Water level prediction hit ratio calculation method and water level prediction hit ratio calculation program

Abstract

To visualize reliability of water level prediction data to support achievement determination to each issuing reference even when extracting "prediction" precipitation data as test data in predicting a water level of a river or the like with machine learning.SOLUTION: An evaluation stage includes: extracting, as evaluation data, water level data, "past observation" precipitation data and "future observation" precipitation data, calculating a probability distribution of a prediction error of a water level and setting an error width of prediction data of the water level relative to the observation data of the water level (S3); and then, as the evaluation data, replacing the "future observation" precipitation data with the "prediction" precipitation data, and calculating, as a prediction hit ratio of a water level, a proportion of the data within the error width around the observation data of the water level of the prediction data of the water level (S5). A prediction stage includes extracting, as test data, the water level data, the "past observation" precipitation data and the "future observation" precipitation data for practical operation, and outputting the prediction hit ratio of the water level together with the prediction data of the water level (S6).SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present disclosure relates to a technology for predicting water levels of rivers and the like using machine learning. [Background technology]

[0002] Patent Document 1 and other publications disclose technology for predicting the water level of rivers and the like using machine learning. In the training stage, a water level prediction model such as a neural network is constructed using training data for water level learning. In the verification stage, verification data for verifying the water level prediction model is input into the water level prediction model, and the generalization performance of the water level prediction model is evaluated. In the prediction stage, test data for water level prediction is input into the water level prediction model, and the water level of rivers and the like is predicted. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2019-095240 Summary of the Invention [Problem to be solved by the invention]

[0004] In actual operation, test data required to predict water levels at a certain time at a water level prediction point includes water level data at that time and at least one of the water level prediction point and a point upstream from the water level prediction point, "past observation" rainfall data for that time and / or in the past, and "forecast" rainfall data for the future from that time. However, because "forecast" rainfall data is less accurate than "future observation" rainfall data, the "forecast" water level data can deviate significantly from the "observed" water level data.

[0005] Therefore, in order to solve the above-mentioned problems, the present disclosure aims to visualize the reliability of water level prediction data and assist in determining whether each issuance standard has been reached when predicting the water level of a river, etc. using machine learning, even when extracting "forecast" rainfall data as test data. [Means for solving the problem]

[0006] To solve the above problem, in the verification stage, water level data, "past observation" rainfall data, and "future observation" rainfall data are extracted as verification data, the probability distribution of water level prediction errors is calculated, and the error range of the water level prediction data relative to the water level observation data is set. Then, as verification data, the "future observation" rainfall data is replaced with the "forecast" rainfall data, and the proportion of the water level prediction data that falls within the error range around the water level observation data is calculated as the water level prediction accuracy rate.

[0007] In the prediction stage, test data is extracted from water level data, "past observation" rainfall data, and "forecast" rainfall data in actual operation, and the water level prediction accuracy rate is output along with the water level prediction data. Here, the water level prediction accuracy rate is output for each issuance standard and / or for each N-hour prediction.

[0008] Specifically, the present disclosure provides a water level prediction accuracy calculation method comprising, in order, a verification data extraction step of extracting, as verification data for verifying a water level prediction model for predicting the water level at a certain time at a water level prediction point, water level data and past observed rainfall data at the water level prediction point and / or at a point upstream of the water level prediction point at that time, and forecast rainfall data for the future from that time; a water level prediction data output step of inputting, as the verification data, the water level data, the past observed rainfall data, and the forecast rainfall data for verifying the water levels at the water level prediction point at multiple times, and outputting forecast data of the water level at the water level prediction point at those multiple times; and a water level prediction accuracy calculation step of calculating, as the prediction accuracy rate of the water level at the water level prediction point, the proportion of the forecast data of the water level at the water level prediction point at those multiple times that falls within an error band around the observed water level data at the water level prediction point at those multiple times.

[0009] According to this configuration, when using machine learning to predict the water levels of rivers, etc., even when "forecast" rainfall data is extracted as test data, the accuracy rate of water level predictions can be calculated during the verification stage, making it possible to visualize the reliability of the water level prediction data.

[0010] The present disclosure also provides a water level prediction accuracy calculation method, characterized in that the verification data extraction step also extracts future observed rainfall data from the time in question at at least one of the water level prediction point and a point upstream of the water level prediction point as the verification data, inputs the water level data, the past observed rainfall data, and the future observed rainfall data to the water level prediction model as the verification data, calculates a probability distribution of the water level prediction error at the water level prediction point, and sets an error range for the water level prediction data relative to the water level observation data in the water level prediction accuracy calculation step, between the verification data extraction step and the water level prediction data output step.

[0011] According to this configuration, when predicting the water level of a river or the like using machine learning, the probability distribution of the water level prediction error can be calculated in the verification stage, the error range of the water level prediction data relative to the water level observation data can be set, and the accuracy rate of the water level prediction can be calculated.

[0012] The present disclosure also provides a water level prediction accuracy calculation method, characterized in that the water level prediction accuracy calculation step calculates the prediction accuracy of the water level at the water level prediction point for each issuance criterion for water level alert at the water level prediction point and / or for each period length from the prediction output time in the water level prediction data output step to each prediction target time.

[0013] According to this configuration, when predicting the water level of a river or other body of water using machine learning, even when "forecast" rainfall data is extracted as test data, the accuracy rate of the water level prediction can be calculated during the verification stage to assist in determining whether each issuance standard will be reached.

[0014] The present disclosure also provides a water level prediction accuracy calculation method, characterized in that the water level prediction accuracy calculation step outputs the prediction accuracy of the water level at the water level prediction point together with the predicted data of the water level at the target time of prediction at the water level prediction point based on the water level data, the past observed rainfall data, and the forecast rainfall data.

[0015] According to this configuration, when using machine learning to predict the water level of a river, etc., even when "forecast" rainfall data is extracted as test data, the accuracy rate of the water level prediction can be output at the prediction stage, making it possible to visualize the reliability of the water level prediction data.

[0016] The present disclosure also provides a water level prediction accuracy calculation program for causing a computer to sequentially execute the processing steps included in the water level prediction accuracy calculation method described above.

[0017] According to this configuration, it is possible to provide a program having the above-described effects.

[0018] The above-disclosed inventions can be combined as much as possible. [Effects of the Invention]

[0019] In this way, when predicting the water levels of rivers, etc. using machine learning, the present disclosure can visualize the reliability of water level prediction data and assist in determining whether each issuance standard has been reached, even when extracting "forecast" rainfall data as test data. [Brief explanation of the drawings]

[0020] [Figure 1] FIG. 1 is a diagram illustrating a configuration of a water level prediction device according to the present disclosure. [Figure 2] FIG. 10 is a diagram showing the procedure of the water level prediction accuracy calculation process of the present disclosure. [Figure 3] FIG. 10 is a diagram showing a specific example of the water level prediction error range setting process of the present disclosure. [Figure 4] FIG. 10 is a diagram showing a specific example of the water level prediction accuracy calculation process of the present disclosure. [Figure 5] FIG. 10 is a diagram showing an example of the water level prediction accuracy rate output process of the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0021]

[0023] The following embodiments of the present disclosure will be described with reference to the accompanying drawings. The embodiments described below are examples of implementation of the present disclosure, and the present disclosure is not limited to the following embodiments.

[0022] (Configuration of the water level prediction device of the present disclosure) The configuration of the water level prediction device of the present disclosure is shown in Figure 1. The procedure for the water level prediction accuracy calculation process of the present disclosure is shown in Figure 2. The water level prediction device W comprises a water level prediction model M, a training data extraction unit 1, a water level prediction model construction unit 2, a verification data extraction unit 3, a water level prediction error range setting unit 4, a water level prediction accuracy calculation unit 5, a test data extraction unit 6, and a water level prediction data output unit 7. The verification data extraction unit 3, the water level prediction error range setting unit 4, and the water level prediction accuracy calculation unit 5 can be realized by installing the water level prediction accuracy calculation program shown in Figure 2 on a computer.

[0023] The training data extraction unit 1 extracts water level data WL at the water level prediction point and / or at least any of the water level prediction point and a point upstream of the water level prediction point at that time, and past observed rainfall data RO, as well as future observed rainfall data RO from that time, as training data L required for learning the water level at the water level prediction point. Here, the training data extraction unit 1 may extract water level data WL at the time of water level fluctuation, etc., past observed rainfall data RO, and future observed rainfall data RO, or may extract water level data WL at times when the water level does not fluctuate, such as immediately before and after a water level fluctuation, past observed rainfall data RO, and future observed rainfall data RO. The water level prediction model construction unit 2 constructs a water level prediction model M using the training data L.

[0024] The test data extraction unit 6 extracts water level data WL at or before the water level prediction point and past observed rainfall data RO, as well as future forecast rainfall data RF from at least one of the water level prediction point and a point upstream of the water level prediction point, as test data T required for predicting the water level at a certain time at the water level prediction point. Here, in actual operation, the test data extraction unit 6 extracts forecast rainfall data RF without extracting future observed rainfall data RO. The water level prediction data output unit 7 outputs water level prediction data P using the water level prediction model M.

[0025] However, since the "forecast" rainfall data RF is less accurate than the "future observation" rainfall data RO, the "prediction" water level data P may deviate significantly from the "observation" water level data. Therefore, this disclosure visualizes the reliability of the "prediction" water level data P.

[0026] In the verification stage, the water level data WL, the "past observation" rainfall data RO, and the "future observation" rainfall data RO are extracted as verification data V, the water level prediction error probability distribution E is calculated, and the error band of the water level prediction data P relative to the water level observation data is set. Then, as verification data V, the "future observation" rainfall data RO is replaced with the "forecast" rainfall data RF, and the proportion of the water level prediction data P that falls within the error band around the water level observation data is calculated as the water level prediction accuracy rate H.

[0027] In the prediction stage, in actual operation, water level data WL, "past observation" rainfall data RO, and "forecast" rainfall data RF are extracted as test data T, and the water level prediction accuracy rate H is output along with the water level prediction data P. Here, the water level prediction accuracy rate H is output for each issuance standard and / or for each prediction N hours later.

[0028] Here, the training data L is known data for water level learning, and since it makes the water level prediction accuracy rate H nearly 100%, it cannot be applied to calculate the water level prediction accuracy rate H. The test data T is unknown data for water level prediction, but since it is data specialized for a specific rainfall situation, it cannot be applied to calculate the water level prediction accuracy rate H. On the other hand, the verification data V is unknown data for verifying the water level prediction model M, and since it is data not specialized for a specific rainfall situation, it can be applied to calculate the water level prediction accuracy rate H.

[0029] (Specific example of water level prediction error range setting process disclosed herein) A specific example of the water level prediction error range setting process of the present disclosure is shown in Fig. 3. The verification data extraction unit 3 extracts water level data WL and past observed rainfall data RO at at least one of the water level prediction point and a point upstream of the water level prediction point at a certain time as verification data V for verifying the water level prediction model M for predicting the water level at the water level prediction point at a certain time (step S1). Here, the verification data extraction unit 3 may extract water level data WL, past observed rainfall data RO, and future observed rainfall data RO when the water level fluctuates, etc., or may extract water level data WL, past observed rainfall data RO, and future observed rainfall data RO when the water level does not fluctuate, etc., immediately before and after the water level fluctuates, etc. In the water level prediction error range setting process, forecast rainfall data RF is not extracted.

[0030] The verification data extraction unit 3 inputs water level data WL, past observed rainfall data RO, and future observed rainfall data RO to the water level prediction model M as verification data V to verify water levels at multiple times at the water level prediction point (step S2).The water level prediction error range setting unit 4 then calculates the prediction error probability distribution E of the water level at the water level prediction point and sets the error range of the water level prediction data P for the water level observation data in the water level prediction accuracy calculation step S5 (step S3).Here, the water level prediction error range setting unit 4 calculates the prediction error probability distribution E for each issuance standard for water level alert at the water level prediction point and / or for each period length from the prediction output time in the water level prediction data output step S2 to each prediction target time, and sets the error range of the water level prediction data P (step S3).

[0031] In the "Prediction error after each N hours" column in Figure 3, the prediction error probability distribution E of the water level at the water level prediction point is calculated for the observed water level of 0 m or more and less than the announcement standard water level A, the observed water level of A or more and less than the announcement standard water level B, and the observed water level of B or more. In the "Error range after each N hours" column in Figure 3, the e of the prediction error probability distribution E is calculated for the observed water level of 0 m or more and less than the announcement standard water level A, the observed water level of A or more and less than the announcement standard water level B, and the observed water level of B or more.lоw-min That's all lоw-max The following and e mid-min That's all mid-max The following and e high-min That's all high-max The error range of the water level forecast data P for the water level observation data is set so that the following is within the error range. Here, the minimum and maximum values ​​of the forecast error probability distribution E are set to 0% and 100%, respectively, and 0.1%≦e high-min ≦e mid-min ≦e lоw-min ≦20% and 80%≦e lоw-max ≦e mid-max ≦e high-max ≦99.9% is satisfied. In other words, the error band of the water level prediction data P is set wider for higher water levels, narrower for lower water levels, and intermediate for medium water levels. The upper and lower fluctuation bands of the error band of the water level prediction data P may be equal or different depending on the distribution shape of the prediction error probability distribution E.

[0032] In addition, e high-min , e mid-min , e lоw-min , e lоw-max , e mid-max , e high-max is tentatively set as above, but is finally set as follows. First, water level data and analyzed rainfall data (radar interpolated with rain gauge) are input as test data into the water level prediction model M. Next, for each of the high water level, low water level, and medium water level, e is set so that 80% or more of the water level prediction data falls within the error range of the water level prediction data. high-min , e mid-min , e lоw-min , e lоw-max , e mid-max , e high-max is set.

[0033] In this way, when predicting the water level of a river or other body of water using machine learning, the water level prediction error probability distribution E can be calculated in the verification stage, the error range of the water level prediction data P relative to the water level observation data can be set, and preparations can be made to calculate the water level prediction accuracy rate H.

[0034] Furthermore, when using machine learning to predict water levels of rivers, etc., even when "forecast" rainfall data RF is extracted as test data T, by preparing to calculate the water level prediction accuracy rate H during the verification stage, it is possible to prepare to support the determination of whether each issuance standard has been reached.

[0035] (Specific example of water level prediction accuracy calculation process disclosed herein) A specific example of the water level prediction accuracy calculation process of the present disclosure is shown in Fig. 4. The verification data extraction unit 3 extracts water level data WL and past observed rainfall data RO for at least one of the water level prediction point and a point upstream of the water level prediction point at a certain time, as verification data V for verifying the water level prediction model M for predicting the water level at the water level prediction point at a certain time (step S1). Here, the verification data extraction unit 3 may extract water level data WL, past observed rainfall data RO, and forecast rainfall data RF for times when the water level fluctuates, etc., or may extract water level data WL, past observed rainfall data RO, and forecast rainfall data RF for times when the water level does not fluctuate, etc., immediately before and after a water level fluctuation. In the water level prediction accuracy calculation process, future observed rainfall data RO is not extracted.

[0036] The verification data extraction unit 3 inputs water level data WL, past observed rainfall data RO, and forecast rainfall data RF to the water level prediction model M as verification data V to verify water levels at multiple times at the water level prediction point (step S4). Then, the water level prediction model M outputs water level prediction data P at the water level prediction point for the multiple times. Furthermore, the water level prediction accuracy calculation unit 5 calculates the proportion of water level prediction data P at the water level prediction point for the multiple times that falls within the error band around the water level observation data at the water level prediction point for the multiple times as the water level prediction accuracy H at the water level prediction point (step S5). Here, the water level prediction accuracy calculation unit 5 calculates the water level prediction accuracy H for each water level alert issuance criterion at the water level prediction point and / or for each period length from the prediction output time in the water level prediction data output step S4 to each prediction target time (step S5).

[0037] In the "Predicted data for each N hours later" column in Figure 4, water level prediction data P for the water level prediction point at multiple times is output for observed water levels of 0 m or more and less than the warning standard water level A, observed water levels of A or more and less than the warning standard water level B, and observed water levels of B or more. For observed water levels of 0 m or more and less than the warning standard water level A, there are 80 pieces of data that fall within the error range, 10 pieces of data that fall below the error range, and 10 pieces of data that exceed the error range. For observed water levels of A or more and less than the warning standard water level B, there are 30 pieces of data that fall within the error range, 15 pieces of data that fall below the error range, and 15 pieces of data that exceed the error range. For observed water levels of B or more, there are 10 pieces of data that fall within the error range, 10 pieces of data that fall below the error range, and 10 pieces of data that exceed the error range.

[0038] In the "Prediction accuracy rate N hours later" column in Figure 4, the water level prediction accuracy rate H at the water level prediction point is calculated for observed water levels of 0 m or more and less than the warning standard water level A, observed water levels of A or more and less than the warning standard water level B, and observed water levels of B or more. For observed water levels of 0 m or more and less than the warning standard water level A, the water level prediction accuracy rate is 80 / (80+10+10)×100 = 80%. For observed water levels of A or more and less than the warning standard water level B, the water level prediction accuracy rate is 30 / (30+15+15)×100 = 50%. For observed water levels of B or more, the water level prediction accuracy rate is 10 / (10+10+10)×100 = 33%.

[0039] Even if test data T is accumulated, the water level prediction model M is not updated, so there is no need to update the water level prediction accuracy rate H. Furthermore, when training data L is accumulated, the water level prediction model M may be updated, and the water level prediction accuracy rate H may also be updated.

[0040] In this way, when using machine learning to predict the water level of a river, etc., even when "forecast" rainfall data RF is extracted as test data T, the reliability of the water level prediction data P can be visualized by calculating the water level prediction accuracy rate H in the verification stage.

[0041] Furthermore, when using machine learning to predict water levels of rivers, etc., even when "forecast" rainfall data RF is extracted as test data T, the accuracy rate H of water level predictions can be calculated in the verification stage to assist in determining whether each issuance standard has been reached.

[0042] (Example of water level prediction accuracy output processing of the present disclosure) An example of the water level prediction accuracy output process of the present disclosure is shown in Fig. 5. When predicting the water level at a water level prediction point at a target prediction time based on the water level data WL, past observed rainfall data RO, and forecast rainfall data RF, the water level prediction accuracy calculation unit 5 outputs the water level prediction accuracy rate H at the water level prediction point along with the water level prediction data P at the target prediction time at the water level prediction point (step S6). Here, the water level prediction accuracy calculation unit 5 outputs the water level prediction accuracy rate H for each issuance criterion for a water level alert at the water level prediction point and / or for each period length from the prediction output time to each target prediction time (step S6).

[0043] In Figure 5, L0, L1, L2, L3, L4, and L5 are set as the water level warning reference levels. In the verification stage, the accuracy rate of water level prediction is H for the observed water levels of warning reference levels L0-L1, L1-L2, L2-L3, L3-L4, and L4-L5. 01 %, H 12 %, H 23 %, H 34 %, H 45 At the forecast stage, the accuracy rate of the water level forecast for around 12:00 on the first day, around 14:00 on the second day, around 8:00 on the third day, and around 12:00 on the fourth day (forecast for 6 hours later) is calculated as H 01 %(hit), H 23 % (miss), H 45 %(hit), H 01 The output will be % (hit).

[0044] In this way, when using machine learning to predict the water level of a river, etc., even when "forecast" rainfall data RF is extracted as test data T, the reliability of the water level prediction data P can be visualized by outputting the water level prediction accuracy rate H at the prediction stage.

[0045] Furthermore, when using machine learning to predict the water levels of rivers, etc., even when "forecast" rainfall data RF is extracted as test data T, the water level prediction accuracy rate H can be output at the prediction stage to assist in determining whether each issuance standard will be reached. [Industrial Applicability]

[0046] The water level prediction accuracy calculation method and water level prediction accuracy calculation program disclosed herein use machine learning to predict the water level of a river, etc., and even when "forecast" rainfall data is extracted as test data, by calculating the water level prediction accuracy rate in the verification stage, the reliability of the water level prediction data can be visualized and judgments about whether each issuance standard will be reached can be supported. [Explanation of symbols]

[0047] W: Water level prediction device M: Water level prediction model 1: Training data extraction part 2: Water level prediction model construction section 3: Verification data extraction section 4: Water level prediction error range setting section 5: Water level prediction accuracy calculation section 6: Test data extraction part 7: Water level forecast data output section WL: Water level data RO: Observed rainfall data RF: Forecast rainfall data L: training data V: Verification data T: Test data E: Forecast error probability distribution P: Water level forecast data H: Water level prediction accuracy rate

Claims

1. a verification data extraction step of extracting, as verification data for verifying a water level prediction model for predicting the water level at a water level prediction point at a certain time, water level data and past observed rainfall data at the water level prediction point and / or a point upstream of the water level prediction point at the certain time and / or in the past, and forecast rainfall data for the future from the certain time; a water level prediction data output step of inputting the water level data, the past observed rainfall data, and the forecast rainfall data for verifying the water levels at the water level prediction point at multiple times as the verification data into the water level prediction model, and outputting predicted data of the water levels at the water level prediction point at the multiple times; a water level prediction accuracy calculation step of calculating the rate of prediction data of water levels at the water level prediction point at the multiple times that fall within an error range around the observed data of water levels at the water level prediction point at the multiple times as the prediction accuracy rate of the water level at the water level prediction point; A water level prediction accuracy calculation method characterized by comprising the steps of:

2. the verification data extraction step also extracts future observed rainfall data from the time point at least one of the water level prediction point and a point upstream of the water level prediction point as the verification data; a water level prediction error range setting step of inputting the water level data, the past observed rainfall data, and the future observed rainfall data for verifying the water levels at the water level prediction point at the multiple times as the verification data into the water level prediction model, calculating a probability distribution of the prediction error of the water level at the water level prediction point, and setting an error range of the water level prediction data relative to the water level observation data in the water level prediction accuracy calculation step; between the verification data extraction step and the water level prediction data output step. The water level prediction accuracy calculation method according to claim 1 .

3. The water level prediction accuracy calculation step calculates a prediction accuracy rate of the water level at the water level prediction point for each issuance criterion for water level alert at the water level prediction point and / or for each period length from the prediction output time in the water level prediction data output step to each prediction target time.

3. The water level prediction accuracy calculation method according to claim 1 or 2.

4. The water level prediction accuracy calculation step outputs a prediction accuracy rate of the water level at the water level prediction point together with prediction data of the water level at the water level prediction point at the prediction target time when predicting the water level at the water level prediction point at the prediction target time based on the water level data, the past observed rainfall data, and the forecast rainfall data.

3. The water level prediction accuracy calculation method according to claim 1 or 2.

5. A water level prediction accuracy rate calculation program for causing a computer to sequentially execute each processing step of the water level prediction accuracy rate calculation method according to claim 1 or 2.

Images