JP2024068121A - Prediction method, prediction device, and prediction program - Google Patents

Abstract

[Problem] To realize a technology that can predict the amount of water stored in a reservoir several days in advance. [Solution] A prediction method (S1) includes an acquisition step (S11) of acquiring information indicating variables that will affect the fluctuation of the amount of water stored in the reservoir from the current time to a date and time a predetermined time has elapsed, and a prediction step (S12) of predicting the amount of water stored in the reservoir at a date and time a predetermined time has elapsed by inputting the information indicating the variables that will affect the fluctuation of the amount of water acquired in the acquisition step (S11) into a trained model. [Selected Figure] Figure 1

Description

Application for exception to loss of novelty has been filed

The present invention relates to a prediction method, a prediction device, and a prediction program for predicting the amount of water stored in a reservoir.

In recent years, floods caused by localized heavy rains have become more frequent. To mitigate damage caused by floods, there are known techniques for predicting water levels in rivers and reservoirs such as dams and reservoirs. Patent Document 1 describes a technique for predicting water levels in rivers using machine learning.

JP 2022-40882 A

However, the conventional techniques described above are based on short-term predictions of up to six hours, and do not take into account inflows and outflows that affect water levels. As a result, it is necessary to predict water levels one to several days in advance, which poses the problem that they are not suitable for predicting water levels in reservoirs where water is frequently taken for agricultural purposes.

One aspect of the present invention was made in consideration of the above problems, and its purpose is to realize a technology that can predict the amount of water (water level) in a reservoir several days in advance.

In order to solve the above problem, a prediction method according to one embodiment of the present invention is a prediction method for predicting the amount of water stored in a reservoir, and includes an acquisition step of acquiring information indicating variables that affect fluctuations in the amount of water stored in the reservoir from the current time to a date and time after a predetermined time has elapsed, and a prediction step of predicting the amount of water stored in the reservoir at a date and time after the predetermined time has elapsed by inputting the information acquired in the acquisition step into a trained model that is trained to use the information indicating the variables that affect fluctuations in the amount of water stored in the reservoir as input information and output a prediction result of the amount of water stored in the reservoir at a date and time after the predetermined time has elapsed.

In order to solve the above problem, a prediction device according to one aspect of the present invention is a prediction device that predicts the amount of water stored in a reservoir, and includes an acquisition unit that acquires information indicating variables that affect fluctuations in the amount of water stored in the reservoir from the current time to a date and time after a predetermined time has elapsed, and a prediction unit that predicts the amount of water stored in the reservoir at a date and time after the predetermined time has elapsed by inputting the information acquired by the acquisition unit into a trained model that is trained to use the information indicating the variables that affect fluctuations in the amount of water stored in the reservoir as input information and output a prediction result of the amount of water stored in the reservoir at a date and time after the predetermined time has elapsed.

To solve the above problems, a prediction program according to one aspect of the present invention is a program for causing a computer to function as the above-mentioned prediction device, and causes the computer to function as each of the above-mentioned parts.

According to one aspect of the present invention, it is possible to predict the amount of water stored in a reservoir several days in advance.

FIG. 1 is a schematic diagram showing a configuration of a prediction system according to an embodiment of the present invention. 1 is a block diagram showing a configuration of a prediction system according to an embodiment of the present invention. 1 is a flow diagram showing the flow of a prediction method implemented by a prediction device according to an embodiment of the present invention. FIG. 2 is a diagram for specifically explaining a prediction method implemented by a prediction device according to an embodiment of the present invention. FIG. 13 is a block diagram illustrating a configuration of a prediction device according to a modified example of the present invention. FIG. 1 is a diagram illustrating an example of a trained model according to an embodiment of the present invention. FIG. 13 is a diagram showing an example of the results of an embodiment of the present invention. FIG. 13 is a diagram showing another example of the results of an embodiment of the present invention. 1 is a table summarizing examples of the present invention.

[Outline of forecasting method]
A prediction method according to an embodiment of the present invention will be described in detail with reference to Figures 1 to 6. The prediction method is a method for predicting the amount of water stored in a reservoir several days ahead based on information indicating variables that affect fluctuations in the amount of water stored in the reservoir.

An example of a variable that affects the fluctuation of the stored water volume is a variable that indicates the cumulative amount of precipitation in the reservoir from a specified time to a date and time a specified time later (hereinafter, the variable is also referred to as "LT XX_Rsum" (XX is an identification number for identifying each of the multiple LT XX_Rsum)). Another example of a variable that affects the fluctuation of the stored water volume is a variable that indicates the amount of precipitation at a specified interval from a specified time to a date and time a specified time later (hereinafter, the variable is also referred to as "LT XX_Rsum" (XX is an identification number for identifying each of the multiple LT XX_Rsum)). Still other examples of variables that affect the fluctuation of the stored water volume are a variable that indicates the amount of water stored in the reservoir a specified time before the current time (hereinafter, the variable is also referred to as "SV"), a variable that indicates the amount of water stored at the current time (hereinafter, the variable is also referred to as "EV"), and a variable that indicates the difference between the amount of water stored at the current time and a specified time before the current time (hereinafter, the variable is also referred to as "InOut"). Yet another example of a variable that affects the fluctuation of the stored water volume is a variable that indicates the maximum value of the stored water volume from a predetermined time before the current time to the current time (hereinafter, this variable is also referred to as "Max") and a variable that indicates the minimum value (hereinafter, this variable is also referred to as "Min") in the reservoir. Yet another example of a variable that affects the fluctuation of the stored water volume is a variable that indicates the accumulated value of the precipitation from a predetermined time before the current time to the current time in the reservoir (hereinafter, this variable is also referred to as "Rsum").

Yet another example of a variable that affects fluctuations in stored water volume is a variable that indicates the cumulative amount of discharge from a given time to a date and time a given time later in a reservoir (hereinafter, this variable is also referred to as "LTxxx_Ssum" (xxx is an identification number for identifying each of the multiple LTxxx_Ssums)). Yet another example of a variable that affects fluctuations in stored water volume is a variable that indicates the cumulative amount of discharge from a given time interval from the current time to a date and time a given time later in a reservoir (hereinafter, this variable is also referred to as "LTxxx_Ssum" (xxx is an identification number for identifying each of the multiple LTxxx_Ssums)). Yet another example of a variable that affects fluctuations in stored water volume is a variable that indicates a tentative amount of stored water a given time later from the current time in a reservoir (hereinafter, this variable is also referred to as "LTxxx_EV" (xxx is an identification number for identifying each of the multiple LTxxx_EVs)). Yet another example of a variable that affects fluctuations in the amount of stored water is a variable that indicates the difference between the amount of stored water at the current time in a reservoir (the above-mentioned "EV") and the provisional amount of stored water after a certain period of time has passed (the above-mentioned "LTXXX_EV") (hereinafter, this variable will also be referred to as "LTXXX_InOut" (XXX is an identification number for identifying each of the multiple LTXXX_InOuts)). The calculation methods for LTXXX_Ssum, LTXXX_Ssum, LTXXX_EV, and LTXXX_InOut will be described later.

[Configuration of prediction system]
The configuration of a prediction system 100 according to an embodiment of the present invention will be described with reference to Fig. 1. Fig. 1 is a schematic diagram showing the configuration of the prediction system according to this embodiment.

As shown in FIG. 1, the prediction system 100 includes a prediction device 1, a sensor 2, a precipitation database 3, and a user terminal 4. The sensor 2, the precipitation database 3, and the user terminal 4 are configured to be able to communicate with the prediction device 1 via a network 5. Note that the number of sensors, precipitation databases, and user terminals each configured to be able to communicate with one prediction device 1 may be one or more.

The prediction system 100 is a system that executes the prediction method described above. More specifically, in the prediction system 100, the prediction device 1 predicts the amount of water stored in the reservoir several days ahead based on information acquired from the sensor 2 and the precipitation database 3. The prediction device 1 also transmits a signal to the user terminal 4 to display the prediction result. The user terminal 4 displays the prediction result of the amount of stored water several days ahead based on the received signal. Therefore, the prediction system 100 displays the prediction result of the amount of stored water several days ahead to the administrator Ua, so that the administrator Ua can take necessary measures such as releasing water in advance.

The prediction device 1, the sensor 2, the precipitation database 3, and the user terminal 4 that constitute the above-mentioned prediction system 100 will be described in detail with reference to FIG. 2. FIG. 2 is a block diagram showing the configuration of the prediction system 100 according to this embodiment.

<Prediction device>
The prediction device 1 is a device that implements a prediction method according to an embodiment of the present invention. The flow of the prediction method will be described later with reference to different drawings.

As shown in FIG. 2, the prediction device 1 includes a device control unit 11, a device communication unit 12, and a device storage unit 13. As shown in FIG. 2, the prediction device 1 is also configured to be able to communicate with a sensor 2, a precipitation database 3, and a user terminal 4 via a network 5.

(Device communication unit)
The device communication unit 12 communicates with the sensor 2, the precipitation database 3, and the user terminal 4. As an example, the device communication unit 12 receives information indicating the measurement results of the amount of water stored in the reservoir from the sensor 2. The device communication unit 12 also receives information indicating precipitation data from the precipitation database 3. Furthermore, the device communication unit 12 transmits to the user terminal 4 a prediction result of the amount of water stored in the reservoir at a date and time a predetermined time after the current time t (hereinafter, the date and time a predetermined time after the current time t is referred to as "time t+Δt").

The device communication unit 12 may transmit the prediction result each time a prediction is completed, may transmit it at a predetermined time interval that can be set arbitrarily (e.g., every 24 hours), or may transmit it in response to a transmission request from the user terminal 4.

(Device Control Unit)
The device control unit 11 controls each of the components included in the prediction device 1. The device control unit 11 also includes an acquisition unit 111 and a prediction unit 112.

The acquisition unit 111 acquires information via the device communication unit 12. As an example, the acquisition unit 111 acquires information necessary for predicting the amount of water stored in the reservoir. One example of the information acquired by the acquisition unit 111 is information indicating the measurement results of the amount of stored water and information indicating precipitation data received by the device communication unit 12 from the sensor 2 and the precipitation database 3, respectively. The information indicating the measurement results of the amount of stored water and the information indicating the precipitation data include variables that affect the fluctuations in the amount of stored water in the reservoir described above. Therefore, it can also be said that the acquisition unit 111 acquires information indicating variables that affect the fluctuations in the amount of stored water in the reservoir from the current time t to the time t+Δt.

The prediction unit 112 predicts the amount of water stored in the reservoir. As an example, the prediction unit 112 predicts the amount of water stored in the reservoir at time t+Δt based on the information acquired by the acquisition unit 111.

An example of a prediction method executed by the prediction unit 112 is a prediction method using a trained model. Specifically, the prediction unit 112 obtains a prediction result of the reservoir's water storage volume at time t + Δt by inputting the information acquired by the acquisition unit 111 into a trained model that has been trained to output a prediction result of the reservoir's water storage volume at time t + Δt using information indicating variables that affect fluctuations in the reservoir's water storage volume as input information.

The prediction unit 112 may have the function of a trained model, or a device different from the prediction device 1 may have the function of a trained model, and the prediction unit 112 may output information indicating variables that affect fluctuations in the water volume to the device, and obtain prediction results from the device.

(Trained model 1)
The trained model in this embodiment is a model trained to output a prediction result of the water volume of the reservoir at time t+Δt using information indicating variables that affect the fluctuation of the water volume in the reservoir as input information. The variables that affect the fluctuation of the water volume in the reservoir are as described above.

When information indicating variables that affect fluctuations in the amount of stored water is input, the trained model outputs a prediction result for the amount of water stored in the reservoir at time t + Δt. Note that the trained model may be a model generated by deep learning, or a model generated by machine learning other than deep learning.

(Trained model 2)
The trained model in this embodiment may be a trained model trained by a stacking ensemble method. The stacking ensemble method is a method for effectively combining multiple prediction results by multiple prediction models (trained models) to derive (output) a final prediction. A trained model M, which is an example of a trained model trained by such a stacking ensemble method, will be described with reference to FIG. 6. FIG. 6 is a diagram showing the configuration of the trained model M.

As shown in FIG. 6, the trained model M is a model trained to combine the prediction results of the reservoir's water volume at time t + Δt by multiple trained models (Model 1 to Model 4 in FIG. 6) to which different combinations of variables are input, and to output the final prediction result of the reservoir's water volume at time t + Δt. That is, each of the multiple trained models Model 1 to Model 4 is a model trained to input a different combination of variables and output a prediction result of the reservoir's water volume at time t + Δt. The trained model M processes the multiple prediction results output from each of Model 1 to Model 4 using a stacking ensemble method, and outputs the final prediction result of the reservoir's water volume at time t + Δt. In FIG. 6, the multiple trained models included in the trained model M are four, Model 1 to Model 4, but the number of models included in the trained model M to which different combinations of variables are input is not limited.

In Figure 6, SV, InOut, LTXXX_Ssum, and LTXXX_Rsum are input to Model1. Model1 also outputs Model1 Output as the predicted water volume of the reservoir at time t+Δt.

6, SV, LTXXX_EV0.5-3, and LTXXX_InOut0.5-3 are input to Model4. Model4 Output is output from Model4 as a prediction result of the water volume of the reservoir at time t+Δt. LTXXX_EV0.5-3 and LTXXX_InOut0.5-3 are variables that indicate the following, respectively.
・LT〇〇_EV0.5-3: A variable indicating the provisional water storage volume after a predetermined time has elapsed from the current time when the coefficient described later is increased from 0.5 to 3 in increments of 0.5 ・LT〇〇_InOut0.5-3: A variable indicating the difference between the water storage volume at the current time and the provisional water storage volume after a predetermined time has elapsed (LT〇〇_EV0.5-3) In the trained model M, Model1 Output to Model4 Output output from Model1 to Model4 are used as Input Data for the AI algorithm. Model1 Output to Model4 Output used as Input Data are output as Output by the AI algorithm, which is the predicted result of the water storage volume of the reservoir at the time t + Δt. In addition, the AI algorithm used in the stacking ensemble method may be a deep learning algorithm, or a machine learning algorithm other than deep learning.

(Example 1 of variable calculation method)
As an example of a variable calculation method, an example of a calculation method of LT00_Ssum (a variable indicating an integrated value of the discharge amount from a predetermined time to a date and time after a predetermined time has elapsed) and LT0_Ssum (a variable indicating an integrated value of the discharge amount at a predetermined interval from a predetermined time to a date and time after a predetermined time has elapsed) will be described. The discharge amount may be a discharge amount for irrigation, such as an irrigation period and an irrigation time.

As an example, the discharge amount may be set as the amount of water required for irrigation based on the above-mentioned variable InOut, the irrigated field area, and the discharge time. In this case, the value indicated by LT〇〇_Rsum or LT〇_Rsum may be referenced, and if the amount of precipitation up to a specified time exceeds a specified amount of rainfall, no discharge (discharge amount is set to zero) may be set. Furthermore, the discharge amount may be set based on the above-mentioned variable InOut and the pattern of agricultural work. For example, the amount corresponding to the plowing period may be discharged once on a specified day of the week (e.g. Saturday) before and after the start of irrigation every year, and after the plowing period, the amount may be set to be discharged every week on a day of the week other than the specified day of the week (e.g. Sunday). Furthermore, it may be set so that no discharge for irrigation is performed during the mid-drying period.

(Example 2 of variable calculation method)
As another example of a variable calculation method, a calculation method of LTOO_EV (a variable indicating a provisional amount of stored water after a predetermined time has elapsed from the current time) will be described.

First, it is assumed that water will be released from the reservoir into the beneficiary paddy fields in accordance with agricultural work and the rice growth process (irrigation, plowing, inter-seasonal drainage), and that no water will be released from the reservoir during non-irrigation periods (discharge volume is set to zero). If there is data on low water level management of the reservoir during non-irrigation periods, that data will be referenced.

Next, LT__EV is calculated by adding the amount of water stored at the specified time (SV or EV) to the amount of precipitation that falls directly on the reservoir during the LT period minus the amount of water discharged from the reservoir during the LT period. This calculation is continued until the end of the target period.

As an example, LT XX_EV one day ahead (after 24 hours have passed) may be calculated using the following formulas (1) to (4). The "coefficients" below are coefficients for correcting calculation errors due to precipitation amounts corresponding to the catchment area of a reservoir. As an example, the coefficients increased in increments of 0.5 from 0.5 to 3 are referred to as the above-mentioned "LT XX_EV 0.5 to 3."
・When SV <= full water storage volume, LT〇〇_Rsum < 5 (mm), LT〇〇_EV = SV - LT〇〇_Ssum + (LT〇〇_Rsum × coefficient) ... (1)
・When SV <= full water storage volume, LT〇〇_Rsum >= 5 (mm), LT〇〇_EV = SV + (LT〇〇_Rsum × coefficient) ... (2)
・When SV>full water storage volume and LT〇〇_Rsum<5 (mm), LT〇〇_EV = full water storage volume - LT〇〇_Ssum + (LT〇〇_Rsum × coefficient) ... (3)
・When SV>full water storage volume and LT〇〇_Rsum>=5 (mm), LT〇〇_EV = full water storage volume + (LT〇〇_Rsum × coefficient) ... (4)
(Example 3 of variable calculation method)
As yet another example of a variable calculation method, a calculation method for LTOO_InOut (a variable indicating the difference between the amount of water stored at the current time and the provisional amount of water stored after a predetermined time has elapsed) will be described.

LT〇〇_InOut is calculated by finding the difference between the above-mentioned EV and LT〇〇_EV.

(Device Memory Unit)
Various information is recorded in the device storage unit 13. As an example, the device storage unit 13 records precipitation amount data 131 which is information indicating precipitation amount data received from the precipitation amount database 3, stored water amount data 132 which is information indicating the measurement result of the stored water amount measured by the sensor 2, and prediction result 133 which is information indicating the prediction result of the stored water amount predicted by the prediction unit 112.

<Sensor>
The sensor 2 measures the amount of water stored in the reservoir. As an example, the sensor 2 measures the amount of water stored at predetermined time intervals (e.g., every hour) and transmits the measurement results to the prediction device 1. The sensor 2 may transmit the prediction results every time a measurement is performed, or may transmit the prediction results at predetermined time intervals (e.g., every 24 hours) that can be set arbitrarily.

<Precipitation Database>
The precipitation database 3 stores precipitation data including forecast data of precipitation in an area including the reservoir. The precipitation database 3 may be a database storing data acquired from the website of the Japan Meteorological Agency or a weather forecast provider. The precipitation database 3 may transmit precipitation data to the prediction device 1 at predetermined time intervals (e.g., every 24 hours) or may transmit the data every time the information stored in the precipitation database 3 is updated.

<User device>
The user terminal 4 is a device that displays the result of the water volume in the reservoir predicted by the prediction device 1. In this embodiment, a notebook PC (Personal Computer) is used as the user terminal 4, but is not limited to this. For example, a desktop PC, a smartphone, or a tablet PC may be used as the user terminal 4. The user terminal 4 also includes an input unit 41, a terminal communication unit 42, a terminal control unit 43, a terminal storage unit 44, and a display unit 45.

(Input section)
The input unit 41 transmits an instruction signal indicating the content of a user operation on the user terminal 4 to the terminal control unit 43. A specific example of the instruction signal is an instruction signal indicating a request to view the prediction result of the water storage volume at the date and time specified by the administrator Ua. Specific examples of the input unit 41 are a mouse, a keyboard, and a touch panel.

(Terminal communication unit)
The terminal communication unit 42 communicates with the prediction device 1 via the network 5. As an example, the terminal communication unit 42 receives a prediction result of the water volume of the reservoir at time t+Δt from the prediction device 1. The terminal communication unit 42 may transmit a request to transmit the prediction result to the prediction device 1.

(Terminal control unit)
The terminal control unit 43 controls each component included in the user terminal 4. The terminal control unit 43 also includes an input receiving unit 431 and a display control unit 432.

The input reception unit 431 acquires the instruction signal transmitted by the input unit 41. The input reception unit 431 transmits the acquired instruction signal to the prediction device 1 via the device communication unit 42.

The display control unit 432 outputs a display image to the display unit 45. As an example, in response to a viewing request from the administrator Ua, the display control unit 432 outputs a display image including the prediction result predicted by the prediction device 1 to the display unit 45.

(Terminal storage unit)
Various types of information are stored in the terminal storage unit 44. As an example, information indicating the prediction result of the stored water volume received from the prediction device 1 is stored in the terminal storage unit 44.

(Display)
The display unit 45 displays the display image output from the terminal control unit 43. As an example, the display unit 45 displays a display image including the predicted result of the stored water volume predicted by the prediction device 1. In addition to the predicted result of the stored water volume, the image may include text or the like recommending the implementation of necessary measures based on the predicted result (e.g., prior release). An example of the display unit 45 is a computer display.

[Flow of prediction method S1]
The prediction method S1 is a method that the prediction device 1 executes to predict the water volume of the reservoir at time t+Δt. The prediction method S1 will be described with reference to Fig. 3. Fig. 3 is a flow chart showing the flow of the prediction method S1.

As shown in FIG. 3, the prediction method S1 includes an acquisition step S11 and a prediction step S12. The timing at which the prediction device 1 performs the prediction method S1 can be set arbitrarily. For example, the prediction device 1 may perform the prediction method S1 each time it receives information from the sensor 2 and the precipitation database 3, or may perform the prediction method S1 each time it receives a request from the user terminal 4 to send the prediction result at time t+Δt.

<Acquisition process S11>
The acquisition step S11 is a step in which the acquisition unit 111 acquires information indicating the measurement result of the amount of water stored in the reservoir received from the sensor 2 and information indicating the precipitation data received from the precipitation database 3. That is, the acquisition step S11 is a step in which the acquisition unit 111 acquires information indicating variables that affect the fluctuation of the amount of water stored in the reservoir from the current time t to the time t+Δt.

<Prediction step S12>
The prediction process S12 is a process in which the prediction unit 112 predicts the water storage volume of the reservoir at time t + Δt by inputting information indicating variables that affect the fluctuation in the water storage volume in the reservoir from the current time t to time t + Δt, which was acquired in the acquisition process S11, into the trained model.

Information indicating the predicted result of the water volume of the reservoir at time t+Δt predicted in the prediction step S12 is transmitted to the device memory unit 13 or the user terminal 4.

Below, the variables input to the trained model will be described in detail with reference to Figure 4. Figure 4 is a diagram for specifically explaining the prediction method implemented by the prediction device 1.

(Example 1)
As an example, in the prediction step S12, the prediction unit 112 inputs a variable indicating an integrated value of the amount of precipitation in the reservoir from the predetermined time to a date and time Δt after the predetermined time to the trained model. The predetermined time may be the current time t as shown in FIG. 4, or may be a time point prior to the current time t (for example, t-24h). As shown in FIG. 4, when the predetermined time is the current time t and Δt (predetermined time) is "72h", the prediction unit 112 inputs a variable indicating an integrated value of the amount of precipitation in the reservoir at a time point t+72h after 72 hours have elapsed from the current time t to the trained model. The prediction unit 112 calculates the integrated value of the amount of precipitation in the reservoir from the predetermined time to a date and time Δt after the predetermined time based on information acquired from the precipitation database 3.

(Example 2)
As another example, in the prediction step S12, the prediction unit 112 inputs variables indicating the amount of precipitation at each predetermined interval from the predetermined time to the date and time after Δt has elapsed to the trained model. The predetermined time may be the current time t as shown in FIG. 4, or may be a time point before the current time t (for example, t-24h). As shown in FIG. 4, when the predetermined time is the current time t, Δt (predetermined time) is "72h", and the predetermined interval is "24h", the prediction unit 112 inputs variables indicating the amount of precipitation at each 24h interval from the current time t to the time point t+72h after 72 hours have elapsed to the trained model. More specifically, the prediction unit 112 inputs variables indicating the amount of precipitation from the current time t to t+24h, the amount of precipitation from t+24h to t+48h, and the amount of precipitation from t+48h to t+72h to the trained model. The prediction unit 112 calculates the amount of precipitation for each predetermined interval from a predetermined time to a date and time after Δt based on information acquired from the precipitation database 3 .

(Example 3)
As yet another example, in the prediction step S12, in addition to (or instead of) the above example, the prediction unit 112 inputs a variable indicating the amount of water stored in the reservoir at a predetermined time before the current time t (hereinafter, the predetermined time before the current time t will be referred to as "t-Δt1") and a variable indicating the difference between the amount of water stored at the current time t and the amount of water stored at a predetermined time before the current time t into the trained model. As shown in FIG. 4, when Δt1 (the predetermined time) is "24h", the prediction unit 112 inputs a variable indicating the amount of water stored at t-24h, which is 24h before the current time t, and a variable indicating the difference between the amount of water stored at the current time t and the amount of water stored at t-24h into the trained model. The prediction unit 112 calculates the variable indicating the amount of water stored at t-Δt1 and the variable indicating the difference between the amount of water stored at the current time t and the amount of water stored at a predetermined time before the current time t based on information acquired from the sensor 2.

(Example 4)
As yet another example, in the prediction step S12, in addition to (or instead of) the above example, the prediction unit 112 inputs variables indicating the maximum and minimum values of the water volume in the reservoir from t-Δt1 to the current time t into the trained model. As shown in FIG. 4, when Δt1 (predetermined time) is "24h", the prediction unit 112 inputs variables indicating the maximum and minimum values of the water volume from t-24h to the current time t into the trained model. The prediction unit 112 calculates the variables indicating the maximum and minimum values of the water volume from t-Δt1 to the current time t based on the information acquired from the sensor 2.

(Example 5)
As yet another example, in the prediction step S12, in addition to (or instead of) the above-mentioned example, the prediction unit 112 inputs a variable indicating the integrated value of the amount of precipitation in the reservoir from t-Δt1 to the current time t into the trained model. As shown in FIG. 4, when Δt1 (a predetermined time) is "24 h", the prediction unit 112 inputs a variable indicating the integrated value of the amount of precipitation from t-24 h to the current time t into the trained model. The prediction unit 112 calculates the variable indicating the integrated value of the amount of precipitation from t-Δt1 to the current time t based on information acquired from the precipitation database 3.

(Example 6)
As yet another example, in the prediction step S12, in addition to (or instead of) the above example, the prediction unit 112 predicts the water volume of the reservoir at time t+Δt using a trained model trained by the stacking ensemble method described above. As an example, as shown in FIG. 6, variables are input to the trained model M described above. The prediction unit 112 predicts the water volume of the reservoir at a date and time after a predetermined time has elapsed based on variables (LTOO_Rsum, LTOO_Rsum, and Rsum) calculated based on information acquired from the precipitation database 3, variables (SV, EV, InOut, Max, and Min) calculated based on information acquired from the sensor 2, and variables (LTOO_Ssum, LTOO_Ssum, LTOO_EV, and LTOO_InOut) calculated by the above method.

For example, during irrigation periods, large fluctuations in the water storage volume occur due to artificial discharges such as discharges for irrigation or low water level management. However, the prediction device 1 predicts the water storage volume by inputting variables related to the discharge amount that affect the water storage volume during the irrigation period (LTXXX_Ssum, LTXXX_Ssum, LTXXX_EV, and LTXXX_InOut) into the trained model. Therefore, the prediction device 1 can accurately predict the water storage volume regardless of whether it is an irrigation period or a non-irrigation period.

The "trained model" in Examples 7 to 10 below is not limited to a trained model machine-learned using the stacking ensemble method, but may be a trained model machine-learned using a method other than the stacking ensemble method.

(Example 7)
As yet another example, in the prediction step S12, in addition to (or instead of) the above-mentioned example, the prediction unit 112 inputs a variable indicating an integrated value of the discharge amount in the reservoir from a predetermined time to a date and time after Δt has elapsed into the trained model. The predetermined time may be the current time t as shown in FIG. 4, or may be a time point before the current time t (for example, t-24h). As shown in FIG. 4, when the predetermined time is the current time t and Δt (predetermined time) is "72h", the prediction unit 112 inputs a variable indicating an integrated value of the discharge amount from the current time t to a time point t+72h after 72 hours have elapsed into the trained model. The method of calculating the integrated value of the discharge amount from the predetermined time to a date and time after Δt has elapsed is as described above.

(Example 8)
As yet another example, in the prediction step S12, in addition to (or instead of) the above-mentioned example, the prediction unit 112 inputs a variable indicating an integrated value of the discharge amount at each predetermined interval from the predetermined time to the date and time after Δt has elapsed in the reservoir to the trained model. The predetermined time may be the current time t as shown in FIG. 4, or may be a time point before the current time t (for example, t-24h). As shown in FIG. 4, when the predetermined time is the current time t, Δt (predetermined time) is "72h", and the predetermined interval is "24h", the prediction unit 112 inputs the discharge amount at each 24h interval from the current time t to the time point t+72h after 72 hours have elapsed to the trained model. More specifically, the prediction unit 112 inputs variables indicating the discharge amount from the current time t to t+24h, the discharge amount from t+24h to t+48h, and the discharge amount from t+48h to t+72h to the trained model. The method for calculating the integrated value of the discharge amount at each predetermined interval from the predetermined time to the date and time after Δt has elapsed is as described above.

(Example 9)
As yet another example, in the prediction step S12, in addition to (or instead of) the above example, the prediction unit 112 inputs a variable indicating a provisional amount of water stored in the reservoir from the current time t until Δt has elapsed into the trained model. As shown in Fig. 4, when the current time t and Δt (predetermined time) are "72h", the prediction unit 112 inputs a variable indicating a provisional amount of water stored at t+72h, 72 hours after the current time t, into the trained model. The method of calculating the variable indicating the provisional amount of water stored from the current time t until Δt has elapsed is as described above.

(Example 10)
As yet another example, in the prediction step S12, in addition to (or instead of) the above example, the prediction unit 112 inputs a variable indicating the difference between the water volume at the current time t and the provisional water volume after Δt has elapsed from the current time t in the reservoir to the trained model. As shown in FIG. 4, when the current time t and Δt (predetermined time) are "72h", the prediction unit 112 inputs a variable indicating the difference between the water volume at the current time t and the provisional water volume at t+72h after 72 hours have elapsed to the trained model. The method of calculating the variable indicating the difference between the water volume at the current time and the provisional water volume after the predetermined time has elapsed from the current time is as described above.

<Effects of prediction method S1>
In this way, in the prediction method S1, the amount of water stored in the reservoir at t+Δt is predicted by inputting information indicating variables that affect the fluctuation in the amount of water stored in the reservoir from the current time t to t+Δt into a trained model that predicts the amount of water stored in the reservoir at t+Δt. Therefore, the prediction method S1 can predict the amount of water stored in the reservoir several days in advance.

[Modifications of prediction system]
A prediction system 100a, which is a modification of the prediction system 100 described above, will be described with reference to FIG.

As shown in FIG. 5, the prediction system 100a includes a prediction device 1a, a sensor 2, and a precipitation database 3. In the prediction system 100a, the sensor 2 and the precipitation database 3 are configured to be able to communicate with the prediction device 1a via a network 5. Note that the number of sensors 2 and precipitation databases 3 configured to be able to communicate with one prediction device 1a may be one or more. The sensor 2 and the precipitation database 3 are as described in the first embodiment.

In the prediction system 100a, the prediction device 1a has the functions of the prediction device 1 in the prediction system 100 described above, as well as the functions of the user terminal 4. In other words, the prediction device 1a predicts the amount of water stored in the reservoir several days ahead based on information acquired from the sensor 2 and the precipitation database 3. Then, the prediction device 1a displays the prediction results of the amount of water stored several days ahead based on the prediction results.

<Prediction device>
The prediction device 1a is a device that performs the prediction method in this modification. The prediction method is as described in the first embodiment.

As shown in Fig. 5, the prediction device 1a includes a device control unit 11a, a device communication unit 12a, a device storage unit 13a, an input unit 14a, and a display unit 15a. The device communication unit 12a, the device storage unit 13a, the input unit 14a, and the display unit 15a are as described in the device communication unit 12, the device storage unit 13, the input unit 41, and the terminal storage unit 44 in the above-mentioned embodiment, respectively.

(Device Control Unit)
The device control unit 11a controls each component included in the prediction device 1a. The device control unit 11a also includes an acquisition unit 111a, a prediction unit 112a, and a display control unit 113a. The acquisition unit 111a, the prediction unit 112a, and the display control unit 113a are as described in the acquisition unit 111, the prediction unit 112, and the display control unit 432 in the above-mentioned embodiment, respectively.

In this way, the prediction system 100a includes a prediction device 1a that has the functions of the user terminal 4 in addition to the functions of the prediction device 1 in the above-described embodiment. Even in this configuration, the prediction system 100a achieves the same effects as the prediction system 100 in the above-described embodiment.

[Software implementation example]
The functions of the prediction device 1 (hereinafter referred to as the "device") can be realized by a program for causing a computer to function as the device, and a program for causing a computer to function as each control block of the device (particularly each part included in the device control unit 11).

In this case, the device includes a computer having at least one control device (e.g., a processor) and at least one storage device (e.g., a memory) as hardware for executing the program. The functions described in each of the above embodiments are realized by executing the program using the control device and storage device.

The program may be recorded on one or more computer-readable recording media, not on a temporary basis. The recording media may or may not be included in the device. In the latter case, the program may be supplied to the device via any wired or wireless transmission medium.

In addition, some or all of the functions of each of the above control blocks can be realized by a logic circuit. For example, an integrated circuit in which a logic circuit that functions as each of the above control blocks is formed is also included in the scope of the present invention. In addition, it is also possible to realize the functions of each of the above control blocks by, for example, a quantum computer.

[Example]
Specific examples of the present invention will be described below with reference to Figs. 7 to 9. Fig. 7 is a diagram showing an example of the results of an example. Fig. 8 is a diagram showing another example of the results of an example. Fig. 9 is a table summarizing the examples. Note that the present invention is not limited to the following examples.

In this embodiment, the following three machine learning models were used to predict the water storage volume after three days, and the models were evaluated by calculating the difference from the observed value using RMSE (Root Mean Squared Error).
A base model with SV, InOut, and Rsum as inputs. In other words, none of the variables that affect the fluctuation in the water volume of the reservoir from the current time to the date and time after a predetermined time has elapsed, LTOO_Rsum, LTOO_Rsum, LTOO_Ssum, LTOO_Ssum, LTOO_EV, and LTOO_InOut, are incorporated into the input variables of the model ("base model" in Figures 7 to 9).
A machine learning model that uses SV, LT03_EV0.5 to 3, and LT03_InOut0.5 to 3 as input ("Improved Model 1" in Figures 7 and 9, for example, "Model 4" in Figure 6).
A machine learning model generated by machine learning using the stacking ensemble method, using SV, LT03_Ssum, LT03_Rsum, LT03_EV0.5-3, and LT03_InOut0.5-3 as input ("Improved Model 2" in Figures 8 and 9).

As an improvement effect of the target improved model to be compared from the base model, the improvement rate of the target improved model was calculated using the following formula (5).
Improvement rate = 100 × (RMSE of base model - RMSE of target improved model) / RMSE of base model ... (5)
As shown in Figure 7, in the improved model 1, the improvement rate of the prediction results in the irrigation season was 42.6%, in the improved model 1, the improvement rate of the prediction results in the non-irrigation season was 24.9%, and in the improved model 1, the improvement rate of the prediction results for the year was 40.4%.

As shown in Figure 8, in the improved model 2, the improvement rate of the prediction results during the irrigation season was 42.0%. In the improved model 2, the improvement rate of the prediction results during the non-irrigation season was 49.0%. In the improved model 2, the improvement rate of the prediction results for the year was 41.7%.

As shown in Figures 7 to 9, both Improved Model 1 and Improved Model 2 showed improved prediction accuracy compared to the base model in both the irrigation and non-irrigation seasons and throughout the year.

[summary]
A prediction method according to aspect 1 of this embodiment is a prediction method for predicting the amount of water stored in a reservoir, and includes an acquisition step of acquiring information indicating variables that will affect fluctuations in the amount of water stored in the reservoir from the current time to a date and time after a predetermined time has elapsed, and a prediction step of predicting the amount of water stored in the reservoir at a date and time after the predetermined time has elapsed by inputting the information acquired in the acquisition step into a trained model that is trained to use the information indicating the variables that will affect fluctuations in the amount of water stored in the reservoir as input information and output a prediction result of the amount of water stored in the reservoir at the date and time after the predetermined time has elapsed.

The above configuration makes it possible to predict the amount of water stored in a reservoir several days in advance.

In the prediction method according to aspect 2 of this embodiment, in the above aspect 1, the information acquired in the acquisition step includes a variable indicating an integrated value of the amount of precipitation in the reservoir from a predetermined time to a date and time after the predetermined time has elapsed.

The above configuration makes it possible to effectively predict the amount of water stored in a reservoir several days in advance.

In the prediction method according to aspect 3 of this embodiment, in aspect 1 or 2 above, the information acquired in the acquisition step includes a variable indicating the amount of precipitation for each predetermined interval from a predetermined time to a date and time after the predetermined time has elapsed.

The above configuration makes it possible to effectively predict the amount of water stored in a reservoir several days in advance.

In the prediction method according to aspect 4 of this embodiment, in any one of aspects 1 to 3 above, the information acquired in the acquisition step further includes a variable indicating the amount of water stored in the reservoir a predetermined time before the current time, and a variable indicating the difference between the amount of water stored at the current time and the amount of water stored a predetermined time before the current time.

The above configuration makes it possible to effectively predict the amount of water stored in a reservoir several days in advance.

In the prediction method according to aspect 5 of this embodiment, in any one of aspects 1 to 4 above, the information acquired in the acquisition step further includes variables indicating the maximum and minimum water volume in the reservoir from a predetermined time before the current time to the current time.

The above configuration makes it possible to effectively predict the amount of water stored in a reservoir several days in advance.

In the prediction method according to aspect 6 of this embodiment, in any one of aspects 1 to 5 above, the information acquired in the acquisition step further includes a variable indicating an accumulated value of the amount of precipitation in the reservoir from a predetermined time before the current time to the current time.

The above configuration makes it possible to effectively predict the amount of water stored in a reservoir several days in advance.

In the prediction method according to aspect 7 of this embodiment, the trained model in any one of aspects 1 to 6 above is a trained model trained by a stacking ensemble method.

The above configuration allows for more accurate prediction of the amount of water stored in a reservoir several days in advance.

The prediction method according to aspect 8 of this embodiment is any one of aspects 1 to 7 above, in which the information acquired in the acquisition step includes a variable indicating an integrated value of the discharge amount in the reservoir from a predetermined time to a date and time after the predetermined time has elapsed.

The above configuration makes it possible to effectively predict the amount of water stored in a reservoir several days in advance.

The prediction method according to aspect 9 of this embodiment is any one of aspects 1 to 8 above, in which the information acquired in the acquisition step includes a variable indicating an integrated value of the discharge amount at each predetermined interval from a predetermined time to a date and time after the predetermined time has elapsed in the reservoir.

The above configuration makes it possible to effectively predict the amount of water stored in a reservoir several days in advance.

The prediction method according to aspect 10 of this embodiment is any one of aspects 1 to 9 above, in which the information acquired in the acquisition step includes a variable indicating a tentative amount of water stored in the reservoir after the predetermined time has elapsed from the current time.

The above configuration makes it possible to effectively predict the amount of water stored in a reservoir several days in advance.

The prediction method according to aspect 11 of this embodiment is any one of aspects 1 to 10 above, in which the information acquired in the acquisition step includes a variable indicating the difference between the amount of water stored in the reservoir at the current time and the provisional amount of water stored after the predetermined time has elapsed from the current time.

The above configuration makes it possible to effectively predict the amount of water stored in a reservoir several days in advance.

The prediction device according to aspect 12 of this embodiment is a prediction device that predicts the amount of water stored in a reservoir, and includes an acquisition unit that acquires information indicating variables that affect fluctuations in the amount of water stored in the reservoir from the current time to a date and time after a predetermined time has elapsed, and a prediction unit that predicts the amount of water stored in the reservoir at a date and time after the predetermined time has elapsed by inputting the information acquired by the acquisition unit into a trained model that is trained to use the information indicating the variables that affect fluctuations in the amount of water stored in the reservoir as input information and output a prediction result of the amount of water stored in the reservoir at a date and time after the predetermined time has elapsed.

The above configuration provides the same effect as in aspect 1.

The prediction program according to aspect 13 of this embodiment is a prediction program for causing a computer to function as the prediction device according to aspect 12 above, and causes the computer to function as each of the above parts.

The above configuration provides the same effect as in aspect 1.

The storage medium according to aspect 14 of this embodiment is computer-readable and stores the prediction program according to aspect 13 above.

The above configuration provides the same effect as in aspect 1.

[Additional Notes]
The present invention is not limited to the above-described embodiments, and various modifications are possible within the scope of the claims. Embodiments obtained by appropriately combining the technical means disclosed in different embodiments are also included in the technical scope of the present invention.

REFERENCE SIGNS LIST 100, 100a Prediction system 1, 1a Prediction device 2 Sensor 3 Precipitation database 4 User terminal 5 Network 111 Acquisition unit 112 Prediction unit 131 Precipitation data 132 Water storage data 133 Prediction result S1 Prediction method S11 Acquisition process S12 Prediction process

Claims (14)

A method for predicting a water volume in a reservoir, comprising the steps of:
An acquisition step of acquiring information indicating a variable that affects a fluctuation in the water volume in the reservoir from a current time to a date and time after a predetermined time has elapsed;
a prediction process for predicting the amount of water stored in the reservoir at a date and time after the predetermined time has elapsed by inputting the information acquired in the acquisition process into a trained model trained to output a prediction result of the amount of water stored in the reservoir at a date and time after the predetermined time has elapsed, using information indicating variables that affect fluctuations in the amount of water stored in the reservoir as input information;
A method for predicting a prediction result. The prediction method according to claim 1 , wherein the information acquired in the acquisition step includes a variable indicating an integrated value of the amount of precipitation in the reservoir from a predetermined time to a date and time after the predetermined time has elapsed. The prediction method according to claim 1 , wherein the information acquired in the acquiring step includes a variable indicating the amount of precipitation for each predetermined interval from a predetermined time to a date and time after the predetermined time has elapsed. The prediction method according to claim 2 or 3, wherein the information acquired in the acquisition step further includes a variable indicating the amount of water stored in the reservoir a predetermined time before the current time, and a variable indicating the difference between the amount of water stored at the current time and the amount of water stored a predetermined time before the current time. The prediction method according to claim 2 or 3, wherein the information acquired in the acquisition step further includes a variable indicating a maximum value and a minimum value of the water volume in the reservoir from a predetermined time before the current time to the current time. The prediction method according to claim 2 or 3, wherein the information acquired in the acquiring step further includes a variable indicating an integrated value of the amount of precipitation in the reservoir from a predetermined time before the current time to the current time. The trained model is a trained model trained by a stacking ensemble method.
The method of claim 1 . The prediction method according to claim 1 or 7, characterized in that the information acquired in the acquisition process includes a variable indicating an integrated value of the discharge amount in the reservoir from a specified time to a date and time after the specified time has elapsed. The prediction method described in claim 1 or 7, characterized in that the information acquired in the acquisition process includes a variable indicating the accumulated value of the discharge amount in the reservoir at each specified interval from a specified time to a date and time after the specified time has elapsed. The prediction method according to claim 1 or 7, characterized in that the information acquired in the acquisition step includes a variable indicating a provisional amount of water stored in the reservoir after the predetermined time has elapsed from the current time. The prediction method according to claim 1 or 7, characterized in that the information acquired in the acquisition step includes a variable indicating the difference between the amount of water stored in the reservoir at the current time and the provisional amount of water stored after the specified time has elapsed from the current time. A prediction device for predicting a water volume in a reservoir, comprising:
An acquisition unit that acquires information indicating a variable that affects a fluctuation in the water volume in the reservoir from a current time to a date and time after a predetermined time has elapsed;
a prediction unit that predicts the amount of water stored in the reservoir at a date and time after the predetermined time has elapsed by inputting the information acquired by the acquisition unit into a trained model that is trained to output a prediction result of the amount of water stored in the reservoir at a date and time after the predetermined time has elapsed, using information indicating a variable that affects fluctuations in the amount of water stored in the reservoir as input information;
A prediction device comprising: A prediction program for causing a computer to function as the prediction device according to claim 12, the prediction program causing a computer to function as the acquisition unit and the prediction unit. A computer-readable recording medium having the prediction program according to claim 13 recorded thereon.

Images