values during the period including such an exceptional time, the prediction accuracy of the prediction model in a specific period that should be predicted accurately decreases. In other words, it is necessary to increase the prediction accuracy for a specific period even at the expense of accuracy at the time of exception. However, the river water level prediction apparatus of Patent Document 2 cannot learn a prediction model in that way.
Therefore, an object of the present invention is to specify a period for which prediction accuracy is to be improved and a range of actually measured values that cannot be ignored, and to locally learn parameters of a prediction model.

The prediction model creation device of the present invention includes an actual measurement value acquisition unit that acquires an actual measurement value indicating a property of raw water acquired by a water purification plant, and an actual measurement value indicating a quantity of a medicine injected into the raw water, and an actual measurement The parameter optimization unit that optimizes the parameters of the prediction model using the value indicating the nature of the raw water as the explanatory variable and the value indicating the amount of the drug as the objective variable, and the prediction model in which the parameter is optimized A period in which the difference between the predicted value indicating the amount of the medicine to be output and the actual measurement value of the drug amount is significantly increased, and / or the period / range in which the range of the actual measurement value excluding outliers is received A reception unit, a parameter correction unit that corrects a parameter using an actual measurement value corresponding to the received period and / or range, an actual measurement value of a value indicating a drug amount, and a prediction model in which the parameter is optimized Output the amount of drug When the period / range accepting unit accepts the user's input of the period and range with respect to the displayed actual measurement value and predicted value, the value indicating the amount of the drug is displayed. And a display processing unit that displays a predicted value of a value indicating the amount of medicine output by the prediction model with the parameter corrected, in time series .
Other means will be described in the embodiment for carrying out the invention.

  According to the present invention, it is possible to locally learn a parameter of a prediction model by designating a period for which prediction accuracy is to be improved and a range of actually measured values that cannot be ignored.

It is a figure explaining the structure etc. of a prediction model production apparatus. It is an example of measured value information. It is a figure explaining the preparation method of a prediction model. It is a figure explaining the preparation method of a prediction model. It is a figure explaining period designation | designated and range designation | designated. It is a figure explaining period designation | designated and range designation | designated. It is a figure explaining period designation | designated and range designation | designated. It is a figure explaining period designation | designated and range designation | designated. It is an example of the screen which displays a measured value and a predicted value. It is an example of the screen which displays a measured value and a predicted value. It is an example of the screen which displays a measured value and a predicted value. It is a flowchart of a processing procedure.

  Hereinafter, a mode for carrying out the present invention (referred to as “the present embodiment”) will be described in detail with reference to the drawings. This embodiment is an example in which a flocculant is injected into raw water at a water purification plant. However, the present invention can also be applied to an example in which a drug other than the flocculant (flocculation aid, oxidizing agent, adsorbent, disinfectant, etc.) is injected into the raw water. Furthermore, the present invention is more generally applicable to an example in which a drug is injected into a liquid.

(Prediction model creation device)
A configuration and the like of the prediction model creation device 1 will be described with reference to FIG. The prediction model creation device 1 is a general computer, and includes a central control device 11, an input device 12 such as a mouse and a keyboard, an output device 13 such as a display, a main storage device 14, an auxiliary storage device 15, and a communication device 16. . These are connected to each other by a bus. The auxiliary storage device 15 stores a prediction model 31 and measured value information 32 (detailed later).

  The measured value acquisition unit 21, the variable reception unit 22, the parameter optimization unit 23, the period / range reception unit 24, the parameter correction unit 25, and the display processing unit 26 in the main storage device 14 are programs. The central control device 11 reads out these programs from the auxiliary storage device 15 and loads them into the main storage device 14, thereby realizing the functions (details described later) of the respective programs. The auxiliary storage device 15 may have a configuration independent of the prediction model creation device 1.

  The water purification plant 3 includes a mixing basin 4, a flocculant injection device 5, and a flock formation pond 7. Raw water 6 a acquired from a natural environment such as a river flows into the mixing basin 4. The flocculant injection device 5 injects the flocculant 9 into the raw water 6a stored in the mixing pond. The raw water 6a infused with the flocculant 9 is rapidly stirred and then flows into the flock formation pond 7 as mixed water 6b. The mixed water 6b stored in the flock formation pond 7 is gently stirred. Then, colloidal pollutants start to aggregate. The agglomerated contaminants are then precipitated or filtered.

  Substances (drugs) generally used as the flocculant 9 are aluminum sulfate, polyaluminum chloride and the like. The pollutants in the raw water are negatively charged and repel each other. When the aggregating agent 9 neutralizes the charge, the pollutants attract each other by intermolecular force and grow greatly to form a floc.

  “Turbidity” exists as a physical quantity indicating the degree of contamination of raw water. Turbidity is the mass (mg) of contaminants per liter of raw water. According to the homepage of a certain prefectural water department, the turbidity of clear river water is normally "5". However, turbidity “2600” has been recorded immediately after the typhoon. The water purification plant determines the coagulant injection rate so that the turbidity falls below a predetermined reference value. The flocculant injection rate is the mass (mg / liter) of the flocculant injected per liter of raw water. As physical quantities indicating the properties of raw water that are factors for determining the coagulant injection rate, there are turbidity, hydrogen ion concentration (pH, neutral = 7), temperature (water temperature, ° C.), and the like.

  Therefore, the turbidity meter 8a, the pH meter 8b, and the thermometer 8c are arranged in the mixing basin 4 and the injection amount meter 8d is arranged in the flocculant injection device 5. These are connected to the prediction model creation apparatus 1 via the network 2.

(Prediction model)
The prediction model 31 of the present embodiment is a linear expression such as Expression 1 below.
y = a ₀ + a ₁ x ₁ + a ₂ x ₂ + a ₃ x ₃ (Formula 1)

Here, y is a flocculant injection rate. x ₁ is the turbidity. x ₂ is a hydrogen ion concentration. x ₃ is the water temperature. a ₀ , a ₁ , a ₂ and a ₃ are constants (parameters). Expression 1 is a function having x ₁ , x ₂ and x ₃ as explanatory variables and y as an objective variable. Then, by changing the values of a ₀ , a ₁ , a ₂ and a ₃ in various ways, the shape and position of the prediction model 31 in the four-dimensional space are changed in various ways. Here, setting the number (type) of variables to “4” is merely an example. The number of variables may be further increased, that is, the dimension of the prediction model may be further increased.

Now, it is assumed that there are many “[Y, X ₁ , X ₂ , X ₃ ]” as combinations of past measured values of the flocculant injection rate, turbidity, hydrogen ion concentration, and water temperature. The physical quantity indicated by each of Y, X ₁ , X ₂ and X ₃ is the same as the physical quantity indicated by each of y, x ₁ , x ₂ and x ₃ . However, for convenience of explanation, actual measurement values are shown in capital letters, and prediction model variables are shown in lower case letters. The output (objective variable) y of the prediction model is a “prediction value”. “Y−y” is called an error. The prediction model creation device 1 uses a combination of measured values, and uses a combination of parameters “[a ₀ , a ₁ , a ₂ , a ₃ ” that minimizes the sum of squares of error “Σ (Y−y) ² ”. ] ”(Details will be described later).

(Measured value information)
FIG. 2 is an example of the actual measurement value information 32. In the actual measurement value information 32, in association with the time stored in the time column 101, the target variable column 102 stores the actual measurement value of the target variable, and the explanatory variable column 103 stores the actual measurement value of the explanatory variable.
The time in the time column 101 is the time when the measured value of the explanatory variable and the measured value of the objective variable were acquired by the turbidimeter 8a, pH meter 8b, thermometer 8c, and injection meter 8d. The times t1, t2, t3,... The larger the number, the later the time.

The actual measurement value of the objective variable in the objective variable column 102 is an actual measurement value of the coagulant injection rate measured by the injection meter 8d. “#” Indicates a different value in an abbreviated manner (the same applies hereinafter).
The measured values of the explanatory variables in the explanatory variable column 103 were measured by the turbidity of raw water measured by the turbidimeter 8a (column 103a), the hydrogen ion concentration of raw water measured by the pH meter 8b (column 103b), and the thermometer 8c. This is the temperature of raw water (column 103c).

(Prediction model creation method)
3 and 4 are diagrams illustrating a method for creating the prediction model 31. FIG. For simplicity of explanation, it is now assumed that the explanatory variable is only turbidity. FIG. 3 is the same as the columns 101, 102, and 103a of the actual measurement value information 32 (FIG. 2). Then, the prediction model (Formula 1) becomes “y = a ₀ + a ₁ x ₁ ”. The prediction model creation device 1 executes the following processes (1) to (7).

(1) predictive model generation apparatus 1 substitutes random manner the values of the parameters a ₀ and a ₁ which is generated in a ₀ and a ₁ of the prediction model.
(2) predictive model creation apparatus 1 substitutes the measured values X in x ₁ prediction model to calculate a y.
(3) The prediction model creation device 1 calculates the error “Y−y”.
(4) The prediction model creation device 1 repeats the processes (2) and (3) while changing the value of [X, Y] at each time.

(5) The prediction model creation device 1 calculates “Σ (Y−y) ² ”, which is the sum of “(Y−y) ² ” at each time.
(6) the prediction model generating apparatus 1, in a random manner other values of the generated parameters _{a 0} and _{a 1} was after having assigned to _{a 0} and _{a 1} in the prediction model, the (2) - (5) This process is repeated a sufficient number of times.
(7) The prediction model creation device 1 determines the values of the parameters a _0S and a _1S that minimize “Σ (Y−y) ² ”. Here, “S” indicates “optimized”.

The horizontal axis of the coordinate plane in FIG. 4 is turbidity, and the vertical axis is the flocculant injection rate. Twenty dots ● indicate the combination [X, Y] of the actually measured values in FIG. The straight line 31 is the prediction model 31, and the equation is “y = a _0S + a _1S x ₁ ”. For each of the points ●, an error “Y−y” is defined. As described above, “Σ (Y−y) ² ” is minimized, but when attention is paid to individual points ●, there are a mixture of a case where there is almost no error and a case where the error is relatively large. Yes.

  Returning to FIG. Now, it is assumed that the time t4 to t7 is a period for which the user wants to improve the prediction accuracy. For example, the period corresponds to a specific season, and the user creates the prediction model 31 using the actual measurement values of the previous year, and tries to predict the coagulant injection rate on the same day of the same month in the following year. At this time, there is no problem that the error of the point ● of the actual measurement value of the previous year is sufficiently small (in the immediate vicinity of the prediction model 31).

  However, if that is not the case, the prediction will be greatly off. Specifically, even if the turbidity on the same day of the same month of the following year is the same as the turbidity of the same month of the previous year, the predicted value y of the flocculant injection rate of the same month of the following year is It deviates greatly from the actual measurement value Y. This divergence is because the parameters of the prediction model 31 are greatly influenced by the actual measurement values in the periods other than the period in which the prediction accuracy is desired to be improved. Therefore, it is necessary to eliminate this influence.

  FIG. 5 is a diagram for explaining period designation and range designation. FIG. 5 is obtained by deleting the actually measured values during the period other than the times t4 to t7 from FIG. The oblique line is drawn in the column of the coagulant injection rate at time t5. This hatched line indicates that the user has determined that this value is an “outlier”. The “○” in the right column will be described later.

  FIG. 6 is obtained by deleting the points ● indicating the measured values in the period other than the times t4 to t7 from FIG. Three ● and one ○ remain. This circle corresponds to time t5 in FIG. And the prediction model production apparatus 1 produces the prediction model 31b using three ● of the remaining points. As a result of the parameters being recalculated (corrected), the prediction model 31b is different in position and inclination from the prediction model 31. In this way, the error in a specific season is smaller than in the case of FIG. In the present embodiment, designating a specific period (t4 to t7) in this way is called “period designation”, and outliers are excluded from the measured values in that period, and the range of what should be left is designated. This is called “range specification”.

  FIG. 7 is also a diagram for explaining period designation and range designation. FIG. 7 is obtained by deleting the actual measurement values in the period other than the times t13 to t17 from FIG. The slanted line at time t15 also indicates “outlier”. The same applies to “◯” in the right column.

  FIG. 8 is obtained by deleting the points ● indicating the measured values in the period other than the times t13 to t17 from FIG. There are 4 ● and 1 ○. This circle corresponds to time t15 in FIG. Then, the prediction model creation device 1 creates a prediction model 31c using four of the remaining points ●. As a result of recalculating (correcting) the parameters, the prediction model 31 c is different in position and inclination from the prediction model 31. In this way, the error in other specific seasons becomes smaller than in the case of FIG.

(Screen display)
9 to 11 are examples of screens that display actual measurement values and predicted values. The prediction model creation device 1 displays a display screen 41 (FIG. 9) on the output device 13. The display screen 41 has a coordinate plane. The horizontal axis represents time, and the vertical axis represents the flocculant injection rate. Two time-series graphs 42 and 43 are drawn on the coordinate plane. The time series graph 42 (solid line) shows the actual measurement value of the coagulant injection rate, and the time series graph 43 (broken line) shows the predicted value of the coagulant injection rate. The time series graph 43 is a drawing of the output value of the prediction model 31.

  Ideally, the time series graph 42 and the time series graph 43 substantially overlap over the entire period. However, there is actually a period in which the error “Y−y” is widened to a degree that cannot be ignored. Note that “the degree that cannot be ignored” is the degree (significant degree) that the user makes a determination according to a predetermined standard.

  The display screen 41 in FIG. 10 displays the same coordinate plane, time series graph 42 and time series graph 43 as in FIG. Here, the user intends to further reduce the error in the periods t4 to t7 and the error in the periods t13 to t17. At the same time, the user wants to ignore extremely large measured values in the period t4 to t7 and extremely small measured values in the period t13 to t17. Therefore, the user draws the dashed-dotted rectangular windows 44a and 44b on the screen with the input device 12 such as a mouse. The prediction model creation apparatus 1 accepts this drawing.

  The left end of the window 44a coincides with time t4, and the right end coincides with time t7. The lower end of the window 44a matches the coagulant injection rate r1, and the upper end matches the coagulant injection rate r2. A part of the time series graph 42 protrudes above the coagulant injection rate r2. The left end of the window 44b coincides with time t13, and the right end coincides with time t17. The lower end of the window 44b matches the coagulant injection rate r1, and the upper end matches the coagulant injection rate r2. A part of the time series graph 42 protrudes below the coagulant injection rate r1.

  The periods t4 to t7 and the periods t13 to t17 are “period designated” periods. The ranges r1 to r2 are “range designated” ranges. Note that the level of the upper end of the window 44a does not have to coincide with the level of the upper end of the window 44b. Similarly, it is not necessary that the level at the lower end of the window 44a matches the level at the lower end of the window 44b.

  When the prediction model creation device 1 accepts that the user draws the window 44a, the prediction model creation device 1 creates the prediction model 31b of the period t4 to t7 by the method described with reference to FIGS. Similarly, when the prediction model creation device 1 accepts that the user draws the window 44b, the prediction model creation device 1 creates the prediction model 31c of the period t13 to t17 by the method described with reference to FIGS. Thereafter, the prediction model creation device 1 displays the display screen 41 (FIG. 11) on the output device 13.

Compared with FIG. 10, FIG. 11 is different as follows.
Inside the window 44a, the time series graph 48 (predicted value) is closer to the time series graph 42 (actually measured value) than the time series graph 43 (FIG. 10) (shifted upward).
Inside the window 44b, the time series graph 48 (predicted value) is closer to the time series graph 42 (actually measured value) than the time series graph 43 (FIG. 10) (shifted downward).
A parameter value column 45 and a root mean square column 46 are added.
Spin buttons 47a and 47b are displayed corresponding to the values of the four parameters.

Assume that the user selects the window 44a with the input device 12 such as a mouse. Then, the prediction model creation device 1 displays the values of the four parameters of the prediction model 31b for the period t4 to t7 in the parameter value column 45, and calculates the root mean square for the period t4 to t7. Displayed in the column 46. For convenience of explanation, examples of two variables (two dimensions) are given in FIGS. However, the prediction model creation device 1 can also extend the processing to four variables (four dimensions). Therefore, the prediction model creation apparatus 1 can calculate parameters that are multiplied by three explanatory variables (turbidity, hydrogen ion concentration, and water temperature). The root mean square is “√ (Σ (Y−y) ² / n)”, and indicates the magnitude of error per actually measured value (n = number of actually measured values).

  Assume that the user selects the window 44b with the input device 12 such as a mouse. Then, the prediction model creation device 1 displays the values of the four parameters of the prediction model 31c for the periods t13 to t17 in the parameter value column 45, and calculates the root mean square for the periods t13 to t17. Displayed in the column 46.

Now, assume that the user selects the window 44a and presses the spin button 47a for the turbidity parameter. Then, the turbidity parameter display changes from “1.2” to “1.3”. At the same time, the prediction model creation apparatus 1 changes the value of the parameter a ₁ of the prediction model 31b from “1.2” to “1.3”, and draws the time series graph 48 using the prediction model 31b. Try again. In this way, the user brings the time series graph 48 closer to the time series graph 42 little by little. The same applies to spin buttons for parameters of other explanatory variables.

(Processing procedure)
FIG. 12 is a flowchart of the processing procedure. As a premise for starting the processing procedure, it is assumed that the actual measurement information 32 (FIG. 2) is stored in the auxiliary storage device 15 in a completed state.
In step S201, the actual measurement value acquisition unit 21 of the prediction model creation device 1 acquires actual measurement values. Specifically, the actual measurement value acquisition unit 21 acquires actual measurement value information 32 (FIG. 2) from the auxiliary storage device 15.

In step S202, the variable reception unit 22 of the prediction model creation device 1 receives a variable. Specifically, the variable accepting unit 22 accepts that the user selects some or all of a plurality of explanatory variable candidates via the input device 12. For example, if the user predicts that the turbidity parameter value a ₁ of the explanatory variables will have the greatest effect on the objective variable among all parameters except a ₀ , “Turbidity” may be selected.

  In step S203, the parameter optimization unit 23 of the prediction model creation device 1 creates the prediction model 31. Specifically, the parameter optimization unit 23 accepts that the user describes a prediction model mathematical expression on the screen, or displays a typical prediction model template on the screen and accepts the user's selection. The prediction model 31 created here does not need to be a linear expression, and may be a high-order expression or a non-linear expression including an exponent, a logarithm, and the like. However, the prediction model 31 includes parameters (values are unknown at this stage) for each variable received in step S202.

In step S204, the parameter optimization unit 23 optimizes the parameters of the prediction model. Specifically, the parameter optimization unit 23 determines the parameters of the prediction model by the method described with reference to FIGS. That is, the parameter optimization unit 23 determines a parameter that minimizes “Σ (Y−y) ² ” using the actual measurement values at times t1 to t20 of the actual measurement value information 32 (FIG. 3).

  In step S205, the parameter optimization unit 23 calculates a prediction value using the prediction model. Specifically, the parameter optimization unit 23 substitutes the measured value X of the explanatory variable for each of the times t1 to t20 with respect to the prediction model in which the parameter is optimized, and acquires the predicted value y of the target variable. . At this point, the time-series measured value of the objective variable and the time-series predicted value of the objective variable are created.

  In step S206, the display processing unit 26 of the prediction model creation device 1 displays the actual measurement value and the prediction value in time series. Specifically, the display processing unit 26 displays the display screen 41 (FIG. 9) on the output device 13.

  In step S207, the period / range receiving unit 24 of the prediction model creating apparatus 1 receives a period specification and a range specification. Specifically, the period / range receiving unit 24 receives that the user draws at least one window 44a or the like (FIG. 10) on the screen.

In step S208, the actual measurement value acquisition unit 21 of the prediction model creation device 1 acquires the actual measurement value in the accepted range. Specifically, first, the actual measurement value acquisition unit 21 deletes the actual measurement values of the explanatory variables other than the explanatory variable received in step S202 from the actual measurement value information 32 (FIG. 2).
Secondly, the actual measurement value acquisition unit 21 deletes actual measurement values at times other than the period designated for the period from the actual measurement value information 32.
Thirdly, the actual measurement value acquisition unit 21 deletes actual measurement values (outliers) outside the range designated range from the actual measurement value information 32.

  In step S209, the parameter correction unit 25 of the prediction model creation device 1 corrects the parameters of the prediction model. Specifically, the parameter correction unit 25 optimizes the parameters of the prediction model using the actually measured values that are not deleted in step S208.

In step S210, the display processing unit 26 of the prediction model creation apparatus 1 redisplays the actual measurement value and the prediction value in time series. Specifically, the display processing unit 26 displays the display screen 41 (FIG. 11) on the output device 13. At this time, the display processing unit 26 may display the following three time series graphs or the time series graph 42 and the time series graph 48 on the same screen of the output device 13. In this way, the user can easily visually recognize that the predicted value of the drug injection rate approaches the actual measurement value.
-Time series graph 42 showing actual measured values of drug injection rate
Time-series graph 43 showing the predicted value of the drug injection rate output by the prediction model with optimized parameters
A time-series graph 48 showing the predicted value of the drug injection rate output by the further corrected prediction model after the parameters are optimized

  When the user changes the value of the parameter by operating the spin button 47a or the like while the display screen 41 (FIG. 11) is being displayed, the parameter correction unit 25 applies the prediction model whose parameter has been corrected in step S209. Substitute the value of the parameter after the change. Then, the parameter correction unit 25 substitutes the actually measured value in the accepted range for the prediction model after the substitution, and calculates the prediction value of the objective variable. The display processing unit 26 further displays a time series graph 49 (not shown) indicating the predicted value of the objective variable.

In step S207, the period / range receiving unit 24 may receive only one of the period designation and the range designation. As a result, in step S208, the actual measurement value acquisition unit 21 omits either “second” or “third” processing. Further, immediately after the end of step S206, the display processing unit 26 may display a parameter value field 45, a root mean square field 46, a spin button 47a, and the like. When the user changes the turbidity from “1.2” to “1.3” at this stage, for example, the parameter optimization unit 23 changes the value of the parameter a ₁ of the prediction model 31 from “1.2” to “1.3”. Then, the time series graph 43 is redrawn using the prediction model 31.

In this embodiment, for convenience of explanation, the parameter optimization unit 23 is responsible for changing the parameters of the prediction model before accepting the period specification and / or range specification, and the processing for changing the parameters of the prediction model after receiving the specification. Is assigned to the parameter correction unit 25.
Thereafter, the processing procedure ends.

(Effect of this embodiment)
The effects of the prediction model creation device of this embodiment are as follows.
(1) The prediction model creation device can create a prediction model with high accuracy locally.
(2) The prediction model creation device enables the user to confirm that the predicted value approaches the actual measured value while looking at the screen.
(3) The prediction model creation device enables the user to confirm more specifically in the time series graph that the predicted value approaches the actual measurement value while looking at the screen.
(4) The prediction model creation device can accept an explanatory variable that the user determines to be particularly important.
(5) The prediction model creation device can predict the amount of flocculant injected into the water purification plant.

  In addition, this invention is not limited to an above-described Example, Various modifications are included. For example, the above-described embodiments have been described in detail for easy understanding of the present invention, and are not necessarily limited to those having all the configurations described. Further, a part of the configuration of one embodiment can be replaced with the configuration of another embodiment, and the configuration of another embodiment can be added to the configuration of one embodiment. Further, it is possible to add, delete, and replace other configurations for a part of the configuration of each embodiment.

Each of the above-described configurations, functions, processing units, processing means, and the like may be realized by hardware by designing a part or all of them with, for example, an integrated circuit. Each of the above-described configurations, functions, and the like may be realized by software by interpreting and executing a program that realizes each function by the processor. Information such as programs, tables, and files for realizing each function can be stored in a recording device such as a memory, a hard disk, an SSD (Solid State Drive), or a recording medium such as an IC card, an SD card, or a DVD.
Further, the control lines and information lines are those that are considered necessary for the explanation, and not all the control lines and information lines on the product are necessarily shown. In practice, it may be considered that almost all the components are connected to each other.

DESCRIPTION OF SYMBOLS 1 Predictive model creation apparatus 2 Network 3 Water purification plant 4 Mixing pond 5 Coagulant injection device 6a Raw water 6b Mixing water 7 Flock formation pond 8a Turbidity meter 8b pH meter 8c Thermometer 8d Injection amount meter 9 Flocculant 11 Central controller 12 Input Device 13 Output device 14 Main storage device 15 Auxiliary storage device 16 Communication device 21 Actual measurement value acquisition unit 22 Variable reception unit 23 Parameter optimization unit 24 Period / range reception unit 25 Parameter correction unit 26 Display processing unit 31 Prediction model 32 Actual measurement value information

Claims (7)

An actual measurement value of the value indicating the nature of the raw water obtained by the water purification plant, and an actual measurement value acquisition unit for acquiring an actual measurement value of the value indicating the amount of the medicine injected into the raw water;
A parameter optimization unit that optimizes the parameters of the prediction model using the actual measurement value and the value indicating the nature of the raw water as an explanatory variable and the value indicating the amount of the drug as an objective variable;
A period in which a difference between a predicted value indicating the amount of the drug output from the prediction model in which the parameter is optimized and an actual measurement value indicating the amount of the drug is significantly increased, and / or an outlier A period / range accepting unit that accepts a range of measured values excluding,
A parameter correction unit that corrects the parameter using the measured value corresponding to the accepted period and / or range;
The measured value indicating the amount of the drug, and the predicted value of the value indicating the amount of the drug output by the prediction model in which the parameter is optimized are displayed in time series,
When the period / range receiving unit accepts that the user inputs the period and the range with respect to the displayed actual measurement value and predicted value, the actual measurement value of the value indicating the amount of the medicine, and the parameter A display processing unit for displaying a predicted value of a value indicating the amount of the drug output by the prediction model corrected in time series,
A prediction model creation device comprising: The display processing unit
Displaying a time series graph of actual values of the values indicating the amount of the drug and a time series graph of predicted values of the value indicating the amount of the drug output by the prediction model with the parameters corrected on the same screen. ,
The prediction model creation apparatus according to claim 1 . A variable reception unit that receives a user's selection of a plurality of candidates for the explanatory variable;
The prediction model creation device according to claim 2 characterized by things. The drug is
Agglomerates pollutants contained in the raw water,
The value indicating the nature of the raw water is
It is at least one of the turbidity in the raw water, the temperature of the raw water, and the hydrogen ion concentration of the raw water,
The prediction model creation device according to claim 3 characterized by things. An actual measurement value acquisition unit for acquiring an actual measurement value of the value indicating the property of the liquid, and an actual measurement value of the value indicating the amount of the medicine injected into the liquid;
A parameter optimization unit for optimizing a parameter of a prediction model using the actual measurement value as an explanatory variable with a value indicating the property of the liquid as an explanatory variable;
A period in which a difference between a predicted value indicating the amount of the drug output from the prediction model in which the parameter is optimized and an actual measurement value indicating the amount of the drug is significantly increased, and / or an outlier A period / range accepting unit that accepts a range of measured values excluding,
A parameter correction unit that corrects the parameter using the measured value corresponding to the accepted period and / or range;
The measured value indicating the amount of the drug, and the predicted value of the value indicating the amount of the drug output by the prediction model in which the parameter is optimized are displayed in time series,
When the period / range receiving unit accepts that the user inputs the period and the range with respect to the displayed actual measurement value and predicted value, the actual measurement value of the value indicating the amount of the medicine, and the parameter A display processing unit for displaying a predicted value of a value indicating the amount of the drug output by the prediction model corrected in time series,
A prediction model creation device comprising: The measured value acquisition unit of the prediction model creation device
Obtain an actual value of the value indicating the nature of the raw water obtained by the water purification plant, and an actual value of the value indicating the amount of the chemical injected into the raw water,
The parameter optimization unit of the prediction model creation device is
Using the measured values, optimize the parameters of the prediction model with the values indicating the properties of the raw water as explanatory variables and the values indicating the amount of the drug as objective variables,
The period / range reception unit of the prediction model creation device is:
A period in which a difference between a predicted value indicating the amount of the drug output from the prediction model in which the parameter is optimized and an actual measurement value indicating the amount of the drug is significantly increased, and / or an outlier Accept the measured value range excluding
The parameter correction unit of the prediction model creation device,
Using said measured value corresponding to the duration and / or range of the accepted, and corrects the parameter,
The display processing unit of the prediction model creation device includes:
The measured value indicating the amount of the drug, and the predicted value of the value indicating the amount of the drug output by the prediction model in which the parameter is optimized are displayed in time series,
When the period / range receiving unit accepts that the user inputs the period and the range with respect to the displayed actual measurement value and predicted value, the actual measurement value of the value indicating the amount of the medicine, and the parameter Displaying a predicted value of a value indicating the amount of the drug output by the prediction model corrected by chronologically,
The prediction model creation method of the prediction model creation apparatus characterized by this. Computer
An actual measurement value of the value indicating the nature of the raw water obtained by the water purification plant, and an actual measurement value acquisition unit for acquiring an actual measurement value of the value indicating the amount of the medicine injected into the raw water;
A parameter optimization unit that optimizes the parameters of the prediction model using the actual measurement value and the value indicating the nature of the raw water as an explanatory variable and the value indicating the amount of the drug as an objective variable;
A period in which a difference between a predicted value indicating the amount of the drug output from the prediction model in which the parameter is optimized and an actual measurement value indicating the amount of the drug is significantly increased, and / or an outlier A period / range accepting unit that accepts a range of measured values excluding,
A parameter correction unit that corrects the parameter using the measured value corresponding to the accepted period and / or range ;
The measured value indicating the amount of the drug, and the predicted value of the value indicating the amount of the drug output by the prediction model in which the parameter is optimized are displayed in time series,
When the period / range receiving unit accepts that the user inputs the period and the range with respect to the displayed actual measurement value and predicted value, the actual measurement value of the value indicating the amount of the medicine, and the parameter predictive modeling program for but to function as a display processing unit for displaying in time series prediction value of a value indicative of the amount of the drug corrected prediction model outputs.

Images