River water level prediction system, river water level prediction method, and river water level prediction program

Abstract

Predict water levels with high accuracy without collecting precipitation or water level data from multiple locations. [Solution] One aspect of the present invention is a river water level prediction system comprising an acquisition unit that acquires water level data indicating the water level of a target river and precipitation data indicating the amount of precipitation in the target river; a calculation unit that calculates cumulative precipitation data and precipitation duration for a specified time based on the precipitation data acquired by the acquisition unit; and a prediction unit that predicts future water levels based on the water level data, precipitation data, cumulative precipitation data, and precipitation duration data, wherein the prediction unit generates water level prediction results for each divided period in the specified time by dividing the specified time calculated by the calculation unit into multiple periods and predicting the water level for each divided period based on the water level prediction results for the divided period prior to each divided period.

Description

[Technical Field]

[0001] The present invention relates to a river water level prediction system, a river water level prediction method, and a river water level prediction program. [Background technology]

[0002] A known conventional technology for determining the water level of a river is, for example, the water level determination system described in Patent Document 1. This water level determination system detects areas where water is reflected in a plurality of water level determination areas set at different positions in an image obtained by capturing an image of a location for determining the water level, determines whether or not each of the plurality of water level determination areas will be used to determine the water level based on the detection results, and determines the water level using the detection results of the water level determination areas that have been determined to be used to determine the water level. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2023-27633 Summary of the Invention [Problem to be solved by the invention]

[0004] The water level determination system described above is a technology for determining current water levels, but it is also desirable to be able to predict future river water levels. One way to improve the accuracy of river water level predictions is to collect precipitation and water level data from multiple points along the river. However, in order to improve the accuracy of water level predictions, it is necessary to collect precipitation and water level data from many points, which increases the cost of introducing the system.

[0005] The present invention has been made in consideration of the above-mentioned problems, and aims to provide a river water level prediction system, a river water level prediction method, and a river water level prediction program that can predict water levels with high accuracy without obtaining precipitation data or water level data from multiple points. [Means for solving the problem]

[0006] (1) One aspect of the present invention is to The current time of the area including the prediction target point Water level data showing the water level of the target river Area including the prediction target point in For each divided period, a specified period from the present to the future is divided into multiple periods. an acquisition unit that acquires precipitation data indicating precipitation; Calculating cumulative precipitation data for each divided period by adding up the precipitation amounts of the precipitation data for the predetermined period, and calculating precipitation duration data indicating the duration of precipitation for each divided period by adding up the time during which the precipitation amount is not zero for the predetermined period. a calculation unit and a prediction unit that predicts a future water level based on the water level data, the precipitation amount data, the accumulated precipitation amount data, and the precipitation duration data, and the prediction unit repeating a process of generating water level forecast data for a second divided period immediately following the first divided period from the current time based on the water level forecast data for the first divided period and the precipitation data for the first divided period, and the precipitation cumulative data and precipitation duration data calculated from the water level forecast data for the first divided period and the precipitation data for the first divided period, until the predetermined period is reached, and outputting water level forecast data for each divided period from the current time to the predetermined period; This is a river water level prediction system.

[0007] (2) In one aspect of the present invention, the prediction unit may input the water level data, precipitation data, cumulative precipitation data, and precipitation duration data of the target river into an individual river water level prediction model trained using the water level data, precipitation data, cumulative precipitation data, and precipitation duration data of the target river, and predict future water levels based on the output of the individual river water level prediction model; input the water level data, precipitation data, cumulative precipitation data, and precipitation duration data of the target river into a general-purpose river water level prediction model trained using the water level data, precipitation data, cumulative precipitation data, and precipitation duration data of a plurality of rivers, and predict future water levels based on the output of the general-purpose river water level prediction model; and predict future water levels in the target river based on the output of the individual river water level prediction model and the output of the general-purpose river water level prediction model.

[0008] (3) In one aspect of the present invention, the prediction unit may input the water level data, precipitation data, cumulative precipitation data, and precipitation duration data of the target river into an individual river water level prediction model trained using the water level data, precipitation data, cumulative precipitation data, and precipitation duration data of the target river, and predict future water levels based on the output of the individual river water level prediction model; input the water level data, precipitation data, cumulative precipitation data, and precipitation duration data of the target river into a general-purpose river water level prediction model trained using the water level data, precipitation data, cumulative precipitation data, and precipitation duration data of a plurality of rivers, and predict future water levels based on the output of the general-purpose river water level prediction model; and select whether to use the individual river water level prediction model or the general-purpose river water level prediction model based on the accuracy of the prediction results according to the output of the individual river water level prediction model and the accuracy of the prediction results according to the output of the general-purpose river water level prediction model.

[0009] (4) In one aspect of the present invention, the prediction unit recursively predicts the water level for each divided period based on the water level prediction result for the divided period before that divided period. specified period The water level data, the precipitation data, the accumulated precipitation data, and the precipitation duration data of the target river are input to a recursive prediction model that generates a water level prediction result for each divided period in the target river, and the water level for each divided period in the future is predicted based on the output of the recursive prediction model. The water level data, the precipitation data, the accumulated precipitation data, and the precipitation duration data at the current time are input, and the water level is predicted from the current time to the specified period The water level data, the precipitation data, the cumulative precipitation data, and the precipitation duration data of the target river may be input into a multiple time point prediction model that predicts water levels at multiple future time points included in the multiple time point prediction model, and water levels at multiple future time points may be predicted based on the output of the multiple time point prediction model, and future water levels in the target river may be predicted based on the output of the recursive prediction model and the output of the time point prediction model.

[0010] (5) One aspect of the present invention may include an evaluation unit that evaluates the difference between the water level predicted by the prediction unit and the actual water level, the difference between the time of arrival of the predicted maximum water level and the time of arrival of the actual maximum water level, and the difference between the predicted maximum water level and the actual maximum water level.

[0011] (6) One aspect of the present invention includes a normalization unit that normalizes the water level data, the precipitation data, the accumulated precipitation data, and the precipitation duration data, respectively, and the prediction unit may predict the water level based on the water level data, the precipitation data, the accumulated precipitation data, and the precipitation duration data normalized by the normalization unit.

[0012] (7) In one aspect of the present invention, an information processing device is The current time of the area including the prediction target point Water level data showing the water level of the target river Area including the prediction target point in For each divided period, a specified period from the present to the future is divided into multiple periods. obtaining precipitation data indicative of an amount of precipitation; Calculating cumulative precipitation data for each divided period by adding up the precipitation amounts of the precipitation data for the predetermined period, and calculating precipitation duration data indicating the duration of precipitation for each divided period by adding up the time during which the precipitation amount is not zero for the predetermined period. and a step of the information processing device predicting a future water level based on the water level data, the precipitation amount data, the accumulated precipitation amount data, and the precipitation duration data, wherein the step of predicting a future water level includes: For each divided period, a process of generating water level forecast data for the second divided period immediately following the first divided period from the current time based on the water level forecast data for the first divided period and the precipitation data for the first divided period, and the precipitation cumulative data and precipitation duration data calculated from the water level forecast data for the first divided period and the precipitation data for the first divided period is repeated until the predetermined period is reached, and water level forecast data for each divided period from the current time to the predetermined period is output. ,River water level prediction method.

[0013] (8) In one aspect of the present invention, the computer of the information processing device is The current time of the area including the prediction target point Water level data showing the water level of the target river Area including the prediction target point in For each divided period, a specified period from the present to the future is divided into multiple periods. obtaining precipitation data indicative of an amount of precipitation; Calculating cumulative precipitation data for each divided period by adding up the precipitation amounts of the precipitation data for the predetermined period, and calculating precipitation duration data indicating the duration of precipitation for each divided period by adding up the time during which the precipitation amount is not zero for the predetermined period. and a step in which the information processing device predicts a future water level based on the water level data, the precipitation amount data, the accumulated precipitation amount data, and the precipitation duration data, wherein the step of predicting a future water level includes: For each divided period, a process of generating water level forecast data for the second divided period immediately following the first divided period from the current time based on the water level forecast data for the first divided period and the precipitation data for the first divided period, and the precipitation cumulative data and precipitation duration data calculated from the water level forecast data for the first divided period and the precipitation data for the first divided period is repeated until the predetermined period is reached, and water level forecast data for each divided period from the current time to the predetermined period is output. , a river water level prediction program. [Effects of the Invention]

[0014] According to one aspect of the present invention, water levels can be predicted with high accuracy without acquiring precipitation data or water level data from multiple points. [Brief explanation of the drawings]

[0015] [Figure 1] 1 is a block diagram showing an example of the configuration of a river water level prediction system 1 according to an embodiment. [Figure 2] FIG. 2 is a diagram for explaining a river water level prediction model 142 in the embodiment. [Figure 3] FIG. 10 is a diagram illustrating an example of normalization processing according to the embodiment. [Figure 4] FIG. 2 is a diagram illustrating an example of processing performed by a prediction unit 140 according to the embodiment. [Figure 5] FIG. 10 is a diagram showing an example of a water level prediction result for a predetermined period in the embodiment. [Figure 6] 10 is a flowchart illustrating an example of data accumulation and expansion processing according to an embodiment. [Figure 7] 10 is a flowchart showing an example of a generation process of a river water level prediction model 142 in the embodiment. [Figure 8] 10A and 10B are diagrams illustrating a process of combining predicted target data with advanced training data according to an embodiment. [Figure 9] 10 is a flowchart illustrating an example of a process for predicting water levels in a time series according to an embodiment. [Figure 10] FIG. 1 is a diagram illustrating ensemble prediction of a river water level prediction model in an embodiment. [Figure 11] FIG. 10 is a diagram for explaining the process of automatically selecting a river water level prediction model in the embodiment. [Figure 12] 10 is a flowchart showing an example of a process for automatically selecting a river water level prediction model in an embodiment. [Figure 13] FIG. 10 is a diagram showing another example of selecting a river water level prediction model in the embodiment. [Figure 14]FIG. 10 is a diagram for explaining the evaluation of the river water level prediction results in the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0016] A river water level prediction system and a river water level prediction method to which the present invention is applied will be described below with reference to the drawings. The river water level prediction system, river water level prediction method, and river water level prediction program to which the present invention is applied acquire the water level of a point (hereinafter referred to as the prediction target point) in a river where the future water level is to be predicted, and predict the future water level of the prediction target point based on the acquisition of the current water level of the prediction target point.

[0017] FIG. 1 is a block diagram showing an example of the configuration of a river water level prediction system 1 according to an embodiment. The river water level prediction system 1 includes, for example, a water level prediction device 100, a water level sensor 200, a precipitation forecast information transmission device 300, and a terminal device 400. The water level prediction device 100, the water level sensor 200, the precipitation forecast information transmission device 300, and the terminal device 400 each include a communication interface (not shown in FIG. 1 ), such as a network interface card (NIC) or a wireless communication module. The water level prediction device 100, the water level sensor 200, and the precipitation forecast information transmission device 300 are connected via a communication network. The communication network includes, for example, the Internet, a wide area network (WAN), a local area network (LAN), a cellular network, etc.

[0018] The water level prediction device 100 is an information processing device such as a server device that acquires various types of information and executes various types of information processing based on the acquired information. The water level prediction device 100 includes, for example, an acquisition unit 110, a calculation unit 130, a prediction unit 140, an output unit 150, and an evaluation unit 160. Note that the evaluation unit 160 is not an essential component of the water level prediction device 100, and will be explained later. Functional units such as the acquisition unit 110, storage unit 120, calculation unit 130, prediction unit 140, output unit 150, and evaluation unit 160 are realized by a processor such as a CPU (Central Processing Unit) executing a program stored in a program memory. Furthermore, some or all of these functional units may be realized by hardware such as an LSI (Large Scale Integration), an ASIC (Application Specific Integrated Circuit), or an FPGA (Field-Programmable Gate Array), or may be realized by a combination of software and hardware. The storage unit 120 is realized by, for example, a hard disk drive (HDD), a flash memory, an electrically erasable programmable read-only memory (EEPROM), a read-only memory (ROM), or a random access memory (RAM), or a hybrid storage device that uses a combination of these. The storage unit 120 may also be realized by an external storage device accessible via various networks. An example of an external storage device is a network attached storage (NAS) device.

[0019] The acquisition unit 110 acquires water level data indicating the water level of the target river and precipitation data indicating the amount of precipitation in the target river. The acquisition unit 110 acquires water level data indicating the current water level at the prediction target point from the water level sensor 200. The acquisition unit 110 acquires precipitation data for the area including the prediction target point from the precipitation forecast information transmission device 300. The storage unit 120 stores the water level data and precipitation data acquired by the acquisition unit 110. The storage unit 120 also stores various data processed by the calculation unit 130 and the prediction unit 140. The calculation unit 130 calculates the cumulative precipitation data and the duration of precipitation for a predetermined time based on the amount of precipitation acquired by the acquisition unit 110. The calculation unit 130 calculates the cumulative precipitation data by adding up the date and time of the precipitation data and the numerical value of the precipitation amount for a predetermined time. The calculation unit 130 calculates the duration of precipitation by adding up the duration of non-zero precipitation using the date and time of the precipitation data and the numerical value of the precipitation amount. The prediction unit 140 predicts future water levels based on water level data, precipitation data, accumulated precipitation data, and precipitation duration. For each divided period into which the predetermined time calculated by the calculation unit 130 is divided, the prediction unit 140 predicts the water level for each divided period based on the water level prediction result for the divided period prior to that divided period, thereby generating water level prediction results for each divided period at the predetermined time. The predetermined time is the period for which the water level prediction device 100 predicts the future water level, and is, for example, a period of six hours from the present time. The output unit 150 outputs water level prediction data indicating the future water level predicted by the prediction unit 140 to the terminal device 400.

[0020] The water level sensor 200 is a sensor device that generates water level data at a target prediction point. The water level sensor 200 includes a camera device 210 that is fixedly installed at the water level prediction point, a detection unit 220, and a determination unit 230. Functional units such as the detection unit 220 and the determination unit 230 are realized by a processor such as a CPU executing a program stored in a program memory. The water level sensor 200 captures an image of the prediction target point using the camera device 210 and determines the water level based on the captured image. The water level is the height of the river's water surface measured from a reference level. The water level sensor 200 sets multiple water level determination areas in the captured image. The water level determination area is an image area that includes the boundary between the water surface and the water level mark. The water level sensor 200 may set multiple (e.g., three) water level determination areas at different positions in the captured image. The water level determination areas may be moved in accordance with the rise or fall of the river water level.

[0021] The detection unit 220 detects portions in which water is reflected in each of the multiple regions for water level determination. The detection unit 220 may detect the water level using, for example, a pre-stored trained model. The trained model is for segmentation generated by machine learning. The trained model inputs a captured image and outputs information indicating the classification of each pixel of the input image. The classification may be any classification that can distinguish between water, obstacles, and other objects. The trained model may detect classified regions in the captured image as polygonal regions, such as overlapping rectangles. A different trained model may be used for each classification. For example, an obstacle region may be detected using a trained model that detects it as an overlapping rectangle region.

[0022] The determination unit 230 determines whether or not each of the multiple water level determination regions will be used for water level determination based on the detection results by the detection unit 220, and determines the water level using the detection results of the water level determination regions determined to be used for water level determination. The determination unit 230 may calculate statistics of the values ​​related to the water level in each of the multiple water level determination regions based on the detection results by the detection unit 220, and determine whether or not each of the multiple water level determination regions will be used for water level determination based on the calculated statistics. The determination unit 230 may store a reference difference value of the values ​​related to the water level between the multiple water level determination regions and determine the water level using the reference difference value. The determination unit 230 may calculate a difference value of the values ​​related to the water level between the multiple water level determination regions based on the detection results by the detection unit 220, and determine whether or not each of the multiple water level determination regions will be used for water level determination based on the calculated difference value. As described above, the water level sensor 200 is equipped with a camera device 210, a detection unit 220, and a judgment unit 230, but it goes without saying that the water level sensor 200 may have other configurations as long as the water level prediction device 100 can acquire water level data indicating the water level of the target river.

[0023] The precipitation forecast information transmission device 300 is a server device that transmits precipitation forecast information in response to a request. The precipitation forecast information includes precipitation amount data indicating the current amount of precipitation. The precipitation forecast information transmission device 300 transmits the precipitation forecast information to the water level prediction device 100 in response to receiving a request from the water level prediction device 100 to transmit precipitation forecast information for a prediction target point.

[0024] The terminal device 400 is an information processing device such as a personal computer operated by a river administrator. The terminal device 400 receives the water level prediction data output by the water level prediction device 100 and displays information indicating future trends.

[0025] FIG. 2 is a diagram for explaining the river water level prediction model 142 according to the embodiment. The prediction unit 140 may predict the river water level using a river water level prediction model 142. The river water level prediction model 142 is a machine learning model created by performing statistical machine learning. The river water level prediction model 142 is trained using, for example, water level data at an arbitrary time point, precipitation forecast information, accumulated precipitation data, precipitation duration data, and future water levels as viewed from the water level at the arbitrary time point as training data. The training data for the river water level prediction model 142 may be, for example, several years' worth of water level data, precipitation forecast information, accumulated precipitation data, and precipitation duration data. When water level data, precipitation forecast information, accumulated precipitation data, and precipitation duration data at an arbitrary time point are input, the parameters of the machine learning model 142 are updated so as to output future water levels as viewed from the water level at the arbitrary time point. As a result, the river water level prediction model 142 outputs water level prediction data indicating future water levels in response to input of water level data at any time at the prediction target point, precipitation forecast information, cumulative precipitation data, and precipitation duration data.

[0026] FIG. 3 is a diagram illustrating an example of normalization processing according to the embodiment. The calculation unit 130 functions as a normalization unit that performs processing to normalize each piece of data, including water level data, precipitation data, accumulated precipitation amount data, and precipitation duration data, that is input to the prediction unit 140. The calculation unit 130 inputs the normalized values ​​to the river water level prediction model 142. The calculation unit 130 converts each piece of data into a numerical value within a preset range. For example, the calculation unit 130 represents each piece of data as data ranging from 0 to 1. In this case, the calculation unit 130 sets a minimum value Xmin and a maximum value Xmax for each piece of data, and performs the calculation (X-Xmin) / (Xmax-Xmin) when acquiring the value X of each piece of data, thereby converting each piece of data into a value between 0 and 1, thereby unifying the scale between the data. Furthermore, when data with large values ​​fluctuate, such as accumulated precipitation data, the fluctuation in the predicted value becomes large, which can lead to errors, and the accuracy of water level prediction can be improved.

[0027] FIG. 4 is a diagram showing an example of processing performed by the prediction unit 140 according to the embodiment. The prediction unit 140 inputs the current water level data and precipitation data, as well as the accumulated precipitation data and precipitation duration data calculated from the current water level data and precipitation data, into the river water level prediction model 142, and generates water level prediction data for five minutes from the current time. In this example, a period obtained by dividing six hours into five-minute intervals is defined as one divided period. The prediction unit 140 inputs the water level prediction data for 5 minutes from the present time and the precipitation data for 5 minutes from the present time contained in the precipitation forecast information, as well as the accumulated precipitation data and precipitation time data calculated from the water level prediction data for 5 minutes from the present time and the precipitation data for 5 minutes from the present time, into the river water level prediction model 142, and generates water level prediction data for 10 minutes from the present time. The prediction unit 140 generates water level prediction results for each divided period of six hours by predicting the water level for each divided period, which is obtained by dividing six hours into five-minute periods, based on the water level prediction results for the divided period prior to each divided period.

[0028] FIG. 5 is a diagram showing an example of a water level prediction result for a predetermined period according to the embodiment. The prediction unit 140 arranges the water level prediction results for each six-hour divided period on the time axis to create a water level prediction result for six hours from the present time, as shown in Fig. 5. The prediction unit 140 outputs the created water level prediction result for six hours from the present time to the terminal device 400. In this way, the prediction unit 140 uses the water level predicted in one divided period to predict the water level in the next divided period, and by continuing this process, it is possible to perform subdivided water level predictions up to a target future time point. The prediction unit 140 can improve the accuracy of water level predictions by using water level and precipitation data for each subdivided divided period, rather than predicting the water level several hours into the future at once. Furthermore, the prediction unit 140 can create water level prediction results that show in detail the progress of the water level prediction up to a specified time, allowing users to accurately guide evacuation.

[0029] FIG. 6 is a flowchart showing an example of data accumulation and expansion processing according to the embodiment. The acquisition unit 110 acquires current water level data (step S100), acquires precipitation forecast information including current and future precipitation data, and stores the information in the storage unit 120 (step S102). The calculation unit 130 calculates cumulative precipitation data for each divided period in a predetermined time from the current precipitation data and future precipitation data (forecast) included in the precipitation forecast information, and records the data in the storage unit 120 (step S104). The calculation unit 130 calculates precipitation duration data for each divided period in a predetermined time from the current precipitation data and future precipitation data (forecast) included in the precipitation forecast information, and records the data in the storage unit 120 (step S106). The calculation unit 130 performs an extended process to automatically calculate cumulative precipitation data and precipitation duration data from the water level data and precipitation forecast information.

[0030] FIG. 7 is a flowchart showing an example of a process for generating the river water level prediction model 142 according to the embodiment. The water level prediction device 100 acquires a certain number of past time series data from the storage unit 120 (step S200). The past time series data is data in which water level data, precipitation data, cumulative precipitation data, and precipitation time data are associated with date and time data. The certain number corresponds to the amount of data required to generate the river water level prediction model 142. The water level prediction device 100 duplicates the water level data from the time series data as prediction target data (step S202). Next, the water level prediction device 100 normalizes the time series data (step S204). Next, the water level prediction device 100 combines the prediction target data with the advanced learning data (step S206), inputs the combined data to the river water level prediction model 142 (step S208), and generates the river water level prediction model 142 (step S210).

[0031] FIG. 8 is a diagram illustrating a process of combining predicted target data with advance training data according to an embodiment. The water level prediction device 100 uses water level data for each date and time data (prediction target data, Figure 8(a)) to advance the prediction target data by five minutes to predict the water level five minutes from now (Figure 8(b)). The water level prediction device 100 combines the prediction target data advanced by five minutes with the learning data (Figure 8(c)) to create data (Figure 8(d)) in which the prediction target data and learning data are associated for each divided period, and inputs the created data into the river water level prediction model 142. In this way, the water level prediction device 100 trains the river water level prediction model 142 so that it inputs current water level data, precipitation data, accumulated precipitation data, and precipitation duration data and outputs prediction target data.

[0032] FIG. 9 is a flowchart showing an example of a process for predicting water levels in time series according to the embodiment. First, the calculation unit 130 acquires current water level data from the storage unit 120 (step S300), and acquires time-series data of precipitation forecast information (current and future precipitation data) from the storage unit 120 (step S302). The calculation unit 130 normalizes the acquired time-series data (current water level data and current and future precipitation data) (step S304). To predict the water level five minutes from now, the calculation unit 130 advances the current water level data by five minutes and combines it with the precipitation data, accumulated precipitation data, and precipitation duration data (step S306).

[0033] The prediction unit 140 inputs the data processed in step S306 into the river water level prediction model 142 (step S308), and sets the water level prediction data output from the river water level prediction model 142 as the water level data for the next divided period (step S310). The prediction unit 140 determines whether the divided period has reached a predetermined period (step S312), and if the divided period has not reached the predetermined period, repeats the processing from step S302 onwards (step S312: NO). If the divided period has reached the predetermined period (step S312: YES), the prediction unit 140 outputs the water level prediction data for each divided period from the current time to the predetermined period to the terminal device 400 (step S314).

[0034] FIG. 10 is a diagram illustrating the ensemble prediction of the river water level prediction model in the embodiment. The prediction unit 140 may predict water level data using an individual river water level prediction model 144 and a general-purpose river water level prediction model 146 as the river water level prediction model 142. The individual river water level prediction model 144 is a statistical machine learning model trained with water level data, precipitation data, accumulated precipitation data, and precipitation duration data for a target river. The general-purpose river water level prediction model 146 is a statistical machine learning model trained with water level data, precipitation data, accumulated precipitation data, and precipitation duration data for multiple rivers. The prediction unit 140 inputs the water level data, precipitation data, cumulative precipitation data, and precipitation time data of the target river into the individual river water level prediction model 144, and predicts future water levels based on the output of the individual river water level prediction model 144. The prediction unit 140 inputs the water level data, precipitation data, cumulative precipitation data, and precipitation time data of the target river into the general-purpose river water level prediction model 146, and predicts future water levels based on the output of the general-purpose river water level prediction model 146.

[0035] The prediction unit 140 predicts the future water level of the target river based on the output of the individual river water level prediction model 144 and the output of the general-purpose river water level prediction model 146. The prediction unit 140 may, for example, output the average value of the water level prediction data for five minutes from now output by the individual river water level prediction model 144 and the water level prediction data for five minutes from now output by the general-purpose river water level prediction model 146 as the final water level prediction data for five minutes from now. This allows the prediction unit 140 to improve the accuracy of the water level prediction data or suppress deterioration in accuracy compared to using either the individual river water level prediction model 144 or the general-purpose river water level prediction model 146.

[0036] For example, the prediction unit 140 may use the general-purpose river water level prediction model 146 when there is little learning data for the target river. When the amount of learning data for the target river increases and the amount of learning for the individual river water level prediction model 144 exceeds a threshold, the prediction unit 140 may use the individual river water level prediction model 144. In this way, the prediction unit 140 can improve the prediction accuracy by using the individual river water level prediction model 144 on the condition that the prediction accuracy of the individual river water level prediction model 144 has been improved.

[0037] There may be multiple general-purpose river water level prediction models 146 depending on the river environment. The general-purpose river water level prediction model 146 may be a statistical machine learning model trained using learning data (water level data, precipitation data, accumulated precipitation data, and precipitation duration data) for multiple rivers classified by river width (large, medium, and small), for example. The general-purpose river water level prediction model 146 may be a statistical machine learning model trained using learning data for multiple rivers classified by river upstream, midstream, and downstream, for example. The prediction unit 140 can select a general-purpose river water level prediction model 146 and improve the accuracy of water level prediction by acquiring information indicating the river width of the target river and information indicating the river's upstream, midstream, and downstream.

[0038] Fig. 11 is a diagram for explaining the process of automatically selecting a river water level prediction model in the embodiment. Fig. 12 is a flowchart showing an example of the process of automatically selecting a river water level prediction model in the embodiment. The prediction unit 140 may automatically select the river water level prediction model 142. The prediction unit 140 selects whether to use the individual river water level prediction model 144 or the general-purpose river water level prediction model 146 based on the accuracy of the prediction results according to the output of the individual river water level prediction model 144 and the accuracy of the prediction results according to the output of the general-purpose river water level prediction model 146.

[0039] For example, suppose that the current water level acquired by the acquisition unit 110 changes in the order 0.2 and 0.5 (step S400), the water level prediction result of the individual river water level prediction model 144 changes to 0.1 and 0.5 (step S402), and the water level prediction result of the general-purpose river water level prediction model 146 changes to 0.3 and 0.6 (step S404). The prediction unit 140 calculates the error between the current water level and the water level prediction result of the individual river water level prediction model 144, and the error between the current water level and the water level prediction result of the general-purpose river water level prediction model 146 (step S406). The prediction unit 140 calculates the average values ​​of the water level prediction results of the individual river water level prediction model 144 and the water level prediction result of the general-purpose river water level prediction model 146, such as 0.2 and 0.55 (step S408). The prediction unit 140 sets the water level prediction result with the smallest error to be output as the water level data for the next divided period (step S410). The prediction unit 140 may select whether to use the individual river water level prediction model 144 or the general-purpose river water level prediction model 146 for each divided period, and may select whether to use the individual river water level prediction model 144 or the general-purpose river water level prediction model 146 once per day, week, or month. The prediction unit 140 may compare the water level prediction results of the individual river water level prediction model 144 and the water level prediction results of the general-purpose river water level prediction model 146 with the water level data obtained after the prediction, and select the river water level prediction model from the individual river water level prediction model 144 or the general-purpose river water level prediction model 146 that outputs a water level prediction result with a smaller error from the water level data. This allows the prediction unit 140 to predict the water level using the river water level prediction model 142 that has the smallest error relative to the current water level.

[0040] FIG. 13 is a diagram showing another example of selecting a river water level prediction model in the embodiment. The prediction unit 140 may use a recursive prediction model 148 and a multiple time point prediction model 149 as the river water level prediction model 142. The recursive prediction model 148 is a statistical machine learning model that generates water level prediction results for each divided period at a predetermined time by recursively predicting the water level for each divided period based on the water level prediction results for the divided period prior to each divided period. The time point prediction model 149 is a statistical machine learning model that inputs current water level data, precipitation data, cumulative precipitation data, and precipitation time data and predicts the water level for each of multiple future time points included in a predetermined time from the current time. The time point prediction model 149 includes, for example, a time point prediction model 149 that predicts the water level five minutes from the current time, a time point prediction model 149 that predicts the water level ten minutes from the current time, and so on, and a time point prediction model 149 that predicts the water level six hours from the current time.

[0041] The prediction unit 140 inputs the water level data, precipitation data, cumulative precipitation data, and precipitation duration of the target river into the recursive prediction model 148, and predicts the water level for each future divided period based on the output of the recursive prediction model. The prediction unit 140 creates water level prediction data including predicted water levels from the present time to a predetermined period based on the output of the recursive prediction model 148. The prediction unit 140 inputs the water level data, precipitation data, cumulative precipitation data, and precipitation duration data of the target river into a multiple time point prediction model 149, and predicts water levels at multiple future time points based on the output of the multiple time point prediction model 149. The prediction unit 140 creates water level prediction data including predicted water levels from the present time to a predetermined period based on the output of the multiple time point prediction model 149. The prediction unit 140 predicts future water levels in the target river based on the output of the recursive prediction model 148 and the output of the time point prediction model 149 .

[0042] The prediction unit 140 may output, as final water level prediction data for a predetermined period, the average value of the water level prediction data for a predetermined period output by the recursive prediction model 148 and the water level prediction data for a predetermined period output by the multiple time point prediction model 149. This allows the prediction unit 140 to improve the accuracy of the water level prediction data or suppress deterioration in accuracy compared to using either the recursive prediction model 148 or the multiple time point prediction model 149. The prediction unit 140 may select and use the water level prediction result of the recursive prediction model 148 or the water level prediction result of the multiple time point prediction model 149, whichever is more accurate.

[0043] FIG. 14 is a diagram for explaining evaluation of the river water level prediction result in the embodiment. As shown in Fig. 1, the water level prediction device 100 includes an evaluation unit 160. The evaluation unit 160 evaluates the difference between the water level predicted by the prediction unit 140 and the actual water level, the difference between the arrival time of the predicted maximum water level and the arrival time of the actual maximum water level, and the difference between the predicted maximum water level and the actual maximum water level. One of the standards for evaluating water level prediction results is the standard given as an example by the National Institute for Land and Infrastructure Management, an organization affiliated with the Ministry of Land, Infrastructure, Transport and Tourism (left column in Fig. 14). Using this standard as a reference, the evaluation unit 160 performs the evaluation shown in the right column in Fig. 14.

[0044] The evaluation unit 160 evaluates the reference water level as acceptable if the difference between the predicted water level at the third hour and the actual measured water level is within ±30 cm within a six-hour water level prediction range, and evaluates it as unacceptable if the difference exceeds ±30 cm. The evaluation unit 160 calculates the time when the highest water level (peak water level) is predicted and the time when the highest water level actually occurs within the range of 0 to 3 hours of the 6-hour water level prediction for the peak occurrence time, and evaluates it as acceptable if the time difference is between -1 hour and 0.5 hours, and evaluates it as unacceptable if the time difference exceeds -1 hour to 0.5 hours. For peak water levels, the evaluation unit 160 determines the time when the highest water level was predicted and the time when the highest water level actually occurred within the range of 0 to 3 hours of the 6-hour water level prediction, and evaluates it as acceptable if the water level difference is between -10 cm and 30 cm, and evaluates it as unacceptable if the water level difference exceeds -10 cm to 30 cm. This allows the local government that manages the river and the prediction provider to use this as an indicator of the standards that must be met for water level prediction accuracy. By outputting information indicating the evaluation results by the evaluation unit 160 to the terminal device 400, the water level prediction device 100 can make the river manager aware that the water level prediction accuracy of the water level prediction device 100 meets the standards.

[0045] (Effects of the embodiment) As described above, the river water level prediction system 1 of the embodiment acquires water level data indicating the water level of a target river and precipitation data indicating the amount of precipitation in the target river, calculates cumulative precipitation data and precipitation duration data for a predetermined time based on the precipitation data, and predicts future water levels based on the water level data, precipitation data, cumulative precipitation data, and precipitation duration data. In this river water level prediction system 1, the prediction unit 140 can generate water level prediction results for each divided period in a predetermined time by predicting the water level for each divided period based on the water level prediction result for the divided period prior to that divided period. This river water level prediction system 1 can predict water levels with high accuracy without acquiring precipitation data or water level data from multiple points.

[0046] Although each embodiment and each variant have been described, these are merely examples and are not intended to limit the scope of the present invention. For example, one aspect of the present invention may be realized by combining any of the embodiments or variants, or a part of each embodiment or a part of each variant, with one or more other embodiments or one or more other variants.

[0047] The various processes described above relating to the water level prediction device 100 may be performed by recording a program for executing each process of the water level prediction device 100 in this embodiment on a computer-readable recording medium, and reading and executing the program recorded on the recording medium into a computer system.

[0048] Note that the term "computer system" here may include hardware such as the OS and peripheral devices. Furthermore, if a WWW system is used, the term "computer system" also includes the homepage provision environment (or display environment). Furthermore, "computer-readable recording media" refers to storage devices such as flexible disks, magneto-optical disks, ROMs, and writable non-volatile memory such as flash memory, portable media such as CD-ROMs, and hard disks built into computer systems.

[0049] Furthermore, the term "computer-readable recording medium" also includes a storage medium that stores a program for a certain period of time, such as a volatile memory (e.g., DRAM (Dynamic Random Access Memory)) within a computer system that serves as a server or client when the program is transmitted via a network such as the Internet or a communication line such as a telephone line. The program may also be transmitted from a computer system that stores the program in a storage device or the like to another computer system via a transmission medium or by transmission waves in the transmission medium.

[0050] Here, the "transmission medium" for transmitting the program refers to a medium that has the function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may also be one that realizes part of the above-mentioned functions. Furthermore, it may be a so-called differential file (differential program) that can realize the above-mentioned functions in combination with a program already recorded in a computer system.

[0051] Although the embodiments of the present invention have been described in detail above with reference to the drawings, the specific configuration is not limited to this embodiment, and the present invention also includes designs within the scope of the gist of the present invention. [Explanation of symbols]

[0052] 1. River water level forecasting system 100 Water level prediction device 110 Acquisition Department 120 Storage section 130 Calculation Unit 140 Prediction Department 142 River water level prediction model 144 Individual River Water Level Prediction Model 146 General-purpose river water level prediction model 148 Recursive Prediction Models 149 Point-in-time forecast model 150 Output section 160 Evaluation Department 200 Water Level Sensor 210 Camera equipment 220 Detector 230 Judgment section 300 Precipitation forecast information transmitter 400 Terminal Equipment

Claims

1. an acquisition unit that acquires water level data indicating the current water level in an area including the prediction target point of the target river, and precipitation data indicating the precipitation for each divided period obtained by dividing a predetermined future period from the present time into multiple periods in the area including the prediction target point of the target river; a calculation unit that calculates cumulative precipitation data for each divided period by summing up the precipitation amounts of the precipitation data for the predetermined period acquired by the acquisition unit, and calculates precipitation duration data indicating the duration of precipitation for each divided period by adding up the time during which the precipitation amount is not zero for the predetermined period; a prediction unit that predicts a future water level based on the water level data, the precipitation amount data, the accumulated precipitation amount data, and the precipitation duration data, The prediction unit repeats the process of generating water level prediction data for the second divided period immediately following the first divided period from the current time based on water level prediction data for the first divided period and precipitation data for the first divided period, and cumulative precipitation data and precipitation time data calculated from the water level prediction data for the first divided period and the precipitation data for the first divided period, for each divided period, until the specified period is reached, and outputs water level prediction data for each divided period from the current time to the specified period.

2. The prediction unit inputting the water level data, the precipitation data, the accumulated precipitation data, and the precipitation duration data of the target river into an individual river water level prediction model trained using the water level data, the precipitation data, the accumulated precipitation data, and the precipitation duration data of the target river, and predicting future water levels based on the output of the individual river water level prediction model; inputting the water level data, the precipitation data, the accumulated precipitation data, and the precipitation duration data of a target river into a general-purpose river water level prediction model trained using the water level data, the precipitation data, the accumulated precipitation data, and the precipitation duration data of a plurality of rivers, and predicting future water levels based on the output of the general-purpose river water level prediction model; predicting a future water level in a target river based on the output of the individual river water level prediction model and the output of the general-purpose river water level prediction model; The river water level prediction system according to claim 1 .

3. The prediction unit inputting the water level data, the precipitation data, the accumulated precipitation data, and the precipitation duration data of the target river into an individual river water level prediction model trained using the water level data, the precipitation data, the accumulated precipitation data, and the precipitation duration data of the target river, and predicting future water levels based on the output of the individual river water level prediction model; inputting the water level data, the precipitation data, the accumulated precipitation data, and the precipitation duration data of a target river into a general-purpose river water level prediction model trained using the water level data, the precipitation data, the accumulated precipitation data, and the precipitation duration data of a plurality of rivers, and predicting future water levels based on the output of the general-purpose river water level prediction model; selecting whether to use the individual river water level prediction model or the general-purpose river water level prediction model based on the accuracy of the prediction result according to the output of the individual river water level prediction model and the accuracy of the prediction result according to the output of the general-purpose river water level prediction model; The river water level prediction system according to claim 1 .

4. The prediction unit inputting the water level data, the precipitation data, the accumulated precipitation data, and the precipitation duration data of the target river into a recursive prediction model that generates water level prediction results for each divided period in the predetermined period by recursively predicting the water level for each divided period based on the water level prediction results for the divided period prior to each divided period, and predicting the water level for each future divided period based on the output of the recursive prediction model; inputting the water level data, precipitation data, cumulative precipitation data, and precipitation duration data of a target river into a multiple time point prediction model that inputs the current water level data, precipitation data, cumulative precipitation data, and precipitation duration data, and predicts water levels at multiple future time points included in the predetermined period from the current time point, and predicting water levels at multiple future time points based on the output of the multiple time point prediction model; predicting future water levels in a target river based on the output of the recursive prediction model and the output of the point-in-time prediction model; The river water level prediction system according to claim 1 .

5. The river water level prediction system of claim 1, further comprising an evaluation unit that evaluates the difference between the water level predicted by the prediction unit and the actual water level, the difference between the time of arrival of the predicted maximum water level and the time of arrival of the actual maximum water level, and the difference between the predicted maximum water level and the actual maximum water level.

6. a normalization unit that normalizes the water level data, the precipitation amount data, the accumulated precipitation amount data, and the precipitation duration data, The river water level prediction system according to claim 1 , wherein the prediction unit predicts the water level based on the water level data, the precipitation data, the accumulated precipitation data, and the precipitation time data normalized by the normalization unit.

7. An information processing device acquires water level data indicating the current water level of an area including the prediction target point of the target river, and precipitation data indicating the precipitation for each divided period obtained by dividing a predetermined period from the present time into a plurality of periods in the area including the prediction target point of the target river; calculating cumulative precipitation data for each divided period by adding up the precipitation amounts of the precipitation data for the predetermined period, and calculating precipitation duration data indicating the duration of precipitation for each divided period by adding up the duration of precipitation for the predetermined period when the precipitation amount is not zero; and a step by the information processing device of predicting a future water level based on the water level data, the precipitation amount data, the accumulated precipitation amount data, and the precipitation duration data, The step of predicting future water levels in this river water level prediction method repeats, for each divided period, a process of generating water level prediction data for the second divided period immediately following the first divided period from the present time based on water level prediction data for the first divided period and precipitation data for the first divided period, and precipitation cumulative data and precipitation duration data calculated from the water level prediction data for the first divided period and the precipitation data for the first divided period, until the specified period is reached, and outputs water level prediction data for each divided period from the present time to the specified period.

8. The computer of the information processing device A step of acquiring water level data indicating the current water level in an area including the prediction target point of the target river, and precipitation data indicating the precipitation for each divided period obtained by dividing a predetermined period from the present time into a plurality of periods in the area including the prediction target point of the target river; calculating cumulative precipitation data for each divided period by adding up the precipitation amounts of the precipitation data for the predetermined period, and calculating precipitation duration data indicating the duration of precipitation for each divided period by adding up the duration of precipitation for the predetermined period when the precipitation amount is not zero; the information processing device executes a step of predicting a future water level based on the water level data, the precipitation amount data, the accumulated precipitation amount data, and the precipitation duration data; The step of predicting future water levels is a river water level prediction program that repeats a process of generating water level prediction data for a second divided period from the present time immediately after the first divided period based on water level prediction data after a first divided period and precipitation data after the first divided period, and precipitation cumulative data and precipitation duration data calculated from the water level prediction data after the first divided period and the precipitation data after the first divided period, for each divided period, until the specified period is reached, and outputs water level prediction data for each divided period from the present time to the specified period.

Images