Learning apparatus, learning method, and prediction apparatus

Abstract

To provide a learning apparatus, a learning method, and a prediction apparatus capable of predicting a water level with higher accuracy than conventional techniques.SOLUTION: A learning apparatus includes: a learning data acquisition unit 12 that acquires, as learning data, a set including a water level of a river at a first time point, representative point information and rainfall amount information at at least one or more second time points earlier than the first time point, and water levels at at least two or more third time points earlier than the first time point; and a learning processing unit 13 that causes a learning machine to perform machine learning by using the learning data. The learning processing unit 13 causes the learning machine to perform learning so as to output a water level of the river at a sixth time point later than the fourth time points and the fifth time points by inputting representative point information and rainfall amount information at at least one or more fourth time points and water levels at two or more fifth time points corresponding to the third time points.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a learning device, a learning method, and a prediction device. [Background technology]

[0002] In river construction, from the perspective of safety management, it is effective to predict river water levels in real time and inform those involved in the construction. When considering only human evacuation, it is sufficient to know whether or not there will be flooding a few hours in advance to ensure safe evacuation. On the other hand, to ensure the safe evacuation of heavy machinery and materials within the construction area, it is necessary to know whether there will be flooding about 24 hours in advance.

[0003] There is a technology that uses machine learning to predict water levels, for example, 1 to 24 hours in the future, using the centroid coordinates of the distribution obtained from a rain cloud image, the distance to the centroid coordinates, the average precipitation amount, and the most recent water level (see Patent Document 1). [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2022-065522 A (Fig. 12) Summary of the Invention [Problem to be solved by the invention]

[0005] When considering evacuation for river construction work, prediction accuracy is extremely important, so improving prediction accuracy is a major issue.

[0006] From this perspective, the present invention provides a learning device, a learning method, and a prediction device that can predict water levels with higher accuracy than conventional devices. [Means for solving the problem]

[0007] The learning device according to the present invention includes a learning data acquisition unit that acquires learning data, and a learning processing unit that has a learning device and performs machine learning on the learning device using the learning data. The learning data acquisition unit acquires, as learning data, sets of the water level of a river at a first time point, representative point information and rainfall information at at least one or more second time points that are earlier than the first time point, and water levels at at least two or more third time points that are earlier than the first time point. The representative point information is information about a representative point that reflects a rainfall intensity distribution within an area that includes at least a part of a catchment area of ​​the river. The rainfall amount information is information about the rainfall amount within the area. The learning processing unit inputs representative point information and rainfall information at at least one or more fourth time points and water levels at two or more fifth time points corresponding to the third time points, and trains the learning device to output the water level of the river at a sixth time point that is in the future than the fourth and fifth time points.

[0008] In the learning device according to the present invention, by inputting the water levels at at least two or more third points in time that are prior to the first point in time, it is possible to learn the water level trend (whether the water level is rising, falling, or remaining flat). Therefore, by using a learning device trained using this learning data, it is possible to predict the water level with high accuracy.

[0009] The representative point information and the rainfall amount information at the second time point and the fourth time point may be two or more pieces of information that are consecutive in time. The representative point information may be the center of gravity of the range when the intensity of rainfall within the range is considered to be weight. The learning data may include the representative point information and the rainfall information at the first time point. In this case, the learning processing unit may input the representative point information and rainfall information at the fourth and sixth time points, and the water level at the fifth time point, to train the learning device to output the water level of the river at the sixth time point.

[0010] The learning data may include variance value information at the second time point, the variance value information being information relating to the variation of rainfall in the range. In this case, the learning processing unit trains the learning device to output the water level of the river at a sixth time point that is in the future than the fourth and fifth time points by inputting representative point information, variance value information, and rainfall information at at least one or more fourth time points, and water levels at two or more fifth time points that correspond to the third time points.

[0011] In this way, by inputting variance information, it is possible to learn about differences in rainfall distribution (for example, whether it is a uniform distribution or a torrential downpour). Therefore, by using a learning machine trained using this training data, it is possible to predict water levels with greater accuracy.

[0012] The learning method according to the present invention includes a learning data acquisition step of acquiring learning data, and a learning processing step of using the learning data to train a learner. In the learning data acquisition step, sets of the water level of a river at a first time point, representative point information and rainfall information at at least one or more second time points that are earlier than the first time point, and water levels at at least two or more third time points that are earlier than the first time point are acquired as learning data. The representative point information is information about a representative point that reflects a rainfall intensity distribution within an area that includes at least a part of a catchment area of ​​the river. The rainfall amount information is information about the rainfall amount within the area. In the learning processing step, the learning device is trained to output the water level of the river at a sixth time point that is in the future than the fourth and fifth time points by inputting representative point information and rainfall information at at least one fourth time point and water levels at two or more fifth time points that correspond to the third time point.

[0013] In the learning method according to the present invention, by inputting water levels at at least two or more third time points that are prior to the first time point, it is possible to learn the water level trend (whether the water level is rising, falling, or remaining flat). Therefore, by using a learning device trained using this learning data, it is possible to predict the water level with high accuracy.

[0014] The prediction device of the present invention is equipped with a trained learning device that has been trained using the learning method described above, and predicts the water level of the river at the sixth point in time by inputting representative point information and rainfall information at the fourth point in time, and water levels at two or more fifth points in time corresponding to the third point in time, into the trained learning device.

[0015] By using the prediction device according to the present invention, it is possible to predict water levels with high accuracy. [Effects of the Invention]

[0016] According to the present invention, the water level can be predicted more accurately than before. [Brief explanation of the drawings]

[0017] [Figure 1] FIG. 1 is a functional configuration diagram of a learning device and a prediction device according to an embodiment of the present invention. [Figure 2] This is an image of a river and its catchment area. [Figure 3] FIG. 10 is a diagram for explaining representative point information. [Figure 4] This is an illustration of the center of gravity in a watershed. (a) shows the center of gravity when rainfall is uniformly distributed within the watershed, (b) shows the center of gravity when heavy rain occurs within the watershed, and (c) shows the center of gravity when heavy rain occurs in two places within the watershed. [Figure 5] 1 is an example of a training data group according to an embodiment of the present invention. [Figure 6] 10 is an image of predicted data assumed in an embodiment of the present invention. [Figure 7]1 is an exemplary flowchart of a training method and a prediction method according to an embodiment of the present invention. [Figure 8] The results of water level predictions using the previously invented method and the method of the present invention are shown. [Figure 9] The results of water level forecasts for 6, 12, 18, and 24 hours ahead are compared between the existing method and the method of the present invention. [Figure 10] The correlation coefficient of the water level prediction results is compared between the previously invented method and the method of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0018] Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. Each drawing is merely a schematic illustration to allow a sufficient understanding of the present invention. Therefore, the present invention is not limited to the illustrated examples. In each drawing, common or similar components are designated by the same reference numerals, and redundant explanations thereof may be omitted.

[0019] <Configuration of the learning device and prediction device according to the embodiment> A learning device 10 and a prediction device 20 according to an embodiment will be described with reference to Fig. 1. Fig. 1 is a functional configuration diagram of the learning device 10 and the prediction device 20 according to an embodiment. Note that the learning device 10 and the prediction device 20 can also be configured as a single device.

[0020] The learning device 10 is a device that learns the relationship between the rainfall conditions in a river catchment area (also called a "watershed") and the water level of the river. The learning device 10 performs machine learning using information obtained by processing the rainfall intensity distribution F. The prediction device 20 is a device that predicts the water level of a river at a predetermined time based on the rainfall conditions in the river's catchment area. The prediction device 20 predicts the water level using information obtained by processing the rainfall intensity distribution F.

[0021] The rainfall intensity distribution F shows the intensity of rainfall in two dimensions. For example, the rainfall intensity distribution F is the "analyzed rainfall" which is the actual rainfall amount with a resolution of "1 km" distributed by the Japan Meteorological Agency. The rainfall intensity distribution F shows the rainfall amounts at multiple points in the catchment area of ​​the river for which learning (or prediction) is being performed. The rainfall amounts that make up the rainfall intensity distribution F may be actual measured values. It is desirable that the rainfall intensity distribution F shows the intensity distribution of the entire river catchment area, but it is not necessary to show the entire area (i.e., it may show only a part of the river catchment area). It is desirable that the rainfall intensity distribution F includes information indicating the point in time at which the intensity distribution is shown (for example, the date and time). In the following, the rainfall intensity distribution F used in the learning stage may be specifically referred to as "rainfall intensity distribution F1," and the rainfall intensity distribution F used in the forecasting stage may be specifically referred to as "rainfall intensity distribution F2."

[0022] (Learning device configuration) 1, the learning device 10 includes a data processing unit 11, a learning data acquisition unit 12, and a learning processing unit 13. The data processing unit 11, the learning data acquisition unit 12, and the learning processing unit 13 are realized by program execution processing by a CPU (Central Processing Unit), dedicated circuits, etc. When these functions are realized by program execution processing, programs for realizing the functions are stored in a storage unit (not shown).

[0023] A rainfall intensity distribution F1 is input to the data processing unit 11. There is no particular limit to the number of rainfall intensity distributions F1 input to the data processing unit 11, and for example, a time-series rainfall intensity distribution F1 for each hour is input. The data processing unit 11 obtains representative point information, variance value information, and rainfall amount information from the rainfall intensity distribution F1.

[0024] The representative point reflects the rainfall intensity distribution in the river catchment area, and is determined from the rainfall intensity (for example, rainfall amount) at multiple points included in the rainfall intensity distribution F. The representative point is, for example, the center of gravity of the catchment area when rainfall intensity (for example, rainfall amount) is considered to be weight. The representative point information is, for example, the position of the representative point and the distance from a predetermined reference point to the representative point.

[0025] The rainfall information is information about the amount of rainfall in the river catchment area, such as a statistical value (e.g., average value) for the entire catchment area. The rainfall information is obtained from the rainfall intensity (e.g., rainfall amount) at multiple points included in the rainfall intensity distribution F.

[0026] The variance information is information about the variation of rainfall in a river catchment area, for example, the variance of the entire catchment area. The variance information is calculated from the rainfall intensity (for example, the rainfall amount) at multiple points included in the rainfall intensity distribution F.

[0027] The processing of the data processing unit 11 will be described with reference to Figures 2 and 3. Figure 2 is an image diagram of a river and its catchment area. Figure 3 is a diagram for explaining representative point information.

[0028] In Figure 2, the river is represented by the symbol K1, and the catchment area is represented by the symbol K2. The catchment area K2 indicates the area where rain that falls flows into the river K1, and the rain that falls in the catchment area K2 flows out into the sea K3 via the river K1. The catchment area K2 may be based on a river basin map published by the Ministry of Land, Infrastructure, Transport and Tourism, for example. In addition, an observation station K4 is installed downstream of the river K1, which can measure the water level of the river K1. The time it takes for rain that falls upstream of the river K1 to affect the water level change at the observation station K4 is determined by the characteristics of the river K1, such as the length and elevation difference of the river.

[0029] The data processing unit 11 calculates the center of gravity G of the watershed K2 as a representative point when the rainfall amounts at multiple points (e.g., points every 1 km) set in the rainfall intensity distribution F1 are likened to weights. The calculation of the center of gravity of the watershed K2 can be found (see FIG. 3) from the distance (e.g., coordinate values) from a reference point (the origin in FIG. 3) set in the rainfall intensity distribution F1 to each set point and the rainfall amount at each point, using, for example, equation (1). The origin here is the water level prediction point, and the x-axis and y-axis are set on the sides that make up the rainfall intensity distribution F1.

[0030] ·X coordinate of center of gravity G = (x1m1+ x2m2+ … + x n m n ) / (m1+ m2+ … + m n ) ···Formula (1) Here, "x1, x2, … , x n " is the distance in the x-axis direction (x coordinate value) from the origin to each point to be set, and "m1, m2, ... , m n " is the amount of rainfall at each point. The y coordinate value of the center of gravity G is calculated in the same way, and the center of gravity coordinate (X, Y) of the center of gravity G is found.

[0031] Referring to Figure 4, an image of the center of gravity in a watershed is shown. Figure 4 is an image diagram of the center of gravity in a watershed. (a) shows the center of gravity when rainfall conditions within the watershed are uniformly distributed, (b) shows the center of gravity when heavy rain occurs within the watershed, and (c) shows the center of gravity when heavy rain occurs in two locations within the watershed. It is also possible that the rainfall distributions are different but the center of gravity is the same. In this embodiment, differences in rainfall distribution are taken into account by using the variance of rainfall in the watershed, as will be described later.

[0032] The data processing unit 11 then sets the position of the center of gravity G (center of gravity coordinates (X, Y)) and the distance L from the reference point (here, the origin) to the center of gravity G as representative value information. The data processing unit 11 also calculates statistical values ​​of the rainfall amounts at each point set in the rainfall intensity distribution F1, and sets the calculated statistical values ​​as rainfall amount information. The statistical value may be, for example, the average value of the rainfall amounts at each point set in the rainfall intensity distribution F1 (average rainfall R). The data processing unit 11 also calculates the variance value Var of the rainfall in the watershed using the rainfall amounts at each point set in the rainfall intensity distribution F1 and the average rainfall R. Specifically, it calculates the square of the difference between the rainfall amount at each point and the average rainfall R, and then calculates the average of the results.

[0033] The learning data acquisition unit 12 shown in FIG. 1 combines the following information and organizes it as learning data. Here, the second and third time points are points in time earlier than the first time point (the second time point may partially overlap with the first time point). The third time point is preferably a time included in the time period affected by water level changes due to rainfall. The time until the water level changes have an effect is determined by the characteristics of the river, such as the length and elevation difference of the river. (a) River water level measured at the observation station (at the first point in time) (a) The position of the center of gravity G (center of gravity coordinates (X, Y)) obtained by the data processing unit 11, the distance L from the origin to the center of gravity G, the average rainfall R in the catchment area, and the variance value Var of the rainfall (at the second time point) (c) Past river water levels measured at the observation station (at the third point in time) (a) The water level of the river measured at the observation station and (c) the past water level of the river measured at the observation station may be, for example, that published by the Ministry of Land, Infrastructure, Transport and Tourism.

[0034] The learning data acquisition unit 12 organizes the learning data according to instructions from, for example, a person who predicts water levels (hereinafter referred to as a "user"). Note that the river water level, the coordinate position of the center of gravity G, the distance L to the center of gravity G, the average rainfall R, the variance value Var, and past river water levels may be registered in advance in a storage unit in a combined form, and the learning data acquisition unit 12 may acquire the registered information as learning data. The storage unit may be included in a device connected via a communication line (for example, a cloud system). Hereinafter, a collection of organized learning data may be referred to as a "learning data group."

[0035] An example of the training data group organized by the training data acquisition unit 12 is shown in Figure 5. The training data group shown in Figure 5 is composed of multiple training data sets with different water level measurement times. The training data group shown in Figure 5 includes information from 2007 to 2022, and the information for each year corresponds to water levels measured every hour from June to November. Each training data set consists of the "water level measurement time," the "centroid coordinates X, Y," the "distance to the centroid L," the "average rainfall within the watershed R," and the "rainfall variance Var" for each hour from 24 hours before to 24 hours after the water level measurement time, and the hourly water level from 6 hours before to 24 hours after the water level measurement time. Note that the "centroid coordinates X, Y," the "distance to the centroid L," the "average rainfall within the watershed R," and the "rainfall variance Var" for times in the future and for the water level measurement time do not necessarily need to be included in the training data. Including this information has the advantage of improving prediction accuracy. In other words, because future rainfall will change the rise and fall of water levels, past information alone may not improve the accuracy of water level predictions in the future, especially at times far from the present. In such cases, it is better to use the "centroid coordinates X, Y," "distance to the centroid L," "average rainfall within the catchment area R," and "variance of rainfall Var" for the future and water level measurement times rather than the water level measurement time.

[0036] In this way, the learning data acquisition unit 12 acquires as learning data a set of the water level of the river at a certain first point in time (a time from 1 hour to 24 hours after the water level measurement time in Figure 5), representative point information, rainfall information, and variance value information at at least one or more second points in time in the past than the first point in time (which may include from 24 hours before the water level measurement time in Figure 5 to the time of water level measurement up to 24 hours later), and water levels at at least two or more third points in time in the past than the first point in time (from 6 hours before the water level measurement time in Figure 5 to the time of water level measurement).

[0037] The learning processing unit 13 shown in FIG. 1 has a learning device, and performs machine learning on the learning device using the learning data acquired by the learning data acquisition unit 12. The learning device is a learning system in machine learning, and can derive good results by comparing the results of classification, prediction, judgment, etc. based on given data with the actual results that are correct answers, and adjusting various parameters. The learning device is, for example, a neural network, and the present embodiment will be described assuming a neural network. The learning device is also called a "learning model," etc.

[0038] As is well known, neural networks are an information processing method created with the aim of having computers imitate the workings of the human brain. Neural networks excel at complex nonlinear processing, and by using learning functions, they are characterized by the fact that there is no need to formulate the relationship between explanatory variables and target variables. Neural networks generally consist of an input layer, a hidden layer, and an output layer. Explanatory variables are input to the input layer, and target variables are output from the output layer. The hidden layer plays a role in relating the two. There are no restrictions on the number of hidden layers or the number of neurons in each hidden layer, and they can be set as desired.

[0039] The method for training the learning device is not particularly limited. For example, the learning device may be trained using a method appropriate for the type of learning device. Specifically, the learning data, which comprise the "centroid coordinates X, Y," "distance to the centroid L," "average rainfall in the watershed R," and "rainfall variance Var" for each hour from 24 hours before to 24 hours after the water level measurement time, as well as the hourly water level from 6 hours before to the water level measurement time, are input to the input layer. The middle layer is adjusted based on the error between the results output from the output layer and the hourly water level from 1 hour to 24 hours after the water level measurement time. In other words, the learning processing unit 13 inputs representative point information, variance information, and rainfall information for at least one fourth time point, and water levels for two or more fifth time points corresponding to the third time point, and trains the learning device to output the river water level at a sixth time point that is later than the fourth and fifth time points. The number of learning data to be trained is not particularly limited. For example, the learning process ends when a desired accuracy is reached.

[0040] (Configuration of prediction device) 1, the prediction device 20 includes a data processing unit 21, a prediction data acquisition unit 22, and a prediction processing unit 23. The data processing unit 21, the prediction data acquisition unit 22, and the prediction processing unit 23 are realized by program execution processing by a CPU (Central Processing Unit), a dedicated circuit, etc. When these functions are realized by program execution processing, a program for realizing the function is stored in a storage unit (not shown).

[0041] A rainfall intensity distribution F2 is input to the data processing unit 21. There is no particular limit to the number of rainfall intensity distributions F2 input to the data processing unit 21; for example, a time-series rainfall intensity distribution F2 for each hour may be input. The data processing unit 21 obtains representative point information, variance value information, and rainfall amount information from the rainfall intensity distribution F2. The processing of the data processing unit 21 is similar to the processing of the data processing unit 11 in the learning stage, so a detailed description will be omitted. The rainfall intensity distribution F2 may be from the past (e.g., measured) or from the future (e.g., forecast). In other words, it is possible to predict the water level several hours later (e.g., the present or near future) from the actual past rainfall intensity distribution F2, or to predict the water level several hours later from the future rainfall intensity distribution F2 published as a forecast.

[0042] The prediction data acquisition unit 22 shown in Fig. 1 combines the following information and organizes it as learning data. Here, the fourth and fifth time points are time points that are earlier than the sixth time point to be predicted (the fourth time point may partially overlap with the sixth time point). The fifth time point is preferably a time included in the time period affected by water level changes due to rainfall. (F) The position of the center of gravity G (center of gravity coordinates (X, Y)) obtained by the data processing unit 21, the distance L from the origin to the center of gravity G, the average rainfall R in the catchment area, and the variance value Var of the rainfall (at the fourth point in time) (G) Past river water levels measured at observation stations (as of the fifth point in time) (G) The past water levels of rivers measured at observation stations may be those published by the Ministry of Land, Infrastructure, Transport and Tourism, for example. When making predictions based on future times, the information in (F) and (G) may be forecast values.

[0043] The coordinate position of the center of gravity G, the distance L to the center of gravity G, the average rainfall R, the variance Var, and past river water levels may be registered in advance in a combination in the storage unit, and the predicted data acquisition unit 22 may acquire the registered information as predicted data. The storage unit may be included in a device connected via a communication line (for example, a cloud system). It is desirable that the predicted data be calculated using a rainfall intensity distribution F2 in the same format as the rainfall intensity distribution F1.

[0044] In this way, the prediction data acquisition unit 22 acquires as prediction data a set of representative point information, rainfall information, and variance value information at at least one or more fourth time points (which in this embodiment may include the period from 24 hours before the water level prediction time up to 24 hours after the water level prediction time) and water levels at at least two or more fifth time points in the past than the water level prediction time (which in this embodiment may include the period from 7 hours before the water level prediction time to 1 hour before the water level prediction time). An image of the prediction data in this embodiment is shown in Figure 6.

[0045] The prediction processing unit 23 shown in Fig. 1 has a trained learning device. The trained learning device is a learning device that has been machine-learned by the learning device 10. In other words, the trained learning device has been trained to output the water level of the river at a sixth time point that is in the future than the fourth time point and the fifth time point by inputting representative point information, variance information, and rainfall information for at least one or more fourth time points and water levels for two or more fifth time points that correspond to the third time point. The prediction processing unit 23 inputs the prediction data acquired by the prediction data acquisition unit 22 into the trained learning device, thereby predicting the water level of the river at a sixth time point that is in the future than the fourth time point and the fifth time point included in the prediction data, and outputs the prediction result.

[0046] <Learning method and prediction method according to the embodiment> A learning method and a prediction method according to an embodiment will be described with reference to Fig. 7 (and also with reference to Figs. 1 to 6 as appropriate). Fig. 7 is a flowchart illustrating the learning method and the prediction method according to an embodiment.

[0047] 7, the process of constructing a prediction model (step S10) corresponding to the learning method according to the embodiment mainly includes a process of preparing input data (past) (step S11), a parameter tuning process (step S12), and a process of determining prediction accuracy (step S13). Also, the process of predicting water level (step S20) corresponding to the prediction method according to the embodiment mainly includes a process of preparing input data (real-time) (step S21), a calculation process using a prediction method (step S22), a confirmation process of prediction results (step S23), and a provision process of the results to construction personnel (step S24).

[0048] (Preparation process of input data (past) (step S11)) The user prepares the rainfall intensity distribution F1 and measured river water level information to be used for training the prediction model (learner). The user prepares, for example, the hourly rainfall intensity distribution F1 and river water level information. The river water level is the water level at a location corresponding to the point to be predicted, and may be published by, for example, the Ministry of Land, Infrastructure, Transport and Tourism. The rainfall intensity distribution F1 may be published by, for example, the Japan Meteorological Agency, and since the information published by the Japan Meteorological Agency covers the entire country of Japan, the rainfall intensity distribution F1 can be prepared without much effort or time.

[0049] The data processing unit 11 of the learning device 10 calculates representative point information, rainfall amount information, and variance value information from the rainfall intensity distribution F1 using the above-mentioned method. The representative point information is, for example, the position of the center of gravity G of the watershed K2 (center of gravity coordinates (X, Y)) and the distance L from the reference point to the center of gravity G. The rainfall amount information is, for example, the average value of rainfall in the watershed K2 (average rainfall R). The variance value information is, for example, the variance value Var of rainfall in the watershed R2. Specifically, the square of the difference between the rainfall amount at each point set in the rainfall intensity distribution F1 and the average rainfall R is calculated, and the average of the results is calculated.

[0050] The learning data acquisition unit 12 combines the water level of the river measured at the observation station with the position of the center of gravity G (center of gravity coordinates (X, Y)) calculated by the data processing unit 11, the distance L from the origin to the center of gravity G, the average rainfall R in the catchment area, and the variance value Var of the rainfall, and organizes the combined data as learning data. The organized learning data is stored in a storage unit (not shown).

[0051] (Parameter tuning process (step S12)) The learning data acquisition unit 12 of the learning device 10 sequentially acquires learning data, and the learning processing unit 13 trains a learning device by machine learning using the acquired learning data. If the learning device is a neural network, the intermediate layer is adjusted using the learning data. The learning processing unit 13 trains the learning device by machine learning to output the water level of the river at a sixth time point that is in the future than the fourth time point and the fifth time point by inputting representative point information, variance information, and rainfall information for at least one or more fourth time points and water levels for two or more fifth time points that correspond to the third time point.

[0052] (Prediction accuracy determination process (step S13)) In this step, the predictive accuracy of the learned learning device is determined, and if the predictive accuracy meets the standard set by the user, the learning step is terminated (if the predictive accuracy is "good"). On the other hand, if the predictive accuracy of the learning device does not meet the set standard (if the predictive accuracy is "poor"), the parameter tuning step (step S12) is continued, and the learning step continues. Note that the parameter tuning step (step S12) and the predictive accuracy determination step (step S13) may be realized by cross-validation. When learning is performed using cross-validation, for example, data from the organized data for the year to be predicted is used as test data, and the rest is classified as learning data to predict the water level for each year.

[0053] (Input data (real-time) preparation process (step S21)) The user prepares the rainfall intensity distribution F2 used to predict the river water level and information on past measured river water levels. The user prepares, for example, hourly rainfall intensity distribution F2 and river water level information. The river water level is the water level at a location corresponding to the prediction point and may be published by, for example, the Ministry of Land, Infrastructure, Transport and Tourism. The rainfall intensity distribution F2 may be published by, for example, the Japan Meteorological Agency. The data processing unit 21 of the prediction device 20 calculates representative point information, rainfall amount information, and variance value information from the rainfall intensity distribution F2 in the same manner as in the preparation process of step S11. The prediction data acquisition unit 22 then combines the information obtained by the data processing unit 21 with past river water level information to organize it as prediction data. Note that when performing real-time predictions from the present onward, the rainfall intensity distribution F2 at future points in time may be forecast values.

[0054] (Calculation process using prediction method (step S22)) The prediction data acquisition unit 22 of the prediction device 20 acquires the prediction data, and the prediction processing unit 23 inputs the prediction data into a trained learning device to predict the water level of the river.

[0055] (Step S23 for confirming the prediction results and step S24 for providing the results to the construction personnel) The user checks the water level prediction result calculated in step S22 and provides the information on the predicted water level to people involved in the construction work, if necessary. Based on the provided predicted water level, the people involved in the construction work can evacuate heavy machinery and materials in the construction yard or take measures such as protecting them.

[0056] <Effects of the learning device and prediction device according to the embodiment> As described above, the learning device 10 and the prediction device 20 according to the embodiment can learn the water level trend (whether the water level is rising, falling, or remaining flat) by inputting the water levels at at least two or more third time points that are earlier than the first time point. Therefore, by using a learning device trained using the learning data, the water level can be predicted with high accuracy.

[0057] Furthermore, by inputting variance value information, the learning device 10 and the prediction device 20 according to the embodiment can learn about differences in rainfall distribution (for example, whether it is a uniform distribution or a torrential downpour). Therefore, by using a learning device trained using the learning data, it is possible to predict water levels with higher accuracy.

[0058] Water levels were predicted using the previously invented method (JP Patent Publication No. 2022-065522) and the method of the present invention at the Tsukigata Observatory on the Ishikari River as an example, and the results were compared and will be explained below. Figure 8 shows the results of water level predictions using the previously invented method and the method of the present invention. The comparison is for floods where the measured water level exceeds 8m. Looking at the prediction results for July 2009, no significant differences can be seen. However, for September 2011 and August 2014, it can be confirmed that the prediction results using the method of the present invention were closer to the measured water level.

[0059] Figure 9 compares the water level forecast results for 6, 12, 18, and 24 hours ahead between the existing invented method and the method of the present invention. For each of the invented method and the method of the present invention, a correlation diagram was created with the measured water level on the horizontal axis and the predicted water level on the vertical axis, and the correlation coefficient was calculated. Figure 10 compares the correlation coefficient of the water level forecast results between the existing invented method and the method of the present invention. Figure 10 shows that the forecast accuracy of the method of the present invention is higher than that of the existing invented method.

[0060] Although the embodiment of the present invention has been described above, the present invention is not limited to this and can be practiced within the scope of the claims.

[0061] In the embodiment, a representative point for the entire watershed is calculated, but a representative point for a part of the watershed may be calculated. In other words, the representative point may reflect the rainfall intensity distribution within an area including at least a part of the watershed of the river, and the representative point information is information reflecting the rainfall intensity distribution within an area including at least a part of the watershed of the river. The same applies to the rainfall amount information and the variance value information.

[0062] Alternatively, the watershed may be divided and a representative point for each divided area may be determined. The method for dividing the watershed is not particularly limited, and for example, the watershed may be divided equally, or if a river branches, the area may be divided for each branching river. In this case, the training data may include representative point information for each divided area, or information determined from each representative point (for example, information obtained by adding or averaging each representative point) may be included in the training data.

[0063] In the embodiment, the representative point information, variance value information, and rainfall amount information were obtained from the rainfall intensity distribution F. The rainfall intensity distribution F is, for example, the "analyzed rainfall" actual rainfall with a resolution of "1 km" distributed by the Japan Meteorological Agency. However, it is also possible to obtain the representative point information, variance value information, and rainfall amount information from a raincloud image. The raincloud image is an RGB image that shows the rainfall intensity distribution with color changes, and may be one published by the Japan Meteorological Agency, for example. A method for calculating the representative point information and rainfall amount information from a raincloud image is described, for example, in "JP 2022-065522 A." Specifically, the RGB values ​​of each pixel in the raincloud image are converted to brightness values ​​(gray values), and the brightness values ​​are used to calculate the representative point information, variance value information, and rainfall amount information. [Explanation of symbols]

[0064] 10 Learning Device 11 Data processing section 12 Learning data acquisition unit 13 Learning processing unit 20 Prediction Device 21 Data processing section 22 Prediction data acquisition section 23 Prediction processing section F,F1,F2 intensity distribution

Claims

1. a learning data acquisition unit that acquires, as learning data, a set of a water level of a river at a certain first time point, representative point information and rainfall information at at least one second time point that is older than the first time point, and a water level at at least two third time points that are older than the first time point; a learning processing unit having a learning device and causing the learning device to perform machine learning using the learning data; the representative point information is information about a representative point that reflects a rainfall intensity distribution within an area that includes at least a part of a catchment area of ​​the river; the rainfall information is information about the rainfall amount in the range; the learning processing unit inputs representative point information and rainfall information at at least one or more fourth time points and water levels at two or more fifth time points corresponding to the third time point, thereby causing the learning device to learn so as to output the water level of the river at a sixth time point that is in the future than the fourth time point and the fifth time point; A learning device characterized by:

2. The representative point information and the rainfall amount information at the second time point and the fourth time point are two or more pieces of time-sequential information.

2. The learning device according to claim 1 .

3. The representative point information is the center of gravity of the range when the intensity of rainfall within the range is considered to be weight.

2. The learning device according to claim 1 .

4. the learning data includes the representative point information and the rainfall amount information at the first time point, the learning processing unit inputs representative point information and rainfall information at the fourth time point and the sixth time point, and the water level at the fifth time point, and thereby causes the learning device to learn so as to output the water level of the river at the sixth time point.

2. The learning device according to claim 1 .

5. the learning data includes variance value information at the second time point; the variance value information is information about the variance of rainfall in the range, the learning processing unit inputs representative point information, variance information, and rainfall information at at least one or more fourth time points, and water levels at two or more fifth time points corresponding to the third time point, and causes the learning device to learn so as to output the water level of the river at a sixth time point that is in the future than the fourth time point and the fifth time point; 2. The learning device according to claim 1 .

6. a learning data acquisition step of acquiring, as learning data, a set of the water level of the river at a certain first time point, representative point information and rainfall information at at least one second time point that is older than the first time point, and water levels at at least two third time points that are older than the first time point; a learning process step of performing machine learning on a learning device using the learning data, the representative point information is information about a representative point that reflects a rainfall intensity distribution within an area that includes at least a part of a catchment area of ​​the river; the rainfall information is information about the rainfall amount in the range; In the learning process, by inputting representative point information and rainfall information at at least one or more fourth time points and water levels at two or more fifth time points corresponding to the third time point, the learning device is trained to output the water level of the river at a sixth time point that is in the future than the fourth time point and the fifth time point. A learning method characterized by:

7. a trained learning device that has been trained by the learning method according to claim 6, and that predicts the water level of the river at the sixth time point by inputting representative point information and rainfall information at the fourth time point and water levels at two or more fifth time points corresponding to the third time point into the trained learning device; A prediction device characterized by:

Images