KR102386532B1 - Methods for learning and testing a multi-layer perceptron based marine environment information correction model for correcting marine environment information using wave prediction information and devices using the same - Google Patents

Abstract

The present invention relates to a method for learning a multi-layered perceptron-based marine environmental information correction model for correcting marine environmental information using wave prediction information, a testing method thereof, a learning device and a testing device using the same. The method of the present invention comprises the steps of: acquiring (i_j)^th wave prediction information and (i_j)^th observation information as a learning data set; outputting (i_j)^th marine environmental correction information; and learning a marine environmental correction information model.

Description

A method for learning and testing a multi-layer perceptron-based marine environment information correction model that calibrates marine environment information using wave prediction information, and a learning and testing apparatus using the same MARINE ENVIRONMENT INFORMATION CORRECTION MODEL FOR CORRECTING MARINE ENVIRONMENT INFORMATION USING WAVE PREDICTION INFORMATION AND DEVICES USING THE SAME}

The present invention relates to a method and a testing method for learning a multi-layer perceptron-based marine environment information correction model that corrects marine environment information using wave prediction information, and a learning apparatus and a testing apparatus using the same.

The Korea Meteorological Administration operates various weather forecasting models to produce and provide ocean and meteorological information. Among them, RWW3 (Regional Wave Watch-3), a wave prediction model in East Asian waters, plays a very important role as a reference data for storm forecasting.

Currently, the Korea Meteorological Administration issues a wind warning when offshore winds with a wind speed of 50.4 km/h (14 m/s) or higher last for more than 3 hours or when the wave height is 3 m or more, and a wind speed of 75.6 km/h (21 m/s) is issued. ) or higher, a wind storm warning is in effect if the sea wind lasts for more than 3 hours or the height of the wave is 5 m or more. When such a warning goes into effect, the operation of ships is restricted in accordance with related laws, so there is an error in the weather forecast and if the forecast is wrong, there is a possibility that damage to human life and property may occur. this is required

Previously, the method of estimating the sea wave height information calculated from the wave prediction model was chosen using the quantile mapping (QM) technique. However, the quantile mapping technique uses only wave height information to correct the model's prediction data using the cumulative distribution function. In the high wave section, the predicted sea wave height information was not corrected, but rather an error occurred.

Therefore, recently, in order to correct the error occurring in the wave prediction model, a method of correcting sea wave height information using an artificial intelligence-based learning model is emerging. Since the AI-based learning model is trained using wave and weather information over time, even if the number of samples in the high wave section is insufficient, only the significant wave height information is used to correct the sea wave height information calculated from the predictive model. Compared to the mapping method, it is a popular trend because it can obtain stable results.

Accordingly, the present applicant intends to provide an algorithm that effectively corrects the error value predicted by the wave prediction model as an artificial intelligence-based optimal learning model.

An object of the present invention is to solve all of the above problems.

Another object of the present invention is to predict marine environment correction information using wave prediction information and observation information of a specific point using a multi-perceptron-based marine environment information correction model and to learn it.

In addition, the present invention predicts environmental correction information for verification by inputting verification wave prediction information and verification observation information to a multi-perceptron-based marine environment information correction model in a state where such learning is completed, and predicts verification wave prediction The deviation and root mean square error (RMSE, Root Mean Square Error) calculated by referring to the information and observation information for verification are compared with the deviation and root mean square error calculated by referring to the environmental correction information for verification and observation information to resolve Another purpose is to verify the performance of the environmental calibration model.

In addition, the present invention is a wave prediction model using the marine environment correction information that is output when the test wave prediction information for a specific point that has not been learned is input to the marine environment correction model in a state where the performance verification of the marine environment correction model is completed Another purpose is to provide more accurate wave prediction information or to evaluate the accuracy of the test wave prediction information by improving and updating the accuracy of the test wave prediction information when the error of the wave prediction information for the test is corrected. .

In order to achieve the object of the present invention as described above and to realize the characteristic effects of the present invention to be described later, the characteristic configuration of the present invention is as follows.

According to an aspect of the present invention, in a method for learning a multi-layer perceptron-based marine environment information correction model for correcting marine environment information using wave prediction information, (a) the learning apparatus includes: m th - The m is an integer greater than or equal to 2 - At each past time point, the first wave prediction information to the m th wave predicted by using each of the first atmospheric environment information to the m th atmospheric environment information of the specific region With reference to the prediction information and the first to n-th points within the specific region, where n is an integer greater than or equal to 2, the first to m_n-th observation information for observing marine environment information at each point, i-th i is It is an integer greater than or equal to 1 and less than or equal to m - The jth of the past time - The j is an integer greater than or equal to 1 and less than or equal to n - The i_j wave prediction information, which is the wave prediction information at the point, and the i_j observation information, which is the corresponding observation information, are used as a learning data set. to obtain; (b) the learning apparatus inputs the i_j wave prediction information into a marine environment information correction model, and causes the marine environment information correction model to perform a neural network operation on the i_j wave prediction information using a preset parameter to calculate the ith outputting the i_j th marine environment correction information that predicts the correction result of the marine environment correction information at the j th point in the past; and (c) the learning device generates a loss by referring to the i_jth marine environment correction information and the i_jth observation information, and updates the parameter of the marine environment information correction model through backpropagation using the loss learning the marine environment information correction model by doing so; A method comprising

As an example, the i_jth observation information and the i_jth marine environment correction information each include at least one of significant wave height information, maximum wave height information, wave period information, and wind speed information.

As an example, the i_j observation information and the i_j sea environment correction information include significant wave height information and maximum wave height information, respectively, when the i_j wave prediction information corresponds to a wave prediction model of RWW3 (Regional Wave Watch-3). , wave period information, and wind speed information, and when the i_j wave prediction information corresponds to a wave prediction model of GWW3 (Global Wave Watch-3), a method including significant wave height information and wind speed information is disclosed.

As an example, the i_j-th wave prediction information includes at least one of significant wave height information, wave direction information, wave period information, wind speed information, water depth information, and maximum wave height information.

As an example, when the wave prediction model is RWW3, a method characterized in that the maximum wave height information is calculated by applying a Rayleigh probability density function with reference to the significant wave height information and the wave period information is disclosed.

As an example, (d) the learning device, (i) at each of the first past time point to the mth past time point, using each of the first atmospheric environment information to the mth atmospheric environment information of the specific area, The first wave prediction information to the m-th wave prediction information for predicting a wave in a specific area, and 1_1 observation information to the 1st_1st observation information to observing the marine environment information at each of the first point to the n-th point within the specific area With reference to m_n observation information, s - The s is an integer greater than or equal to 1 and less than or equal to m - The t th of the past time - The t is an integer greater than or equal to 1 and less than or equal to n - The s_t th wave prediction information, which is the wave prediction information at the point, and Acquire s_t observation information, which is the corresponding observation information, as a verification data set, (ii) input the s_t wave prediction information into the marine environment information correction model to cause the marine environment information correction model to use the updated parameters Neural network operation on the s_t wave prediction information to output the s_t sea environment correction information that predicts the correction result of the marine environment correction information at the t th point of the s th past time point, and (iii) the s_t wave A first bias and a mean squared error (RMSE) are calculated with reference to the prediction information and the s_t observation information, and a second deviation and average are calculated by referring to the s_t sea environment correction information and the s_t observation information. calculating a root-square error and verifying the performance of the marine environment information correction model with reference to the first deviation and root-mean-square error and the second deviation and root-mean-square error; Disclosed is a method characterized in that it further comprises.

According to another aspect of the present invention, in a method for testing a multi-layer perceptron-based marine environment information correction model that corrects marine environment information using wave prediction information, (a) by a learning device, (i) 1 Past time point to m - The m is an integer greater than or equal to 2 - At each past time point, the wave of the specific area for learning is predicted using each of the first atmospheric environment information for learning or the mth atmospheric environment information for learning of the specific area for learning. 1st wave prediction information or m-th wave prediction information for learning, and a first point for learning within the specific area to n-th for learning - wherein n is an integer greater than or equal to 2 - 1 _1 observation information for learning or learning for observing marine environment information at each point With reference to m_n-th observation information, i-th i is an integer greater than or equal to 1 and less than or equal to m - j-th of the past time - the j is an integer greater than or equal to 1 and less than or equal to n - i_j wave prediction information for learning that is wave prediction information at a point and a process of acquiring i_j observation information for learning, which is observation information corresponding thereto, as a learning data set; (ii) input the i_j wave prediction information for learning into the marine environment information correction model to cause the marine environment information correction model to calculate the i_j wave prediction information for learning by using a preset parameter with a neural network of the i-th past time point a process of outputting the i_j-th marine environment correction information for learning that predicted the correction result of the marine environment correction information for learning at the j-th point, and (iii) refer to the i_j sea environment correction information for learning and the i_j observation information for learning to generate a loss, and in a state in which the process of learning the marine environment information correction model by updating the parameters of the marine environment information correction model through backpropagation using the loss is performed, the testing device is configured in the specific area obtaining test wave prediction information for predicting a wave in the specific area by using the test atmospheric environment information; And (b) the testing device, with reference to the wave prediction information for the test, the first point for the test to the k-th for testing within the specific area - The k is an integer greater than or equal to 1 - First wave prediction information for testing at the point to obtain the k-th wave prediction information for the test, and input the first wave prediction information for the test to the k-th wave prediction information for the test into the marine environment information correction model to cause the marine environment information correction model to update the parameters By using the neural network calculation of the first wave prediction information for the test to the k-th wave prediction information for the test, the first marine environment correction information for testing corresponding to each of the first point for the test to the k-th point for the test using outputting k-th marine environment correction information for testing; A method comprising

As an example, (c) the testing device updates the wave prediction information for the test by using the first sea environment correction information for the test to the k-th sea environment correction information for the test, or the first sea environment for the test evaluating the accuracy of the blue prediction information for the test with reference to the correction information or the k-th marine environment correction information for the test; A method further comprising the is disclosed.

As an example, the test wave prediction information, significant wave height information, wave direction information, wave period information, wind speed information, water depth information, and a method including at least one of the maximum wave height information is disclosed.

According to another aspect of the present invention, there is provided a learning apparatus for learning a multi-layer perception-based marine environment information correction model that corrects marine environment information using wave prediction information, comprising: at least one memory for storing instructions; and at least one processor configured to execute the instructions, wherein the processor comprises: (I) a first past time point to m th - wherein m is an integer greater than or equal to 2 - at each past time point, a first wait in a specific region 1st wave prediction information to mth wave prediction information for predicting a wave in the specific area using each of the environment information to mth atmospheric environment information, and 1st to nth - the n in the specific area, are integers greater than or equal to 2 - With reference to the 1st_1st observation information to m_nth observation information observing the marine environment information at each point, i - The i is an integer greater than or equal to 1 and less than or equal to m - The jth of the past time - The j is greater than or equal to 1 and less than or equal to n Integer - a process of acquiring, as a learning data set, i_j-th wave prediction information, which is wave prediction information at a point, and i_j-th observation information, which is observation information corresponding thereto; (II) input the i_j wave prediction information into a marine environment information correction model to cause the marine environment information correction model to perform a neural network operation on the i_j wave prediction information using a preset parameter to calculate the j-th of the i-th past time point a process of outputting the i_jth marine environment correction information in which the correction result of the marine environment correction information at the point is predicted; and (III) generating a loss with reference to the i_jth marine environment correction information and the i_j observation information, and updating the parameters of the marine environment information correction model through backpropagation using the loss. Disclosed is a learning apparatus characterized by performing; a process of learning a correction model.

As an example, the i_jth observation information and the i_jth marine environment correction information each include at least one of significant wave height information, maximum wave height information, wave period information, and wind speed information.

As an example, the i_j observation information and the i_j sea environment correction information include significant wave height information and maximum wave height information, respectively, when the i_j wave prediction information corresponds to a wave prediction model of RWW3 (Regional Wave Watch-3). , wave period information, and wind speed information, and when the i_j wave prediction information corresponds to a wave prediction model of GWW3 (Global Wave Watch-3), a learning apparatus including significant wave height information and wind speed information is disclosed. .

As an example, the i_j wave prediction information, a learning apparatus including at least one of significant wave height information, wave direction information, wave period information, wind speed information, water depth information, and maximum wave height information is disclosed.

As an example, when the wave prediction model is RWW3, the maximum wave height information is applied to a Rayleigh probability density function with reference to the significant wave height information and the wave period information, and a learning apparatus is disclosed.

As an example, the processor (IV) (i) at each of the first past time point to the mth past time point, using each of the first atmospheric environment information to the mth atmospheric environment information of the specific area, The first wave prediction information to the m-th wave prediction information for predicting a wave of an area, and 1_1 first observation information to m_n-th observation information for observing the marine environment information at each of the first point to the n-th point within the specific area With reference to the observation information, s th - The s is an integer greater than or equal to 1 and less than or equal to m - The t th of the past time - The t is an integer greater than or equal to 1 and less than or equal to n - The s_t wave prediction information corresponding to the wave prediction information at the point Obtaining the s_t observation information, which is the observation information to be obtained as a verification data set, (ii) inputting the s_t wave prediction information into the marine environment information correction model to cause the marine environment information correction model to use the updated parameters Neural network operation on the s_t wave prediction information to output the s_t sea environment correction information that predicted the correction result of the marine environment correction information at the t th point of the s th past time point, and (iii) the s_t wave prediction A first bias and a mean squared error (RMSE) are calculated with reference to the information and the s_t observation information, and a second deviation and a root mean squared error are calculated with reference to the s_t sea environment correction information and the s_t observation information. An error is calculated, and a process of verifying the performance of the marine environment information correction model with reference to the first deviation and the root mean square error and the second deviation and the root mean square error is further performed. .

According to another aspect of the present invention, there is provided a testing apparatus for testing a multi-layer perception-based marine environment information correction model that corrects marine environment information using wave prediction information, comprising: at least one memory for storing instructions; and at least one processor configured to execute the instructions, wherein the processor is configured to: (I) by the learning device, (i) a first past time point to m th, wherein m is an integer greater than or equal to 2 - past time points, respectively In, the first wave prediction information for learning or the m-th wave prediction information for learning that predicted the wave of the specific area for learning using each of the first atmospheric environment information for learning and the m-th atmospheric environment information for learning of the specific area for learning, and the specific area 1st point for learning to nth for learning in - Where n is an integer of 2 or more It is an integer less than or equal to m - The jth of the past time - The j is an integer greater than or equal to 1 and less than or equal to n - The i_j wave prediction information for learning that is the wave prediction information at the point and the i_j observation information for learning that is the corresponding observation information are used as a learning data set the process of obtaining; (ii) input the i_j wave prediction information for learning into the marine environment information correction model to cause the marine environment information correction model to calculate the i_j wave prediction information for learning by using a preset parameter with a neural network of the i-th past time point a process of outputting the i_j-th marine environment correction information for learning that predicted the correction result of the marine environment correction information for learning at the j-th point, and (iii) refer to the i_j sea environment correction information for learning and the i_j observation information for learning to generate a loss, and in a state in which the process of learning the marine environment information correction model by updating the parameters of the marine environment information correction model through backpropagation using the loss is performed, the test standby of the specific area a process of obtaining wave prediction information for a test in which a wave of the specific area is predicted using environment information; and (II) the first point for testing to the k-th for testing within the specific area with reference to the wave prediction information for the test - The k is an integer greater than or equal to 1 - The first wave prediction information for testing at the point to the k-th for testing Obtaining wave prediction information, inputting the first wave prediction information for the test to the k-th wave prediction information for the test into the marine environment information correction model causes the marine environment information correction model to perform the test using the updated parameters For the first wave prediction information for the test, the neural network calculates the k-th wave prediction information for the test, and the first sea environment correction information for the test or the k-th sea for the test corresponding to each of the first point for the test to the k-th point for the test A testing apparatus for performing a process of outputting environmental correction information is disclosed.

As an example, the processor (III) updates the wave prediction information for the test using the first sea environment correction information for the test to the k-th sea environment correction information for the test, or corrects the first sea environment for the test a process of evaluating the accuracy of the wave prediction information for the test with reference to the information or the k-th marine environment correction information for the test; A testing apparatus for further performing is disclosed.

As an example, the wave prediction information for the test, a testing apparatus including at least one of significant wave height information, wave direction information, wave period information, wind speed information, water depth information, and maximum wave height information is disclosed.

The present invention has an effect of predicting marine environment correction information using wave prediction information and observation information of a specific point using a multi-perceptron-based marine environment information correction model and learning it.

In addition, the present invention predicts environmental correction information for verification by inputting verification wave prediction information and verification observation information to a multi-perceptron-based marine environment information correction model in a state where such learning is completed, and predicts verification wave prediction The deviation and root mean square error (RMSE, Root Mean Square Error) calculated by referring to the information and observation information for verification are compared with the deviation and root mean square error calculated by referring to the environmental correction information for verification and observation information to resolve It has the effect of verifying the performance of the environmental correction model.

In addition, the present invention is a wave prediction model using the marine environment correction information that is output when the test wave prediction information for a specific point that has not been learned is input to the marine environment correction model in a state where the performance verification of the marine environment correction model is completed If the error of the wave prediction information for the test calculated from is corrected, the accuracy of the wave prediction information for the test is improved and updated, thereby providing more accurate wave prediction information or evaluating the accuracy of the wave prediction information for the test.

1 is a diagram illustrating an apparatus for learning a multi-layer perceptron-based marine environment information correction model for correcting marine environment information using wave prediction information according to an embodiment of the present invention.
FIG. 2 is a diagram illustrating a flow chart in which a learning apparatus according to an embodiment of the present invention learns a multi-layer perceptron-based marine environment information correction model for correcting marine environment information using wave prediction information.
3 is a diagram illustrating a process in which a multi-layer perceptron-based marine environment correction information model for correcting a marine environment using wave prediction information according to an embodiment of the present invention is learned.
4A and 4B are examples of a list of specific points selected for learning and verification of a multi-layer perceptron-based marine environment correction information model that corrects the marine environment using wave prediction information according to an embodiment of the present invention. the drawing shown.
5a to 5f show significant wave height, maximum wave height, period, wind speed for each verification point and prediction time when the wave prediction model is RWW3 in a state where the learning of the marine environment correction information model according to an embodiment of the present invention is completed; It is a diagram verifying the performance of the marine environment correction model by calculating the deviation and root mean square error of the wave prediction model and the marine environment information correction model for .
6a to 6f show a wave prediction model for significant wave height and wind speed by verification point and prediction time when the wave prediction model is GWW3 in a state where the learning of the marine environment correction information model according to an embodiment of the present invention is completed; It is a diagram verifying the performance of the marine environment correction model by calculating the deviation and root mean square error of the marine environment correction information model.
7A and 7B are diagrams illustrating an example of a list of specific points selected for verification of unlearned points of a multi-layer perceptron-based marine environment correction information model that corrects the marine environment using wave prediction information.
8a to 8d show significant wave heights for each selected verification point and prediction time among non-learning points when the wave prediction model is RWW3 in a state where the learning of the marine environment correction information model according to an embodiment of the present invention is completed. It is a diagram showing the verification of the performance of the marine environment information correction model by calculating the deviation and root mean square error between the wave prediction model and the marine environment correction model for Korea.
9 is a diagram illustrating an apparatus for testing a multi-layer perceptron-based marine environment information correction model that corrects marine environment information using wave prediction information according to an embodiment of the present invention.
10 is a process of testing a multi-layer perceptron-based marine environment correction information model that corrects the marine environment using wave prediction information in a state in which verification of the marine environment correction information model according to an embodiment of the present invention is completed; the drawing shown.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS [0012] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS [0010] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS [0010] Reference is made to the accompanying drawings, which show by way of illustration specific embodiments in which the present invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the present invention. It should be understood that the various embodiments of the present invention are different but need not be mutually exclusive. For example, certain shapes, structures, and characteristics described herein with respect to one embodiment may be implemented in other embodiments without departing from the spirit and scope of the invention. In addition, it should be understood that the location or arrangement of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the present invention. Accordingly, the detailed description set forth below is not intended to be taken in a limiting sense, and the scope of the present invention, if properly described, is limited only by the appended claims, along with all scope equivalents as those claimed. Like reference numerals in the drawings refer to the same or similar functions throughout the various aspects.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings in order to enable those of ordinary skill in the art to easily practice the present invention.

1 is a diagram illustrating an apparatus for learning a multi-layer perceptron-based marine environment information correction model for correcting marine environment information using wave prediction information according to an embodiment of the present invention.

The memory 110 of the learning apparatus 100 may store instructions to be executed by the processor 120 . Specifically, the instructions are a code generated for the purpose of causing the learning apparatus 100 to function in a specific manner. , which may be stored in a computer-usable or computer-readable memory that may be directed to a computer or other programmable data processing equipment. The instructions may perform processes for performing the functions described herein.

In addition, the processor 120 of the learning apparatus 100 may include a hardware configuration such as a micro processing unit (MPU) or a central processing unit (CPU), a cache memory, and a data bus. In addition, the operating system may further include a software configuration of an application for performing a specific purpose.

Also, the learning apparatus 100 may be linked with a database (not shown). Here, the database (not shown) includes a flash memory type, a hard disk type, a multimedia card micro type, and a card type memory (eg, SD or XD memory). , Random Access Memory (RAM), Static Random Access Memory (SRAM), ReadOnly Memory, ROM, Electrically Erasable Programmable ReadOnly Memory (EEPROM), Programmable ReadOnly Memory (PROM), Magnetic Memory, Magnetic Disk, Optical Disk It may include at least one type of storage medium, but is not limited thereto, and may include any medium capable of storing data. In addition, the database (not shown) may be installed separately from the learning device 100 , or alternatively, it may be installed inside the learning device 100 to transmit data or record received data. may be implemented separately, which may vary depending on the operating conditions of the invention.

FIG. 2 is a diagram illustrating a flow chart in which a learning apparatus according to an embodiment of the present invention learns a multi-layer perceptron-based marine environment information correction model for correcting marine environment information using wave prediction information.

Referring to FIG. 2 , the learning apparatus 100 obtains the blue wave of the specific region using each of the first atmospheric environment information to the mth atmospheric environment information of the specific region at each of the first past time point to the mth past time point. With reference to the predicted first wave prediction information to the m-th wave prediction information, and the first to m_n-th observation information for observing the marine environment information at each of the first to n-th points within the specific area, the ith past The i_j-th wave prediction information, which is the wave prediction information at the j-th point of the viewpoint, and the i_j-th observation information, which is observation information corresponding thereto, may be acquired as a learning data set ( S210 ).

Then, by inputting the i_j wave prediction information to the marine environment information correction model, the marine environment information correction model uses a preset parameter to perform a neural network operation on the i_j wave prediction information to the j-th point of the i-th past time point It is possible to output the i_j th marine environment correction information in which the correction result of the marine environment correction information is predicted ( S220 ).

In addition, the marine environment information correction model by generating a loss with reference to the i_j sea environment correction information and the i_j observation information, and updating the parameters of the marine environment information correction model through backpropagation using the loss can be learned (S230).

In order to describe the process of FIG. 2 in more detail, the following description will be made with reference to FIGS. 3 to 6F .

3 is a diagram illustrating a process in which a multi-layer perceptron-based marine environment correction information model for correcting a marine environment using wave prediction information according to an embodiment of the present invention is learned.

Referring to FIG. 3 , the wave prediction information 310 and observation information 320 input as a training data set to the marine environment information correction model 330 is a first past time point to m-th where m is an integer greater than or equal to 2 - In each of the past time points, the first wave prediction information to the mth wave prediction information for predicting the wave of the specific area using each of the first atmospheric environment information to the mth atmospheric environment information of the specific area, and the first in the specific area Point to nth - The n is an integer of 2 or more - With reference to the 1_1st observation information to the m_nth observation information observing the marine environment information at each point, i - The i is an integer greater than or equal to 1 and less than or equal to m - of the past time point j-th - The j is an integer greater than or equal to 1 and less than or equal to n - It can be divided into i_j-th wave prediction information, which is wave prediction information at a point, and i_j-th observation information, which is observation information corresponding thereto.

For example, in the present invention, the wave prediction information 310 and the observation information 320 for learning utilize data from 2017 to 2019, and the j may be 16 with reference to FIGS. 4A and 4B . . The m and j may be used by adding or subtracting an appropriate number according to the judgment of those skilled in the art.

In addition, the wave prediction information 310 includes at least one of significant wave height information, wave direction information, wave period information, wind speed information, water depth information, and maximum wave height information, and the observation information 320 includes significant wave height information and maximum wave height information. It may include at least one of information, wave period information, and wind speed information.

Among the factors of the wave prediction information 310, the depth information is directly input to the marine environment correction model 330, and the significant wave height information, the wave direction information, the wave period information, and the wind speed information are calculated from the wave prediction model. can be

In addition, when the wave prediction model is Regional Wave Watch-3 (RWW3), the maximum wave height information is to be calculated by applying to a Rayleigh probability density function with reference to the significant wave height information and the wave period information. can

The marine environment correction model 330 to which the learning data set of the wave prediction information 310 and the observation information 320 is input is a multi-layer perceptron (MLP, Multi-Layer Perceptron)-based deep learning model was applied. The marine environment correction information 340 that predicts the correction result of the marine environment correction information may be output by performing a neural network operation on the wave prediction information 310 using a preset parameter.

The outputted marine environment correction information 340 includes at least one of significant wave height information, maximum wave height information, wave period information, and wind speed information when the wave prediction model is the RWW3, and the wave prediction model is GWW3 (Global Wave Watch-3) may include at least one of significant wave height information and wind speed information.

The outputted marine environment correction information 340 generates a loss by referring to the observation information 320 in the loss layer 350, and the marine environment information correction model 330 through backpropagation using the loss. When the parameter is optimized by repeating the process of updating the parameter, the learning of the marine environment information correction model 330 may be completed.

4A and 4B are examples of a list of specific points selected for learning and verification of a multi-layer perceptron-based marine environment correction information model that corrects the marine environment using wave prediction information according to an embodiment of the present invention. the drawing shown.

First, referring to FIGS. 4A and 4B , it is an example of a total of 16 observation points 410 for learning, and for each of the observation points 410, an observation point id, a point name (English name), latitude, longitude, and observation element information etc. can be checked.

In the case of the observation element, wave environment information such as significant wave height information, maximum wave height information, average wave height information, period and wave direction, and weather environment information such as wind direction, wind speed, atmospheric pressure, humidity, temperature and water temperature may be observed.

In order to verify the marine environment information correction model learned using the training data set as described above, a learning device or a verification device corresponding thereto, at each of the first past time point to the mth past time point, is the The first wave prediction information to the m-th wave prediction information for predicting a wave in the specific region using each of the first atmospheric environment information to the m-th atmospheric environment information, and the first point to the n-th point within the specific region With reference to the 1st_1st observation information to the m_nth observation information that observed the marine environment information at each point, s - the s is an integer greater than or equal to 1 and less than or equal to m - t of the past time - the t is greater than or equal to 1 and less than or equal to n Integer - It is possible to obtain s_t wave prediction information, which is wave prediction information at a point, and s_t-th observation information, which is observation information corresponding thereto, as a verification data set.

In this case, the verification data set may be sampled from the same training data database from which the training data set is sampled so that data classification in each data set is the same or similar.

Thereafter, the learning device or the verification device inputs the s_t wave prediction information to the marine environment information correction model to cause the marine environment information correction model to perform a neural network operation on the s_t wave prediction information using the updated parameters, and the The s_t th marine environment correction information, which predicts the correction result of the marine environment correction information at the t th point of the s th time in the past, is output.

Then, the learning device or verification device calculates a first bias and a mean square error (RMSE) with reference to the s_t wave prediction information and the s_t observation information, and the s_t sea environment correction information and A second deviation and root mean square error are calculated with reference to the s_t observation information, and the performance of the marine environment information correction model is verified by referring to the first deviation and root mean square error and the second deviation and root mean square error. can

5a to 5f show significant wave height, maximum wave height, period, wind speed for each verification point and prediction time when the wave prediction model is RWW3 in a state where the learning of the marine environment correction information model according to an embodiment of the present invention is completed; It is a diagram verifying the performance of the marine environment correction model by calculating the deviation and root mean square error of the wave prediction model and the marine environment information correction model for .

5A and 5B, the same point as the observation point 410 applied to learning in FIGS. 4A and 4B was verified, and wave prediction information and observation information predicted by RWW3 for significant wave height information for each point By comparing each of the deviation and root mean square error (RMSE) values calculated by referring to the marine environment correction information and the deviation and root mean square error values calculated by referring to the observation information, The calibration model can be verified.

In addition, when checking the graph, the deviation and the root mean square error value calculated by referring to the marine environment correction information and the observation information for each point are calculated by referring to the wave prediction information and the observation information predicted by the RWW3 Comparing with the deviation and the root mean square error, since the deviation and the root mean square error calculated from the marine environment correction information as a whole are small, the correction performance of the marine environment correction model may be confirmed.

5C, the deviation and the root mean square error value calculated by referring to the wave prediction information predicted by RWW3 and the observation information for the maximum wave height information, period information, and wind speed information for each point, and marine environment correction information and The marine environment correction model may be verified by comparing each of the deviation calculated with reference to the observation information and the root mean square error value.

In addition, when checking the graph, the deviation calculated by referring to the marine environment correction information and the observation information for the maximum wave height information and the period information was compared with the deviation calculated by referring to the RWW3 and the observation information. , decreased at most points, and the wind speed information showed a characteristic that the deviation was converted from a positive value to a negative value, and when the root mean square error value was compared, the maximum wave height information, the period information, and the wind speed All of the information is reduced, so that it is possible to confirm the correction performance of the marine environment correction model.

5D and 5E , the deviation and root mean square error values calculated by referring to the wave prediction information predicted by the RWW3 and the observation information for the significant wave height information for each significant wave height section and for each prediction time, and the The marine environment correction model may be verified by comparing each of the deviation and the root mean square error value calculated with reference to the marine environment correction information and the observation information.

In addition, when checking the graph, the deviation of the prediction information corrected in the marine environment correction information increased slightly in the 0 to 3 m and entire section, and the deviation slightly increased at the beginning of the prediction in the 3 to 5 m section, but decreased thereafter It can be confirmed that, in the section of 5 m or longer, the deviation decreased at the beginning of the prediction, but it was confirmed that the deviation increased as time passed. The root mean square error value may decrease for each prediction time in all wave height sections, thereby confirming the correction performance of the marine environment correction model.

Referring to FIG. 5F , the deviation and root mean square error values calculated by referring to the wave prediction information and the observation information predicted by the RWW3 for the maximum wave height information, period information, and wind speed information for each predicted time, and the marine environment correction The marine environment correction model may be verified by comparing each of the deviation and the root mean square error value calculated with reference to the information and the observation information.

In addition, when the graph is checked, the deviation and the root mean square error value are reduced in all of the maximum wave height information, the period information, and the wind speed information regardless of the predicted time, thereby confirming the correction performance of the marine environment correction model.

6a to 6f show a wave prediction model for significant wave height and wind speed by verification point and prediction time when the wave prediction model is GWW3 in a state where the learning of the marine environment correction information model according to an embodiment of the present invention is completed; It is a diagram verifying the performance of the marine environment correction model by calculating the deviation and root mean square error of the marine environment correction information model.

Referring to FIG. 6A, the same point as the observation point 410 applied to learning in FIGS. 4A and 4B was verified, and the wind speed information for each point was calculated by referring to the wave prediction information and observation information predicted by GWW3. The marine environment correction model may be verified by comparing the calculated deviation and root mean square error value, the marine environment correction information, and the deviation and the root mean square error value calculated with reference to the observation information, respectively.

In addition, when checking the graph, the deviation and the root mean square error value calculated by referring to the marine environment correction information and the observation information for each point are calculated by referring to the wave prediction information and the observation information predicted by the GWW3 Compared with the deviation and the root mean square error, the deviation value slightly increased only at some points, but the deviation calculated from the marine environment correction information and the root mean square error value decreased as a whole, so that the correction of the marine environment correction model You can check the performance.

6b and 6c, the deviation and root mean square error values calculated by referring to the wave prediction information and observation information predicted by GWW3 for the significant wave height information for each point, and the marine environment correction information and the observation information are calculated by referring to the observation information The marine environment correction model may be verified by comparing each of the calculated deviation and the root mean square error value.

In addition, when checking the graph, the deviation calculated by referring to the marine environment correction information and the observation information and the wave prediction information predicted by the GWW3 for the root mean square error value and the deviation calculated by referring to the observation information and compared with the root mean square error, the deviation of the marine environment correction information and the root mean square error value are reduced for all wave height sections, thereby confirming the correction performance of the marine environment correction model.

6D and 6E, the deviation and root mean square error values calculated by referring to the wave prediction information predicted by the GWW3 with respect to the significant wave height information for each prediction time and the observation information, and the marine environment correction information and The marine environment correction model may be verified by comparing each of the deviation calculated with reference to the observation information and the root mean square error value.

In addition, when checking the graph, the deviation of the prediction information corrected in the marine environment correction information was converted from a negative value to a positive value in the 0 to 3m section. In the interval of 3-5 m, the deviation decreased regardless of the prediction time, and in the interval of 5 m or longer, the deviation slightly increased at the beginning of the prediction, but it can be seen that the deviation decreased as time passed. The root mean square error value may decrease for each prediction time in all wave height sections, thereby confirming the correction performance of the marine environment correction model.

6F, the deviation and root mean square error values calculated by referring to the wave prediction information predicted by the GWW3 and the observation information for the wind speed information for each prediction time, the marine environment correction information, and the observation information The marine environment correction model may be verified by comparing each of the calculated deviation and the root mean square error value with reference to each other.

Also, when the graph is checked, the wind speed information reduces the deviation and the root mean square error value regardless of the predicted time, so that the correction performance of the marine environment correction model may be confirmed.

On the other hand, after the above verification, it can be confirmed whether the marine environment correction model operates correctly at points other than the respective points.

As an example, FIGS. 7A and 7B are diagrams illustrating an example of a list of specific points selected for verification of unlearned points of a multi-layer-based marine environment correction information model that corrects the marine environment using wave prediction information.

Referring to FIGS. 7A and 7B , a total of 16 observation points 410 used for the learning and verification can be identified, and 12 additionally selected non-learning points are selected for verification of non-learning points, of which 10 are the oceans. The surveyor observation point 710, and the other two locations are the Korea Meteorological Administration observation point 720. Observation point id, point name (English name), latitude, longitude information, etc. can be confirmed for each of the observation point 710 and the observation point 720 of the Korea Meteorological Agency.

8a to 8d show significant wave heights for each selected verification point and prediction time among non-learning points when the wave prediction model is RWW3 in a state where the learning of the marine environment correction information model according to an embodiment of the present invention is completed. It is a diagram showing the verification of the performance of the marine environment information correction model by calculating the deviation and root mean square error between the wave prediction model and the marine environment correction model for Korea.

Referring to FIGS. 8A and 8B , the 12 marine surveyor observation points 710 and the Korea Meteorological Administration observation points 720 selected in FIGS. 7A and 7B were verified, and RWW3 for significant wave height information for each point A marine environment correction model by comparing the deviation and root mean square error values calculated with reference to the wave prediction information predicted in the can check the calibration performance.

In addition, when checking the graph, the deviation and the root mean square error value calculated with reference to the marine environment correction information and the observation information for each of the observation point 710 and the observation point 720 of the Korea Meteorological Agency are calculated in the RWW3. Comparing the predicted wave prediction information and the observed information with reference to the deviation and the root mean square error, the deviation value slightly increased in the 0~1m and 1~2m sections, and the deviation value increased toward the high wave section This decreasing trend can be observed. In addition, the root mean square error value slightly increased in the 0~1m, 1~2m, and 2~3m sections, and it was confirmed that the root mean square error value decreased more significantly as the high wave section increased, so that the correction of the marine environment correction model performance can be verified.

8C and 8D , the deviation and root mean square error values calculated by referring to the wave prediction information predicted by the RWW3 and the observation information for the significant wave height information for each prediction time, and the marine environment correction information and the observation The correction performance of the marine environment correction model may be confirmed by comparing each of the deviation calculated with reference to the information and the root mean square error value.

In addition, when checking the graph, the average of the deviation and the root mean square error calculated by referring to the marine environment correction information and the observation information at the observation point 710 and the observation point 720 of the Korea Meteorological Agency for each predicted time When the value is compared with the average value of the deviation and the root mean square error calculated by referring to the wave prediction information predicted by the RWW3 and the observation information, the deviation and the root mean square error are irrespective of the prediction time in the 0 to 3m section. It can be seen that both the average values of the errors slightly increased, and in the 3-5 m section, the average values of the deviations increased slightly at the beginning of the prediction and decreased after 36 hours, and it can be seen that the average values of the root mean square errors decreased. In addition, similar to the 3m section in the 5m or longer section, it was confirmed that the average value of the deviation slightly increased at the beginning of the prediction and then decreased after a certain period of time, and the average value of the root mean square error tends to decrease regardless of the prediction time. This is confirmed to confirm the correction performance of the marine environment correction model.

As a result, it can be seen that the marine environment information correction model according to the present invention outputs a more accurate correction value for specific elements of the wave than the wave prediction model for the entire specific region.

9 is a diagram illustrating a testing apparatus for a multi-layer perceptron-based marine environment information correction model that corrects marine environment information using wave prediction information according to an embodiment of the present invention.

The memory 910 of the testing apparatus 900 may store instructions to be performed by the processor 920 , specifically, the instructions are code generated for the purpose of causing the testing apparatus 900 to function in a particular manner. , which may be stored in a computer-usable or computer-readable memory that may be directed to a computer or other programmable data processing equipment. The instructions may perform processes for performing the functions described herein.

In addition, the processor 920 of the testing apparatus 900 may include a hardware configuration such as a micro processing unit (MPU) or a central processing unit (CPU), a cache memory, and a data bus. In addition, the operating system may further include a software configuration of an application for performing a specific purpose.

Also, the testing device 900 may be linked with a database (not shown). Here, the database (not shown) includes a flash memory type, a hard disk type, a multimedia card micro type, and a card type memory (eg, SD or XD memory). , Random Access Memory (RAM), Static Random Access Memory (SRAM), ReadOnly Memory, ROM, Electrically Erasable Programmable ReadOnly Memory (EEPROM), Programmable ReadOnly Memory (PROM), Magnetic Memory, Magnetic Disk, Optical Disk It may include at least one type of storage medium, but is not limited thereto, and may include any medium capable of storing data. In addition, the database (not shown) may be installed separately from the testing device 900 , or alternatively, it may be installed inside the testing device 900 to transmit data or record data received, and, unlike shown, two or more may be implemented separately, which may vary depending on the operating conditions of the invention.

10 is a diagram illustrating a process of testing a multi-layer-based marine environment correction information model that corrects the marine environment using wave prediction information in a state where verification of the marine environment correction information model according to an embodiment of the present invention is completed; It is a drawing.

Referring to FIG. 10 , the marine environment correction model 330 predicts a wave for a test in which the testing device predicts the wave of the specific area by using the atmospheric environment information for the test of the specific area in a state where learning and verification are completed information can be obtained.

And, the testing device, with reference to the wave prediction information for the test, the first point for the test in the specific area to the k-th for the test - the k is an integer greater than or equal to 1 - the first wave prediction information for the test at the point Obtain the k-th wave prediction information, and input the first wave prediction information for the test to the k-th wave prediction information for the test into the marine environment information correction model to cause the marine environment information correction model to use the updated parameters Neural network calculation of the first wave prediction information for the test to the k-th wave prediction information for the test, and the first marine environment correction information for the test corresponding to each of the first point for the test to the k-th point for the test k It is possible to output marine environment correction information.

In this case, the first point for the test to the kth point for the test may be points corresponding to each grid in the grid of the specific area.

In addition, the first point for the test to the k-th point for the test may be points sampled from the grid of the specific region including the respective points at which observation information is obtained from the training data set.

In addition, the test wave prediction information 1010 may include at least one of significant wave height information, wave direction information, wave period information, wind speed information, water depth information, and maximum wave height information.

Among the factors of the test wave prediction information 1010, the depth information is directly input to the marine environment correction model 330, and the significant wave height information, the wave direction information, the wave period information, and the wind speed information are a wave prediction model can be calculated from

Also, when the wave prediction model is RWW3, the maximum wave height information may be calculated by applying the Rayleigh probability density function with reference to the significant wave height information and the wave period information.

The outputted test marine environment correction information 1020 may be used to update the test wave prediction information 1010 .

That is, when the error of the test wave prediction information 1010 calculated from the wave prediction model is corrected using the test marine environment correction information 1020, the accuracy of the test wave prediction information is improved and updated. It becomes possible to provide more accurate wave prediction information.

In addition, the outputted test marine environment correction information 1020 may be used to evaluate the accuracy of wave prediction information.

That is, by verifying the wave prediction information using the marine environment correction information 1020 for the test, the accuracy of the wave prediction information is confirmed, and it can be determined whether to provide the wave information according to the wave prediction information. In addition, it is possible to check a factor that degrades the performance of the wave prediction information to update the wave prediction model that provides the wave prediction information.

The artificial intelligence-based correction information calculation process according to the present invention may provide the following advantages.

Specifically, (i) it may be possible to provide reliable prediction information through correction information with improved accuracy than the conventional wave prediction model calculation data, and (ii) improve the accuracy of relatively low accuracy elements among the wave prediction model calculation data (iii) Since waves are generated and grown by wind, more accurate wave prediction can be expected if the accuracy of wind speed is improved through the correction process and then input as an external force of the wave prediction model.

That is, in the present invention, the process of calculating the marine environment correction information by inputting the wave prediction information into the marine environment correction model 330 may be applied to the atmospheric prediction model.

For example, the atmospheric prediction model is a model for predicting wind speed information in the atmosphere, and has been calculated using a numerical analysis method until now, and is calculated if the correction process using the AI model of the present invention is applied to the atmospheric prediction model. The accuracy of wind speed information can be improved.

In addition, when the calculated wind speed information is input to the wave prediction model, the accuracy of the calculated wave prediction information is improved, and more accurate wave prediction can be expected.

In addition, the embodiments according to the present invention described above may be implemented in the form of program instructions that can be executed through various computer components and recorded in a computer-readable recording medium. The computer-readable recording medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the computer-readable recording medium may be specially designed and configured for the present invention, or may be known and available to those skilled in the computer software field. Examples of the computer-readable recording medium include a hard disk, a magnetic medium such as a floppy disk and a magnetic tape, an optical recording medium such as a CD-ROM and DVD, and a magneto-optical medium such as a floppy disk. media), and hardware devices specially configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules for carrying out the processing according to the present invention, and vice versa.

In the above, the present invention has been described with specific matters such as specific components and limited embodiments and drawings, but these are provided to help a more general understanding of the present invention, and the present invention is not limited to the above embodiments. , various modifications and variations can be devised from these descriptions by those of ordinary skill in the art to which the present invention pertains.

Accordingly, the spirit of the present invention should not be limited to the above-described embodiments, and not only the claims described below, but also all modifications equivalently or equivalently to the claims described below belong to the scope of the spirit of the present invention. will do it

Claims (18)

A method of learning a multi-layer perceptron-based marine environment information correction model that corrects marine environment information using wave prediction information, the method comprising:
(a) the learning apparatus uses each of the first to mth atmospheric environment information of the specific area in each of the past points in time - the first past time point to the mth - where m is an integer greater than or equal to 2 - the blue wave of the specific area 1st wave prediction information to m-th wave prediction information that predicted With reference to the observation information, i - the i is an integer greater than or equal to 1 and less than or equal to m - j th of the past time - the j is an integer greater than or equal to 1 and less than or equal to n - i_j wave prediction information, which is wave prediction information at the point, and its correspondence obtaining, as a learning data set, i_j observation information, which is observation information to be used;
(b) the learning apparatus inputs the i_j wave prediction information into a marine environment information correction model, and causes the marine environment information correction model to perform a neural network operation on the i_j wave prediction information using a preset parameter to calculate the ith outputting the i_j th marine environment correction information that predicts the correction result of the marine environment correction information at the j th point in the past; and
(c) the learning device generates a loss with reference to the i_jth marine environment correction information and the i_jth observation information, and updates the parameter of the marine environment information correction model through backpropagation using the loss learning the marine environment information correction model;
How to include. According to claim 1,
The i_jth observation information and the i_jth marine environment correction information each include at least one of significant wave height information, maximum wave height information, wave period information, and wind speed information. According to claim 1,
The i_j observation information and the i_j sea environment correction information are significant wave height information, maximum wave height information, and wave period, respectively, when the i_j wave prediction information corresponds to a wave prediction model of RWW3 (Regional Wave Watch-3). information and wind speed information, and when the i_j wave prediction information corresponds to a wave prediction model of GWW3 (Global Wave Watch-3), a method of including significant wave height information and wind speed information. According to claim 1,
The i_j wave prediction information includes at least one of significant wave height information, wave direction information, wave period information, wind speed information, water depth information, and maximum wave height information. 5. The method of claim 4,
When the wave prediction model is RWW3, the maximum wave height information is calculated by applying a Rayleigh probability density function with reference to the significant wave height information and the wave period information. According to claim 1,
(d) the learning apparatus, (i) at each of the first past time point to the mth past time point, using each of the first atmospheric environment information to the mth atmospheric environment information of the specific area, The first wave prediction information to the mth wave prediction information for predicting a wave, and 1_1st observation information to m_nth observation information for observing the marine environment information at each of the first point to the nth point within the specific area With reference to, s-th - s is an integer greater than or equal to 1 and less than or equal to m - th of a past time - t is an integer greater than or equal to 1 and less than or equal to n - s_t wave prediction information and corresponding observations that are wave prediction information at a point Obtaining s_t observation information, which is information, as a verification data set, (ii) inputting the s_t wave prediction information into the marine environment information correction model to cause the marine environment information correction model to use the updated parameter to make the s_t Neural network operation on the wave prediction information to output the s_t sea environment correction information that predicts the correction result of the marine environment correction information at the t th point of the s th past time point, and (iii) the s_t wave prediction information and A first bias and a mean square error (RMSE) are calculated with reference to the s_t observation information, and a second deviation and a root mean square error are calculated with reference to the s_t sea environment correction information and the s_t observation information. calculating and verifying the performance of the marine environment information correction model with reference to the first deviation and root mean square error and the second deviation and root mean square error;
Method, characterized in that it further comprises. A method of testing a multi-layer perceptron-based marine environment information correction model that corrects marine environment information using wave prediction information, the method comprising:
(a) by the learning device, (i) the first past time point to the mth - wherein m is an integer greater than or equal to 2 - At each of the past time points, each of the first atmospheric environment information for learning to the mth atmospheric environment information for learning of a specific area for learning The first wave prediction information for learning to the m-th wave prediction information for learning that predicted the wave of the specific region for learning by using the With reference to the first observation information for learning observing the environmental information and the m_n-th observation information for learning, i - the i is an integer greater than or equal to 1 and less than or equal to m - j-th of the past time - the j is an integer greater than or equal to 1 and less than or equal to n - point a process of obtaining, as a training data set, i_j-th wave prediction information for learning, which is wave prediction information, and i_j-th observation information for learning, which is observation information corresponding thereto; (ii) input the i_j wave prediction information for learning into the marine environment information correction model to cause the marine environment information correction model to calculate the i_j wave prediction information for learning by using a preset parameter with a neural network of the i-th past time point a process of outputting the i_j-th marine environment correction information for learning that predicted the correction result of the marine environment correction information for learning at the j-th point, and (iii) refer to the i_j sea environment correction information for learning and the i_j observation information for learning to generate a loss, and in a state in which the process of learning the marine environment information correction model by updating the parameters of the marine environment information correction model through backpropagation using the loss is performed, the testing device is configured in the specific area obtaining test wave prediction information for predicting a wave in the specific area by using the test atmospheric environment information; and
(b) the testing device, with reference to the wave prediction information for the test, the first point for the test within the specific area to the k-th for the test - the k is an integer greater than or equal to 1 - the first wave prediction information for the test at the point Obtain the k-th wave prediction information for testing, and input the first wave prediction information for the test to the k-th wave prediction information for the test into the marine environment information correction model to cause the marine environment information correction model to update the parameters By using the neural network calculation of the first wave prediction information for the test to the k-th wave prediction information for the test, the first marine environment correction information or test for the test corresponding to each of the first point for the test to the k-th point for the test for outputting k-th marine environment correction information;
How to include. 8. The method of claim 7,
(c) the testing device updates the wave prediction information for the test by using the first sea environment correction information for the test to the k-th sea environment correction information for the test, or the first sea environment correction information for the test evaluating the accuracy of the wave prediction information for the test with reference to the k-th marine environment correction information for the test;
How to include more. 8. The method of claim 7,
The test wave prediction information includes at least one of significant wave height information, wave direction information, wave period information, wind speed information, water depth information, and maximum wave height information. A learning apparatus for learning a multi-layer perception-based marine environment information correction model that corrects marine environment information using wave prediction information,
at least one memory storing instructions; and
at least one processor configured to execute the instructions;
The processor, (I) a first past time point to mth - where m is an integer greater than or equal to 2 - At each past time point, the blue wave of the specific area using each of the first atmospheric environment information to the mth atmospheric environment information of the specific area 1st wave prediction information to m-th wave prediction information that predicted With reference to the observation information, i - the i is an integer greater than or equal to 1 and less than or equal to m - j th of the past time - the j is an integer greater than or equal to 1 and less than or equal to n - i_j wave prediction information, which is wave prediction information at the point, and its correspondence a process of acquiring i_j observation information, which is the observation information to be used, as a learning data set; (II) input the i_j wave prediction information into a marine environment information correction model to cause the marine environment information correction model to perform a neural network operation on the i_j wave prediction information using a preset parameter to calculate the j-th of the i-th past time point a process of outputting the i_jth marine environment correction information that predicts the correction result of the marine environment correction information at the point; and (III) generating a loss with reference to the i_jth marine environment correction information and the i_j observation information, and updating the parameters of the marine environment information correction model through backpropagation using the loss. A learning apparatus characterized by performing; a process of learning the correction model. 11. The method of claim 10,
The i_j observation information and the i_j sea environment correction information each include at least one of significant wave height information, maximum wave height information, wave period information, and wind speed information. 11. The method of claim 10,
The i_j observation information and the i_j sea environment correction information are significant wave height information, maximum wave height information, and wave period, respectively, when the i_j wave prediction information corresponds to a wave prediction model of RWW3 (Regional Wave Watch-3). Information and wind speed information, and when the i_j wave prediction information corresponds to a wave prediction model of GWW3 (Global Wave Watch-3), a learning apparatus including significant wave height information and wind speed information. 11. The method of claim 10,
The i_j wave prediction information is a learning apparatus including at least one of significant wave height information, wave direction information, wave period information, wind speed information, water depth information, and maximum wave height information. 14. The method of claim 13,
When the wave prediction model is RWW3, the maximum wave height information is calculated by applying the Rayleigh probability density function with reference to the significant wave height information and the wave period information. 11. The method of claim 10,
the processor, (IV) (i) in each of the first past time point to the mth past time point, using each of the first atmospheric environment information to the mth atmospheric environment information of the specific area, the blue wave of the specific area The first wave prediction information to the m-th wave prediction information that predicted For reference, s th - the s is an integer greater than or equal to 1 and less than or equal to m - th of the past time - t is an integer greater than or equal to 1 and less than or equal to n - s_t th wave prediction information at a point and corresponding observation information s_t observation information is obtained as a verification data set, and (ii) the s_t wave prediction information is input to the marine environment information correction model to cause the marine environment information correction model to use the updated parameter to make the s_t wave Neural network calculation of the prediction information to output the s_t-th marine environment correction information that predicts the correction result of the marine environment correction information at the t-th point of the s-th past time point, and (iii) the s_t wave prediction information and the A first bias and a mean square error (RMSE) are calculated with reference to the s_t observation information, and a second deviation and a root mean square error are calculated with reference to the s_t sea environment correction information and the s_t observation information. and a process of verifying the performance of the marine environment information correction model with reference to the first deviation and root mean square error and the second deviation and root mean square error. In the testing apparatus for testing a multi-layer perception-based marine environment information correction model that corrects marine environment information using wave prediction information,
at least one memory storing instructions; and
at least one processor configured to execute the instructions;
The processor, (I) by the learning device, (i) the first past time point to the m-th - The m is an integer greater than or equal to 2 - At each of the past time points, the first standby environment information for learning or the m-th standby for learning of the specific area for learning The first wave prediction information for learning to the m-th wave prediction information for learning that predicted the wave of the specific area for learning by using each of the environmental information, and the first point for learning and the n-th for learning within the specific area - The n is an integer of 2 or more - With reference to the first 1_1 observation information for learning to the m_n-th observation information for learning that observed the marine environment information at each point, i - the i is an integer greater than or equal to 1 and less than or equal to m - the jth of the past time - the j is greater than or equal to 1 and n an integer equal to or less than - a process of acquiring, as a learning data set, i_j wave prediction information for learning that is wave prediction information at a point and i_j observation information for learning that is observation information corresponding thereto; (ii) input the i_j wave prediction information for learning into the marine environment information correction model to cause the marine environment information correction model to calculate the i_j wave prediction information for learning by using a preset parameter with a neural network of the i-th past time point a process of outputting the i_j-th marine environment correction information for learning that predicted the correction result of the marine environment correction information for learning at the j-th point, and (iii) refer to the i_j sea environment correction information for learning and the i_j observation information for learning to generate a loss, and in a state in which the process of learning the marine environment information correction model by updating the parameters of the marine environment information correction model through backpropagation using the loss is performed, the test standby of the specific area a process of obtaining wave prediction information for a test in which a wave of the specific area is predicted using environment information; and (II) the first point for testing to the k-th for testing within the specific area with reference to the wave prediction information for the test - The k is an integer greater than or equal to 1 - The first wave prediction information for testing at the point to the k-th for testing Obtaining wave prediction information, inputting the first wave prediction information for the test to the k-th wave prediction information for the test into the marine environment information correction model causes the marine environment information correction model to perform the test using the updated parameters For the first wave prediction information for the test, the neural network calculates the k-th wave prediction information for the test, and the first sea environment correction information for the test or the k-th sea for the test corresponding to each of the first point for the test to the k-th point for the test A testing device that performs a process for outputting environmental correction information. 17. The method of claim 16,
The processor (III) updates the wave prediction information for the test by using the first sea environment correction information for the test to the k-th sea environment correction information for the test, or the first sea environment correction information for the test or the test a process of evaluating the accuracy of the wave prediction information for the test with reference to the k-th marine environment correction information for the test; A testing device to perform further. 17. The method of claim 16,
The test wave prediction information includes at least one of significant wave height information, wave direction information, wave period information, wind speed information, water depth information, and maximum wave height information.

Images