Water level prediction model construction method, water level prediction model construction program, water level prediction data output method, water level prediction data output program, test data interpolation method, test data interpolation program, water level prediction hitting rate calculation method and water level prediction hitting rate calculation program

Abstract

To properly extract training data and test data so that, upon predicting a water level of a backwater section of a branch using machine learning, precision in learning of a water level of the backwater section of the branch is higher and predicting the water level of the backwater section of the branch can be appropriately executed.SOLUTION: The present disclosure is a water level prediction model construction method comprises: a training data extraction step of extracting, as training data L required for learning of a water level at certain time at a water level prediction point of a backwater section of a branch, (1) branch upstream data indicative of a past rainfall and water level with respect to the time, at a water level prediction point, and at an upstream point of a branch relative to the water level prediction point, and (2) mainstream upstream data indicative of a past rainfall and water level ahead of the time, at a confluence point of the branch and a mainstream, and at an upstream point of the mainstream relative to the confluence point; and a water level prediction model construction step of constructing a water level prediction model M for predicting a water level at the water level prediction point, using the branch upstream data and the mainstream upstream data.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present disclosure relates to a technique for predicting water levels in the backwater reaches of a tributary using machine learning. [Background technology]

[0002] A technology for predicting the water level of a river using machine learning is disclosed in Patent Document 1 and the like. 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 to 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 to the water level prediction model, and the water level of the river is predicted.

[0003] In Patent Document 1, upstream data indicating at least one of the rainfall and water level in the past from the water level prediction point and / or a point upstream from the water level prediction point is extracted as training data required for learning the water level at a certain time at the water level prediction point. Then, upstream data indicating at least one of the rainfall and water level in the past from the water level prediction point and / or a point upstream from the water level prediction point is extracted as test data required for predicting the water level at a certain time at the water level prediction point. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2019-095240 A Summary of the Invention [Problem to be solved by the invention]

[0005] During floods, when the water level of the main stream is high, it becomes difficult for the water in the tributary to flow downstream, and the water in the main stream may flow back into the tributary, causing the water level in the backwater section of the tributary to rise. In other words, the water level in the backwater section of a tributary is not only affected by the amount of rainfall and water level upstream of the tributary from the backwater section, but also by the amount of rainfall and water level upstream of the main stream from the confluence of the tributary and the main stream, and may also be affected by the amount of rainfall and water level downstream of the confluence of the tributary and the main stream.

[0006] The extraction process of training data and test data in the conventional technology is shown in Figure 1. In the conventional technology, as training data required for learning the water level at a certain time at the water level prediction point P1 in the backwater section of the tributary, upstream tributary data showing at least one of the rainfall and water level in the past from the relevant time at the water level prediction point P1 and at least one of the upstream points P2 and P3 of the tributary from the water level prediction point P1 is extracted. Then, as test data required for predicting the water level at a certain time at the water level prediction point P1 in the backwater section of the tributary, upstream tributary data showing at least one of the rainfall and water level in the past from the relevant time at the water level prediction point P1 and at least one of the upstream points P2 and P3 of the tributary from the water level prediction point P1 is extracted.

[0007] However, with regard to the water level in the backwater section of a tributary, only the influence of rainfall and water level upstream of the tributary from the backwater section is taken into consideration, and the influence of rainfall and water level upstream of the main stream from the confluence of the tributary with the main stream is not taken into consideration, nor is the influence of rainfall and water level downstream of the confluence of the tributary with the main stream taken into consideration. As a result, the accuracy of learning the water level in the backwater section of the tributary is low, and it is not possible to properly predict the water level in the backwater section of the tributary.

[0008] Therefore, in order to solve the above problem, the present disclosure aims to appropriately extract training data and test data when predicting the water level in the backwater section of a tributary using machine learning, so that the accuracy of learning the water level in the backwater section of a tributary is improved and prediction of the water level in the backwater section of a tributary can be performed appropriately. [Means for solving the problem]

[0009] In order to solve the above problem, the water level in the backwater section of a tributary is not only influenced by the amount of rainfall and water level upstream of the tributary from the backwater section, but also by the amount of rainfall and water level upstream of the main stream from the confluence of the tributary with the main stream as a first influence, and by the amount of rainfall and water level downstream of the confluence of the tributary with the main stream as a second influence.

[0010] Specifically, the present disclosure is a water level prediction model construction method comprising: a training data extraction step of extracting, as training data necessary for learning the water level at a certain time at a water level prediction point in a backwater section of a tributary, (1) tributary upstream data indicating at least one of the rainfall and water level in the past from the time at the water level prediction point and at least one of a point upstream of the tributary from the water level prediction point, and (2) main stream upstream data indicating at least one of the rainfall and water level in the past from the time at the confluence of the tributary and the main stream and at least one of a point upstream of the main stream from the confluence; and a water level prediction model construction step of constructing a water level prediction model for predicting the water level at the water level prediction point using the tributary upstream data and the main stream upstream data.

[0011] According to this configuration, data on the upstream of the tributary and the upstream of the main stream can be appropriately extracted as training data so as to increase the accuracy of learning the water level in the backwater section of the tributary.

[0012] The present disclosure also relates to a water level prediction model construction method, wherein the training data extraction step extracts, as the training data, in addition to the tributary upstream data and the mainstream upstream data, (3) mainstream downstream data indicating at least one of past rainfall and water level from a point downstream of the mainstream from the confluence, and the water level prediction model construction step constructs the water level prediction model using the tributary upstream data, the mainstream upstream data, and the mainstream downstream data.

[0013] According to this configuration, data on the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream can be appropriately extracted as training data so as to increase the accuracy of learning the water level in the backwater section of the tributary.

[0014] The present disclosure also relates to a water level prediction model construction program for causing a computer to sequentially execute each processing step of the water level prediction model construction method described above.

[0015] According to this configuration, it is possible to provide a program having the above-described effects.

[0016] The present disclosure also relates to a water level prediction data output method, comprising a water level prediction data output step of inputting the tributary upstream data and the main stream upstream data into the water level prediction model constructed by the water level prediction model construction method described above as test data required for predicting the water level at a certain time at the water level prediction point, and outputting predicted data of the water level at the water level prediction point at that time.

[0017] According to this configuration, data on the upstream of the tributary and the upstream of the main stream can be appropriately extracted as test data so that the water level of the backwater section of the tributary can be appropriately predicted.

[0018] The present disclosure also relates to a water level prediction data output method, comprising a water level prediction data output step of inputting the tributary upstream data, the main stream upstream data, and the main stream downstream data into the water level prediction model constructed by the water level prediction model construction method described above as test data required for predicting the water level at a certain time at the water level prediction point, and outputting predicted data of the water level at the water level prediction point at that time.

[0019] According to this configuration, data on the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream can be appropriately extracted as test data so that the water level of the backwater section of the tributary can be appropriately predicted.

[0020] The present disclosure also provides a water level prediction data output program for causing a computer to execute the water level prediction data output step of the water level prediction data output method described above.

[0021] According to this configuration, it is possible to provide a program having the above-described effects.

[0022] In order to solve further issues, as test data, when water level data from an upstream tributary at the water level prediction point is missing, the water level data from the upstream tributary at the missing time at the water level prediction point is interpolated using predicted water level data from the missing time at an arbitrary time after the predicted time at the water level prediction point, which is output after reflecting rainfall and actual water level conditions at a predicted time an arbitrary time before the missing time.

[0023] Specifically, the present disclosure is a test data interpolation method that, when using the water level prediction data output method described above, when the upstream tributary data for the water level at the water level prediction point is missing, interpolates the upstream tributary data for the water level at the missing time at the water level prediction point, and is characterized by including a test data interpolation step of interpolating with the predicted data of the water level at the missing time after the predicted time at the water level prediction point, which was output at a predicted time an arbitrary time before the missing time in the water level prediction data output step.

[0024] According to this configuration, even when water level test data at the water level prediction point is missing, the water level prediction accuracy at the water level prediction point can be improved by interpolating the water level test data at the water level prediction point while reflecting the rainfall and actual water level conditions.

[0025] The present disclosure also relates to a test data interpolation method, characterized in that the water level prediction data output step of the water level prediction data output method described above inputs at least one of the tributary upstream data, the mainstream upstream data, and the mainstream downstream data as the test data into the water level prediction model for predicting the water level at at least one of the upstream point of the tributary, the confluence, the upstream point of the mainstream, and the downstream point of the mainstream as the test data, and also outputs the predicted data of the water level at the relevant point at the relevant time, and when the test data of the water level at the relevant point is missing during the use of the water level prediction data output method described above, the water level prediction data output step of the water level prediction data output method described above interpolates with the predicted data of the water level at the missing time at the relevant point, which is output at a predicted time an arbitrary time before the missing time by the water level prediction data output step of the water level prediction data output method described above.

[0026] According to this configuration, even when water level test data is missing at the upstream point of the tributary, the confluence, the upstream point of the main stream, and the downstream point of the main stream, the accuracy of water level prediction at the water level prediction point can be improved by interpolating the water level test data at the upstream point of the tributary, the confluence, the upstream point of the main stream, and the downstream point of the main stream while reflecting the rainfall and actual water level conditions.

[0027] The present disclosure also provides a test data interpolation method, characterized in that, when the test data for rainfall at at least any one of the water level prediction point, the upstream point of the tributary, the confluence point, the upstream point of the main stream, and the downstream point of the main stream is missing, the test data for rainfall at the missing time at the point is interpolated using rainfall forecast data at the missing time at the point.

[0028] According to this configuration, even when rainfall test data is missing at the water level prediction point, the upstream point of the tributary, the confluence, the upstream point of the main stream, and the downstream point of the main stream, the accuracy of water level prediction at the water level prediction point can be improved by reflecting the rainfall forecast data and interpolating the rainfall test data at the water level prediction point, the upstream point of the tributary, the confluence, the upstream point of the main stream, and the downstream point of the main stream.

[0029] The present disclosure also relates to a test data interpolation method, characterized in that the arbitrary time set by the test data interpolation step is a time equal to the time interval at which the water level prediction data output step outputs the prediction data for water level.

[0030] With this configuration, the water level prediction data for the missing time after the predicted time at the water level prediction point, a short time (the time interval for water level prediction) before the missing time, can be interpolated using the rainfall and actual water level conditions reflected and output.

[0031] The present disclosure also provides a test data interpolation program for causing a computer to execute each processing step of the test data interpolation method described above.

[0032] According to this configuration, it is possible to provide a program having the above-described effects.

[0033] In actual operation, the upstream tributary data and upstream main stream data showing water level, "past observed" rainfall, and "forecast" rainfall are extracted as test data. However, the upstream tributary data and upstream main stream data showing "forecast" rainfall are less accurate than the upstream tributary data and upstream main stream data showing "future observed" rainfall, so the "forecast" water level data can be significantly different from the "observed" water level data.

[0034] In order to solve the above problem, in the verification stage, upstream tributary data and upstream main stream data indicating water level, "past observation" rainfall, and "future observation" rainfall are extracted as verification data, a probability distribution of water level prediction errors is calculated, and an error band of predicted water level data relative to the observed water level data is set. Then, upstream tributary data and upstream main stream data indicating "future observation" rainfall are replaced with upstream tributary data and upstream main stream data indicating "forecast" rainfall as verification data, and the proportion of predicted water level data that falls within the error band around the observed water level data is calculated as the water level prediction accuracy rate.

[0035] In the prediction stage, as test data, tributary upstream data and main upstream data showing water level, "past observation" rainfall, and "forecast" rainfall are extracted in actual operation, and the prediction accuracy rate of water level is output together with the prediction data of water level. Here, the prediction accuracy rate of water level is output for each issuance standard and / or for each prediction after N hours.

[0036] Specifically, the present disclosure is a water level prediction accuracy calculation method characterized by comprising, in order, a verification data extraction step of extracting the tributary upstream data and the main upstream data indicating the water level, past observed rainfall, and future forecast rainfall from a verification target time as verification data for verifying the water level prediction model constructed by the water level prediction model construction method described above; a water level prediction data output step of inputting the tributary upstream data and the main upstream data indicating the water level, past observed rainfall, and forecast rainfall to verify the water level at multiple times at the water level prediction point as the verification data to the water level prediction model, and outputting predicted data of the water level at the water level prediction point at the multiple times; and a water level prediction accuracy calculation step of calculating the proportion of the predicted data of the water level at the multiple times at the water level prediction point that falls within an error band around the observed data of the water level 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.

[0037] According to this configuration, when predicting water levels in the backwater sections of tributaries using machine learning, even when tributary upstream data and main stream upstream data showing "forecasted" rainfall are extracted as test data, the accuracy rate of water level predictions can be calculated in the verification stage, making it possible to visualize the reliability of the water level prediction data.

[0038] The present disclosure also provides a water level prediction accuracy calculation method, characterized in that the verification data extraction step also extracts, as the verification data, the upstream tributary data and the upstream main stream data indicating future observed rainfall from the verification time, inputs, as the verification data, the upstream tributary data and the upstream main stream data indicating water levels, past observed rainfall and future observed rainfall for verifying the water levels at the water level prediction point at the multiple times, calculates a probability distribution of the water level prediction error at the water level prediction point, and includes a water level prediction error range setting step between the verification data extraction step and the water level prediction data output step.

[0039] According to this configuration, when predicting the water level in the backwater section of a tributary using machine learning, in the verification stage, the probability distribution of the water level prediction error can be calculated, an error range of the water level prediction data relative to the observed water level data can be set, and the accuracy rate of the water level prediction can be calculated.

[0040] The present disclosure also relates to 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.

[0041] According to this configuration, when using machine learning to predict the water level of the backwater section of a tributary, even when tributary upstream data and main river upstream data showing ``forecast'' rainfall are extracted as test data, the accuracy rate of the water level prediction can be calculated in the verification stage to assist in determining whether each issuance standard will be reached.

[0042] The present disclosure also provides a water level prediction accuracy calculation method, characterized in that the water level prediction accuracy calculation step outputs a prediction accuracy rate of the water level at the water level prediction point together with predicted data of the water level at the prediction time at the water level prediction point based on the tributary upstream data and the main stream upstream data indicating the water level, past observed rainfall, and forecast rainfall.

[0043] According to this configuration, when using machine learning to predict the water level in the backwater section of a tributary, even when tributary upstream data and main stream upstream data showing ``forecast'' rainfall are 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.

[0044] The present disclosure also provides a water level prediction accuracy calculation program for causing a computer to sequentially execute each of the processing steps included in the water level prediction accuracy calculation method described above.

[0045] According to this configuration, it is possible to provide a program having the above-described effects.

[0046] The above disclosed inventions can be combined as much as possible. Effect of the Invention

[0047] In this way, the present disclosure can use machine learning to predict the water level in the backwater section of a tributary, and can appropriately extract training data and test data so as to increase the accuracy of learning the water level in the backwater section of a tributary and appropriately predict the water level in the backwater section of a tributary. [Brief description of the drawings]

[0048] [Figure 1] FIG. 1 illustrates a prior art training and test data extraction process. [Diagram 2] FIG. 1 is a diagram illustrating a configuration of a water level prediction device according to the present disclosure. [Diagram 3] FIG. 2 is a diagram showing the procedure of the water level prediction process of the present disclosure. [Figure 4] FIG. 2 illustrates a training data and test data extraction process of the present disclosure. [Diagram 5] 1 is a diagram showing an example of a water level prediction process according to the prior art and the present disclosure; [Figure 6] 1 is a diagram showing an example of a water level prediction process according to the prior art and the present disclosure; [Figure 7] FIG. 13 is a diagram showing a specific example of the water level test data interpolation process of the comparative example. [Figure 8] FIG. 1 is a diagram illustrating a configuration of a water level prediction device according to the present disclosure. [Figure 9] FIG. 1 is a diagram showing a procedure of a test data interpolation process according to the present disclosure. [Figure 10] FIG. 13 illustrates a specific example of the water level test data interpolation process of the present disclosure. [Figure 11] FIG. 13 is a diagram showing a specific example of the rainfall test data interpolation process of the present disclosure. [Figure 12] FIG. 1 is a diagram illustrating a configuration of a water level prediction device according to the present disclosure. [Figure 13] FIG. 11 is a diagram showing the procedure of the water level prediction accuracy calculation process of the present disclosure. [Figure 14] FIG. 13 is a diagram showing a specific example of the water level prediction error range setting process of the present disclosure. [Figure 15] FIG. 13 is a diagram showing a specific example of the water level prediction accuracy rate calculation process of the present disclosure. [Figure 16] FIG. 13 is a diagram showing an example of the water level prediction accuracy rate output process of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0049] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present disclosure will be described with reference to the accompanying drawings. The embodiments described below are examples of the present disclosure, and the present disclosure is not limited to the following embodiments.

[0050] (Overview of the water level prediction process of the present disclosure) The configuration of the water level prediction device of the present disclosure is shown in Fig. 2. The procedure of the water level prediction process of the present disclosure is shown in Fig. 3. The extraction process of the training data and test data of the present disclosure is shown in Fig. 4.

[0051] 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 test data extraction unit 3 and a water level prediction data output unit 4, and can be realized by installing the water level prediction program shown in Figure 3 on a computer.

[0052] During floods, when the water level of the main stream is high, it becomes difficult for the water in the tributary to flow downstream, and the water in the main stream may flow back into the tributary, causing the water level in the backwater section of the tributary to rise. In other words, the water level in the backwater section of a tributary is not only affected by the amount of rainfall and water level upstream of the tributary from the backwater section, but also by the amount of rainfall and water level upstream of the main stream from the confluence of the tributary and the main stream, and can also be affected by the amount of rainfall and water level downstream of the confluence of the tributary and the main stream.

[0053] In this disclosure, not only is the influence of the rainfall and water level upstream of the tributary from the backwater section considered with respect to the water level in the backwater section of the tributary, but also the influence of the rainfall and water level upstream of the main stream from the confluence of the tributary and the main stream is considered as a first influence, and the influence of the rainfall and water level downstream of the confluence of the tributary and the main stream is considered as a second influence. Below, the details of the training data extraction process will be explained, and then the test data extraction process will be explained in detail.

[0054] (Details of the training data extraction process of the present disclosure) The training data extraction unit 1 acquires rainfall data RF and water level data WL, and extracts the following data as training data L required for learning the water level at a certain time at the water level prediction point P1 in the backwater section of the tributary (step S1): (1) tributary upstream data showing at least one of the rainfall and water level in the past from the water level prediction point P1 and at least one of the upstream points P2 and P3 of the tributary from the water level prediction point P1, (2) mainstream upstream data showing at least one of the rainfall and water level in the past from the time at the confluence point P4 of the tributary and the mainstream and at least one of the upstream points P5 and P6 of the mainstream from the confluence point P4, (3) mainstream downstream data showing at least one of the rainfall and water level in the past from the time at the downstream point P7 of the mainstream from the confluence point P4. However, it is preferable that the downstream point P7 of the mainstream is close to the confluence point P4 to such an extent that it has an effect on the water level in the backwater section of the tributary.

[0055] The water level prediction model construction unit 2 uses training data L (basically tributary upstream data and main stream upstream data, and better also main stream downstream data) to construct a water level prediction model M (neural network, etc.) for predicting the water level at the water level prediction point P1 (step S2).

[0056] Here, it is desirable for the training data extraction unit 1 to extract, as training data L, data on rainfall and water level for a period of time with high correlation compared to the water level during flooding at the water level prediction point P1 in the backwater section of the tributary, and it is desirable to discard data on rainfall and water level for a period of time with low correlation.

[0057] For example, the training data extraction unit 1 may select, as the training data L, (1) a water level peak discrimination threshold for rejecting periods of no water level fluctuation and then select an extraction period for water level data, (2) a water level peak extraction width for rejecting periods of no water level fluctuation and then select an extraction period for water level data, (3) a specified water level period for extracting periods of water level rise and fall and then select an extraction period for water level data, (4) a specified rainfall period taking into account the time it takes for rainfall to reach the water level prediction point P1 and then select an extraction period for rainfall data, or (5) a specified accumulation period taking into account the time it takes for rainfall to flow out of the soil or from a reservoir, etc. into a river and then select an extraction period for rainfall data (see Patent Document 1).

[0058] In this way, in order to increase the accuracy of learning the water level in the backwater section of the tributary, data on rainfall and water level upstream of the tributary and upstream of the main stream can basically be appropriately extracted as training data L, and even better, data on rainfall and water level downstream of the main stream can also be appropriately extracted.

[0059] (Details of the test data extraction process of this disclosure) The test data extraction unit 3 acquires the rainfall data RF and the water level data WL, and extracts the following data as test data T required for predicting the water level at a certain time at the water level prediction point P1 in the backwater section of the tributary (step S3): (1) tributary upstream data showing at least one of the rainfall and water level in the past from the water level prediction point P1 and at least one of the upstream points P2 and P3 of the tributary from the water level prediction point P1, (2) mainstream upstream data showing at least one of the rainfall and water level in the past from the time at the confluence point P4 of the tributary and the mainstream and at least one of the upstream points P5 and P6 of the mainstream from the confluence point P4, (3) mainstream downstream data showing at least one of the rainfall and water level in the past from the time at the downstream point P7 of the mainstream from the confluence point P4. However, it is preferable that the downstream point P7 of the mainstream is close to the confluence point P4 to such an extent that it has an effect on the water level in the backwater section of the tributary.

[0060] The test data extraction unit 3 inputs the test data T (basically tributary upstream data and main stream upstream data, and more preferably, main stream downstream data) to the water level prediction model M (neural network, etc.). Then, the water level prediction data output unit 4 outputs the water level prediction data P at the water level prediction point P1 at the relevant time from the water level prediction model M (step S4).

[0061] In this way, in order to properly predict the water level in the backwater section of the tributary, data on rainfall and water level upstream of the tributary and upstream of the main stream can basically be properly extracted as test data T, and better yet, data on rainfall and water level downstream of the main stream can also be properly extracted.

[0062] (Examples of water level prediction processing in the prior art and the present disclosure) Examples of water level prediction processing of the conventional technology and the present disclosure are shown in Figures 5 and 6. In the conventional technology, only tributary upstream data is extracted as training data L and test data T required for learning and predicting the water level at the water level prediction point P1 in the backwater section of the tributary. In the present disclosure, tributary upstream data, main stream upstream data, and main stream downstream data are all extracted as training data L and test data T required for learning and predicting the water level at the water level prediction point P1 in the backwater section of the tributary.

[0063] In the conventional technology shown in the upper part of Fig. 5, during flooding after rainfall, the predicted water level at water level prediction point P1 is lower than the actual water level at water level prediction point P1 and shows a slower rise. In the present disclosure shown in the lower part of Fig. 5, during flooding after rainfall, the predicted water level at water level prediction point P1 is almost equal to the actual water level at water level prediction point P1 (the upward deviation is due to the selection of such a water level prediction model M) and shows an appropriate rise.

[0064] In the conventional technology shown in the upper part of Fig. 6, during flooding after rainfall, the predicted water level at water level prediction point P1 is lower than the actual water level at water level prediction point P1 (the rate of rise is approximately the same). In the present disclosure shown in the lower part of Fig. 6, during flooding after rainfall, the predicted water level at water level prediction point P1 is approximately the same as the actual water level at water level prediction point P1 (the rate of rise is approximately the same).

[0065] (Procedure for water level test data interpolation processing in comparative example) However, it is possible that the upstream tributary data of the water level at the water level prediction point P1 may be missing as the test data T. This is because a code error may be detected in the input data from the telemeter of the water level observation station, preventing normal reception. In that case, the test data T cannot be input to the water level prediction model M, and the water level at the water level prediction point P1 cannot be predicted. Therefore, it is possible to interpolate the upstream tributary data of the water level at the water level prediction point P1.

[0066] A specific example of the water level test data interpolation process of the comparative example is shown in Fig. 7. When upstream tributary data of water level at water level prediction point P1 is missing, in order to interpolate the upstream tributary data of water level at the missing time at water level prediction point P1, the upstream tributary data of water level at the time before and / or after the missing time at water level prediction point P1 that was transmitted by the telemeter of the water level observation station at a time before and / or after the missing time is referenced.

[0067] In the first row of Fig. 7, to predict the water level at 12:30, 30 minutes from the current time, the following water level data are extracted as test data: 3.0m at 11:00, one hour before the current time, 4.0m at 11:30, 30 minutes before the current time, and 5.0m at 12:00, the current time. In the second to fourth rows of Fig. 7, some water level data is missing as test data.

[0068] In the second row of Figure 7, the water level data of 3.0m at 11:00 and the water level data of 4.0m at 11:30 are missing, and the water level data of 5.0m is observed at 12:00. Therefore, the water level data of 5.0m at 12:00 is time shifted to 3.0m at 11:00 and 4.0m at 11:30, resulting in a large error in the interpolation of the water level data of 5.0m.

[0069] In the third row of Figure 7, the water level data of 4.0m at 11:30 and the water level data of 5.0m at 12:00 are missing, and the water level data of 3.0m is observed at 11:00. Therefore, the water level data of 3.0m at 11:00 is time shifted to 4.0m at 11:30 and 5.0m at 12:00, resulting in a large error in the interpolation of the water level data of 3.0m.

[0070] In the fourth row of Figure 7, the water level data of 4.0m at 11:30 is missing, and the water level data of 3.0m at 11:00 and the water level data of 5.0m at 12:00 are observed. Therefore, the water level data of 4.0m at 11:30 is linearly interpolated between the water level data of 3.0m at 11:00 and the water level data of 5.0m at 12:00, and the water level data of 4.0m is almost correctly interpolated.

[0071] In the fourth row of Figure 7, the tributary upstream data for water levels both before and after the missing data time are referenced, so the rainfall conditions are reflected. In the second and third rows of Figure 7, the tributary upstream data for water levels only before or after the missing data time are referenced, so the rainfall conditions are not reflected.

[0072] (Water Level Test Data Interpolation Procedure of the Present Disclosure) The configuration of the water level prediction device of the present disclosure is also shown in Fig. 8. The procedure of the test data interpolation process of the present disclosure is shown in Fig. 9. A specific example of the water level test data interpolation process of the present disclosure is shown in Fig. 10. The water level prediction device W also includes a test data interpolation unit 5, and the test data interpolation program shown in Fig. 9 can also be installed and realized in a computer.

[0073] When performing the water level test data interpolation process (particularly, step S14) of the present disclosure, the water level prediction model M, which predicts the water levels at at least any of the upstream tributary points P2, P3, the confluence point P4, the upstream main stream points P5, P6, and the downstream main stream point P7, inputs at least any of the upstream tributary data, upstream main stream data, and downstream main stream data as test data T, and also outputs water level prediction data P for the relevant point and time.

[0074] When the upstream tributary water level data at the water level prediction point P1 is missing (step S11), the test data interpolation unit 5 interpolates the upstream tributary water level data at the missing time at the water level prediction point P1 (step S12).

[0075] Here, the test data interpolation unit 5 interpolates using the water level prediction data P at a missing time any time after the predicted time at the water level prediction point P1, which was output by the water level prediction data output unit 4 at a predicted time any time before the missing time (step S12).

[0076] When water level test data T is missing at at least any of the upstream points P2, P3 of the tributary, the confluence point P4, the upstream points P5, P6 of the main stream, and the downstream point P7 of the main stream (step S13), the test data interpolation unit 5 interpolates the water level test data T at the time of the missing data at the relevant points (step S14).

[0077] Here, the test data interpolation unit 5 interpolates with the water level prediction data P at a missing time any time after the predicted time at at least any of the upstream points P2 and P3 of the tributary, the confluence point P4, the upstream points P5 and P6 of the main stream, and the downstream point P7 of the main stream, which are output by the water level prediction data output unit 4 at a predicted time any time before the missing time (step S14).

[0078] In addition, when water level test data T is missing at at least any of the upstream points P2 and P3 of the tributary, the confluence point P4, the upstream points P5 and P6 of the main stream, and the downstream point P7 of the main stream, even if the water level test data T at the water level prediction point P1 is not missing, the water level prediction data P at the water level prediction point P1 cannot be output. Therefore, the water level test data T at the missing time at at least any of the upstream points P2 and P3 of the tributary, the confluence point P4, the upstream points P5 and P6 of the main stream, and the downstream point P7 of the main stream is interpolated with the water level prediction data P at that point.

[0079] Moreover, the arbitrary time set by the test data interpolation unit 5 is equal to the time interval at which the water level prediction data output unit 4 outputs the water level prediction data P. In other words, the rainfall forecast for a short time later is highly accurate, and the arbitrary time set by the test data interpolation unit 5 is short.

[0080] The test data extraction unit 3 inputs test data T (including test data T of the water level at the missing time interpolated by the test data interpolation unit 5; test data T of the rainfall at the missing time will be described later) to the water level prediction model M (step S17). The water level prediction data output unit 4 outputs water level prediction data P at the relevant time at the water level prediction point P1 from the water level prediction model M (step S18). The water level prediction data output unit 4 also outputs water level prediction data P at the relevant time at least one of the upstream points P2 and P3 of the tributaries, the confluence point P4, the upstream points P5 and P6 of the main stream, and the downstream point P7 of the main stream from the water level prediction model M (step S18).

[0081] In the first row of Fig. 10, to predict the water level at 12:30, 30 minutes from the current time, the water level data of 3.0 m at 11:00, one hour before the current time, the water level data of 4.0 m at 11:30, 30 minutes before the current time, and the water level data of 5.0 m at the current time of 12:00 are extracted as test data. In the third to fifth rows of Fig. 10, some water level data is missing as test data.

[0082] In the second stage of Fig. 10, as the results of predicting the water level 30 minutes after the current time at each time, the water level prediction data of 2.9 m at 11:00, one hour before the current time, the water level prediction data of 3.9 m at 11:30, 30 minutes before the current time, and the water level prediction data of 4.8 m at 12:00, the current time, are output. In the second stage of Fig. 10, the arbitrary time set by the test data interpolation unit 5 is 30 minutes.

[0083] In the third row of Figure 10, the water level data of 3.0m at 11:00 and the water level data of 4.0m at 11:30 are missing, and the water level data of 5.0m is observed at 12:00. Therefore, the water level data of 3.0m at 11:00 and the water level data of 4.0m at 11:30 are interpolated with the predicted water level data of 2.9m at 11:00 and the predicted water level data of 3.9m at 11:30, respectively, and the water level data of 2.9m and 3.9m are almost correctly interpolated.

[0084] In the fourth row of Figure 10, the water level data of 4.0m at 11:30 and the water level data of 5.0m at 12:00 are missing, and the water level data of 3.0m is observed at 11:00. Therefore, the water level data of 4.0m at 11:30 and the water level data of 5.0m at 12:00 are interpolated with the predicted water level data of 3.9m at 11:30 and the predicted water level data of 4.8m at 12:00, respectively, and the water level data of 3.9m and 4.8m are almost correctly interpolated.

[0085] In the fifth row of Figure 10, the water level data of 4.0 m at 11:30 is missing, and the water level data of 3.0 m at 11:00 and the water level data of 5.0 m at 12:00 are observed. Therefore, the water level data of 4.0 m at 11:30 is interpolated with the predicted water level data of 3.9 m at 11:30, and the water level data of 3.9 m is almost correctly interpolated.

[0086] In this way, even when water level test data at a water level prediction point is missing, the accuracy of water level prediction at the water level prediction point can be improved by interpolating the water level test data at the water level prediction point while reflecting the rainfall and actual water level conditions.

[0087] Furthermore, even when water level test data is missing at the upstream point of a tributary, the confluence, the upstream point of the main stream, and the downstream point of the main stream, the accuracy of water level prediction at the water level prediction point can be improved by interpolating the water level test data at the upstream point of a tributary, the confluence, the upstream point of the main stream, and the downstream point of the main stream, while reflecting the rainfall and actual water level conditions.

[0088] Furthermore, the water level prediction data for the missing time after the predicted time at the water level prediction point, a short time (the water level prediction time interval) before the missing time, is output after reflecting the rainfall and actual water level conditions, and then interpolated using water level prediction data.

[0089] (Procedure for the Rainfall Test Data Interpolation Process of the Present Disclosure) A specific example of the rainfall test data interpolation process of the present disclosure is shown in Fig. 11. When rainfall test data T is missing at least one of the water level prediction point P1, the upstream points P2 and P3 of the tributaries, the confluence point P4, the upstream points P5 and P6 of the main stream, and the downstream point P7 of the main stream (step S15), the test data interpolation unit 5 interpolates the rainfall test data T at the missing time at the relevant points (step S16).

[0090] Here, the test data interpolation unit 5 interpolates using rainfall forecast data RD for the missing time at at least one of the water level prediction point P1, the upstream points P2 and P3 of the tributaries, the confluence point P4, the upstream points P5 and P6 of the main stream, and the downstream point P7 of the main stream (step S16).

[0091] In addition, when rainfall test data T is missing at at least any of the upstream points P2 and P3 of the tributary, the confluence point P4, the upstream points P5 and P6 of the main stream, and the downstream point P7 of the main stream, even if the rainfall test data T at the water level prediction point P1 is not missing, the water level prediction data P at the water level prediction point P1 cannot be output. Therefore, the rainfall test data T at the missing time at at least any of the upstream points P2 and P3 of the tributary, the confluence point P4, the upstream points P5 and P6 of the main stream, and the downstream point P7 of the main stream is interpolated using the rainfall forecast data RD at the relevant point.

[0092] The test data extraction unit 3 inputs test data T (including test data T of rainfall at the missing time interpolated by the test data interpolation unit 5, and the test data T of water level at the missing time is described above) to the water level prediction model M (step S17). The water level prediction data output unit 4 outputs water level prediction data P at the relevant time at the water level prediction point P1 from the water level prediction model M (step S18). The water level prediction data output unit 4 also outputs water level prediction data P at the relevant time at least one of the upstream points P2 and P3 of the tributaries, the confluence point P4, the upstream points P5 and P6 of the main stream, and the downstream point P7 of the main stream from the water level prediction model M (step S18).

[0093] In the first row of Fig. 11, to predict the water level at 12:30, 30 minutes from the current time, the following rainfall data are extracted as test data: 30 mm at 11:00, one hour before the current time, 40 mm at 11:30, 30 minutes before the current time, and 50 mm at 12:00, the current time. In the third to fifth rows of Fig. 11, some rainfall data is missing as test data.

[0094] In the second row of Figure 11, as rainfall forecast data RD (short-term precipitation forecast), rainfall forecast data of 29 mm at 11:00, one hour before the current time, rainfall forecast data of 39 mm at 11:30, 30 minutes before the current time, and rainfall forecast data of 48 mm at 12:00, the current time, are input.

[0095] In the third row of Figure 11, the rainfall data of 30 mm at 11:00 and 40 mm at 11:30 are missing, and the rainfall data of 50 mm was observed at 12:00. Therefore, the rainfall data of 30 mm at 11:00 and 40 mm at 11:30 are interpolated with the rainfall forecast data of 29 mm at 11:00 and 39 mm at 11:30, respectively, and the rainfall data of 29 mm and 39 mm are almost correctly interpolated.

[0096] In the fourth row of Figure 11, the rainfall data of 40 mm at 11:30 and the rainfall data of 50 mm at 12:00 are missing, and the rainfall data observed at 11:00 is 30 mm. Therefore, the rainfall data of 40 mm at 11:30 and the rainfall data of 50 mm at 12:00 are interpolated with the rainfall forecast data of 39 mm at 11:30 and the rainfall forecast data of 48 mm at 12:00, respectively, and the rainfall data of 39 mm and 48 mm are almost correctly interpolated.

[0097] In the fifth row of Figure 11, the rainfall data of 40 mm at 11:30 is missing, and the rainfall data of 30 mm at 11:00 and the rainfall data of 50 mm at 12:00 are observed. Therefore, the rainfall data of 40 mm at 11:30 is interpolated with the rainfall forecast data of 39 mm at 11:30, and the rainfall data of 39 mm is almost correctly interpolated.

[0098] In this way, even when rainfall test data is missing at the water level prediction point, the upstream point of the tributary, the confluence, the upstream point of the main stream, and the downstream point of the main stream, the accuracy of water level prediction at the water level prediction point can be improved by reflecting the rainfall forecast data and interpolating the rainfall test data at the water level prediction point, the upstream point of the tributary, the confluence, the upstream point of the main stream, and the downstream point of the main stream.

[0099] (Configuration of the water level prediction device of the present disclosure) The configuration of the water level prediction device disclosed herein is shown in Fig. 12. The procedure for the water level prediction accuracy calculation process disclosed herein is shown in Fig. 13. In addition to Figs. 2 and 8, the water level prediction device W includes a verification data extraction unit 6, a water level prediction error range setting unit 7, and a water level prediction accuracy calculation unit 8. The verification data extraction unit 6, the water level prediction error range setting unit 7, and the water level prediction accuracy calculation unit 8 can be realized by installing the water level prediction accuracy calculation program shown in Fig. 13 in a computer.

[0100] The training data extraction unit 1 (not shown in FIG. 12) extracts, as training data L required for learning the water level at a certain time at a water level prediction point, at least any of water level data WL at the tributary upstream, the main stream upstream, and the main stream downstream at the water level prediction point and at least any of water level data WL at the tributary upstream, the main stream upstream, and the main stream downstream at the water level prediction point and at the time and in the past, and observed rainfall data RO in the future from the time. Here, the training data extraction unit 1 may extract water level data WL at the tributary upstream, the main stream upstream, and the main stream downstream 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 the tributary upstream, the main stream upstream, and the main stream downstream at the time of water level non-fluctuation, etc., immediately before and immediately after the water level fluctuation. The water level prediction model construction unit 2 constructs a water level prediction model M using the training data L.

[0101] The test data extraction unit 3 extracts, as test data T required for predicting the water level at a certain time at a water level prediction point, at least any of water level data WL at the tributary upstream, main upstream, and main downstream at the water level prediction point and at least one of the water level data WL at the water level prediction point and the water level data WL at the water level prediction point and the water level data WL at the water level prediction point and the water level data WL at the water level prediction point and the water level data WL at the water level prediction point. The test data extraction unit 3 extracts, as test data T required for predicting the water level at a certain time at a water level prediction point, the water level prediction point and the water level data WL at the water level prediction point and the water level data WL at the water level prediction point and the water level data WL at the water level prediction point. The test data extraction unit 3 extracts, as test data T required for predicting the water level at a certain time at a tributary upstream, main upstream, and main downstream at .... The test data extraction unit 3 extracts, as test data T required for predicting the water level at a certain time at a tributary upstream, main upstream, and main downstream at the water level prediction point. The test data extraction unit 3 extracts, as test data T required for predicting the water level at a certain time at a tributary upstream, main upstream, and main downstream at the water level prediction point. The test data extraction unit 3 extracts, as test data T required for predicting the water level at a certain time at a tributary upstream, main upstream, and main downstream at the water level prediction point. The test data extraction unit 3 extract

[0102] However, the "forecast" rainfall data RP for the upstream of the tributaries, the upstream of the main stream, and the downstream of the main stream is less accurate than the "future observation" rainfall data RO for the upstream of the tributaries, the upstream of the main stream, and the downstream of the main stream, so the "prediction" water level data P may deviate significantly from the "observation" water level data. Therefore, in this disclosure, the reliability of the "prediction" water level data P is visualized.

[0103] In the verification stage, water level data WL in the upstream of the tributaries, the upstream of the main stream, and the downstream of the main stream, "past observation" rainfall data RO, and "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 for the water level observation data is set. Then, as verification data V, the "future observation" rainfall data RO in the upstream of the tributaries, the upstream of the main stream, and the downstream of the main stream is replaced with the "forecast" rainfall data RP in the upstream of the tributaries, the upstream of the main stream, and the downstream of the main stream, 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.

[0104] In the prediction stage, water level data WL from upstream tributaries, upstream main stream and downstream main stream are extracted as test data T in actual operation, "past observation" rainfall data RO and "forecast" rainfall data RP are extracted, and the water level prediction accuracy rate H is output together 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.

[0105] Here, the training data L is known data for water level learning, and the water level prediction accuracy rate H is nearly 100%, so 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 is specialized for a specific rainfall condition, so 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 is not specialized for a specific rainfall condition, so it can be applied to calculate the water level prediction accuracy rate H.

[0106] (Specific example of water level prediction error range setting process of the present disclosure) A specific example of the water level prediction error range setting process of the present disclosure is shown in Fig. 14. The verification data extraction unit 6 extracts water level data WL and past observed rainfall data RO at the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream, and future observed rainfall data RO from the verification target time as verification data V for verifying the water level prediction model M constructed in the water level prediction model construction process of the present disclosure (step S21). Here, the verification data extraction unit 6 may extract water level data WL, past observed rainfall data RO, and future observed rainfall data RO at the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream during water level fluctuations, etc., or may extract water level data WL, past observed rainfall data RO, and future observed rainfall data RO at the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream during water level non-fluctuations, etc. immediately before and immediately after water level fluctuations, and the water level prediction error range setting process does not extract forecast rainfall data RP at the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream.

[0107] The verification data extraction unit 6 inputs the water level data WL of the tributary upstream, the main stream upstream, and the main stream downstream, the past observed rainfall data RO, and the future observed rainfall data RO as verification data V to the water level prediction model M for verifying the water level at multiple times at the water level prediction point (step S22). Then, the water level prediction error range setting unit 7 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 S25 (step S23). Here, the water level prediction error range setting unit 7 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 S22 to each prediction target time, and sets the error range of the water level prediction data P (step S23).

[0108] In the "Prediction error N hours later" column in Fig. 14, 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 N hours later" column in Fig. 14, 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 prediction data P for the water level observation data is set so that is within the error range. Here, the minimum and maximum values ​​of the prediction error probability distribution E are set to 0% and 100%, 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 ranges 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.

[0109] 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 will be finally set as follows. First, water level data and analyzed rainfall data (radar interpolated with rain gauge) from the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream are input as test data to the water level prediction model M. Next, e is set so that 80% or more of the water level prediction data for each of the high water level, low water level, and medium water level falls within the error band 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.

[0110] In this way, when predicting water levels in the backwater sections of tributaries using machine learning, in the verification stage, the water level prediction error probability distribution E can be calculated, the error band 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.

[0111] Furthermore, when using machine learning to predict water levels in the backwater sections of tributaries, even when "forecast" rainfall data RP is extracted for the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream as test data T, by preparing to calculate the water level prediction accuracy rate H in the verification stage, it is possible to prepare to support the judgment of whether each issuance standard will be reached.

[0112] (Specific example of water level prediction accuracy calculation process of the present disclosure) A specific example of the water level prediction accuracy calculation process of the present disclosure is shown in Fig. 15. The verification data extraction unit 6 extracts water level data WL and past observed rainfall data RO at the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream, and forecast rainfall data RP from the verification target time as verification data V for verifying the water level prediction model M constructed in the water level prediction model construction process of the present disclosure (step S21). Here, the verification data extraction unit 6 may extract water level data WL, past observed rainfall data RO, and forecast rainfall data RP at the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream during water level fluctuations, or may extract water level data WL, past observed rainfall data RO, and forecast rainfall data RP at the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream during water level non-fluctuations, etc. immediately before and after water level fluctuations, and the water level prediction accuracy calculation process does not extract future observed rainfall data RO at the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream.

[0113] The verification data extraction unit 6 inputs the water level data WL of the tributary upstream, the main stream upstream, and the main stream downstream, the past observed rainfall data RO, and the forecast rainfall data RP to the water level prediction model M as the verification data V for verifying the water level at the water level prediction point at multiple times (step S24). Then, the water level prediction model M outputs the water level prediction data P at the water level prediction point at the multiple times. Furthermore, the water level prediction accuracy calculation unit 8 calculates the proportion of the water level prediction data P at the multiple times at the water level prediction point that falls within the error band around the water level observation data at the multiple times at the water level prediction point as the water level prediction accuracy H at the water level prediction point (step S25). Here, the water level prediction accuracy calculation unit 8 calculates the water level prediction accuracy H 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 S24 to each prediction target time (step S25).

[0114] In the "Predicted data after each N hours" column in FIG. 15, water level prediction data P for the multiple times at the water level prediction point is output for observed water levels of 0 m or more and less than the alert standard water level A, observed water levels of 0 m or more and less than the alert standard water level A and observed water levels of 0 m or more and less than the alert standard water level B, and observed water levels of 0 m or more and less than the alert 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 0 m or more and less than the alert standard water level A, 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 0 m or more and less than the alert standard water level B, 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.

[0115] In the "Prediction accuracy rate after each N hours" column in Figure 15, the water level prediction accuracy rate H at the water level prediction point is calculated for observed water levels 0m or more and less than the announcement standard water level A, observed water levels A or more and less than the announcement standard water level B, and observed water levels B or more. For observed water levels 0m or more and less than the announcement standard water level A, the water level prediction accuracy rate is 80 / (80+10+10)×100=80%. For observed water levels A or more and less than the announcement standard water level B, the water level prediction accuracy rate is 30 / (30+15+15)×100=50%. For observed water levels B or more, the water level prediction accuracy rate is 10 / (10+10+10)×100=33%.

[0116] Even if the 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. Also, when the 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.

[0117] In this way, when using machine learning to predict the water level in the backwater section of a tributary, even when ``forecast'' rainfall data RP for the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream are 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.

[0118] Furthermore, when using machine learning to predict water levels in the backwater sections of tributaries, even when "forecast" rainfall data RP for the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream are extracted as test data T, the water level prediction accuracy rate H can be calculated in the verification stage to assist in determining whether each issuance standard will be reached.

[0119] (Example of the water level prediction accuracy output process of the present disclosure) An example of the water level prediction accuracy rate output process of the present disclosure is shown in Fig. 16. When predicting the water level at the water level prediction point at the prediction target time based on the water level data WL at the tributary upstream, main stream upstream and main stream downstream, the past observed rainfall data RO and the forecast rainfall data RP, the water level prediction accuracy rate calculation unit 8 outputs the water level prediction accuracy rate H at the water level prediction point together with the water level prediction data P at the prediction target time at the water level prediction point (step S26). Here, the water level prediction accuracy rate calculation unit 8 outputs the water level prediction accuracy rate H 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 to each prediction target time (step S26).

[0120] In FIG. 16, L0, L1, L2, L3, L4, and L5 are set as the water levels for issuing warnings. In the verification stage, the accuracy rate of water level prediction for the observed water levels for issuing warnings, L0-L1, L1-L2, L2-L3, L3-L4, and L4-L5, is H 01 %, H 12 %, H 23 %, H 34 %, H 45 In the prediction stage, the accuracy rate of the water level prediction 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 (prediction six hours later) is calculated as H 01 %(hit), H 23 % (miss), H 45 %(hit), H 01 The output will be % (hit).

[0121] In this way, when using machine learning to predict the water level in the backwater section of a tributary, even when ``forecast'' rainfall data RP for the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream are 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 in the prediction stage.

[0122] Furthermore, when using machine learning to predict the water level in the backwater section of a tributary, even when ``forecast'' rainfall data RP for the upstream of the tributary, the upstream of the main stream, and the downstream of the main stream are 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]

[0123] The water level prediction model construction method, water level prediction model construction program, water level prediction data output method, and water level prediction data output program disclosed herein use machine learning to predict the water level in the backwater section of a tributary, and can appropriately extract training data and test data so as to increase the accuracy of learning the water level in the backwater section of a tributary and to appropriately predict the water level in the backwater section of a tributary. [Explanation of symbols]

[0124] W: Water level prediction device M: Water level prediction model 1: Training data extraction part 2: Water level prediction model construction section 3: Test data extraction part 4: Water level prediction data output section 5: Test data interpolation part 6: Verification data extraction section 7: Water level prediction error range setting section 8: Water level prediction accuracy calculation section RF: Rainfall data WL: Water level data RO: Observed rainfall data RP: Predicted 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 RD: Rainfall forecast data P1: Water level prediction point P2, P3: Upstream points of the tributary P4: Meeting point P5, P6: upstream of the main stream P7: Downstream of the main stream

Claims

1. a training data extraction step for extracting, as training data necessary for learning the water level at a certain time at a water level prediction point in a backwater section of a tributary, (1) tributary upstream data indicating at least one of the rainfall and water level in the past from the time at the water level prediction point and / or at a point upstream of the tributary from the water level prediction point, and (2) mainstream upstream data indicating at least one of the rainfall and water level in the past from the time at the confluence of the tributary and mainstream and / or at a point upstream of the mainstream from the confluence; a water level prediction model construction step of constructing a water level prediction model for predicting the water level at the water level prediction point using the tributary upstream data and the mainstream upstream data; A water level prediction model construction method comprising:

2. The training data extraction step includes extracting, as the training data, in addition to the tributary upstream data and the mainstream upstream data, (3) mainstream downstream data indicating at least one of a rainfall amount and a water level in the past from the confluence to a point downstream of the mainstream, and The water level prediction model construction step constructs the water level prediction model using the tributary upstream data, the mainstream upstream data, and the mainstream downstream data. The method for constructing a water level prediction model according to claim 1 .

3. A water level prediction model construction program for causing a computer to sequentially execute each processing step of the water level prediction model construction method according to claim 1 or 2.

4. a water level prediction data output step of inputting the tributary upstream data and the mainstream upstream data as test data required for predicting the water level at a certain time at the water level prediction point into the water level prediction model constructed by the water level prediction model construction method according to claim 1, and outputting predicted data of the water level at the water level prediction point at that time; A water level prediction data output method comprising:

5. a water level prediction data output step of inputting the tributary upstream data, the mainstream upstream data, and the mainstream downstream data as test data required for predicting the water level at a certain time at the water level prediction point into the water level prediction model constructed by the water level prediction model construction method according to claim 2, and outputting predicted data of the water level at the water level prediction point at that time; A water level prediction data output method comprising:

6. A water level prediction data output program for causing a computer to execute the water level prediction data output step of the water level prediction data output method according to claim 4 or 5.

7. a test data interpolation step of interpolating the upstream tributary data of the water level at the water level prediction point at the missing time when the upstream tributary data of the water level at the water level prediction point is missing, the test data interpolation step being performed using the predicted data of the water level at the missing time after the predicted time at the water level prediction point, the predicted data being output at a predicted time an arbitrary time before the missing time in the water level prediction data output step; 11. A method for test data interpolation comprising:

8. The water level prediction data output step of the water level prediction data output method according to claim 5 includes inputting at least one of the tributary upstream data, the mainstream upstream data, and the mainstream downstream data as the test data into the water level prediction model for predicting water levels at at least one of the upstream point of the tributary, the confluence, the upstream point of the mainstream, and the downstream point of the mainstream, and outputting the prediction data of the water level at the point at the time, When using the water level prediction data output method according to claim 5, when the test data of the water level at the relevant point is missing, the test data of the water level at the missing time at the relevant point is interpolated by a test data interpolation step of interpolating with the predicted data of the water level at the missing time after the predicted time at the relevant point, which is output at a predicted time an arbitrary time before the missing time by the water level prediction data output step of the water level prediction data output method according to claim 5; 11. A method for test data interpolation comprising:

9. The test data interpolation step, when the test data of the rainfall at at least one of the water level prediction point, the upstream point of the tributary, the confluence point, the upstream point of the main stream, and the downstream point of the main stream is missing, interpolates the test data of the rainfall at the missing time at the point using rainfall forecast data at the missing time at the point.

9. The method of test data interpolation according to claim 7 or 8, characterized in that:

10. The arbitrary time set in the test data interpolation step is equal to a time interval at which the water level prediction data output step outputs the prediction data of the water level.

9. The method of test data interpolation according to claim 7 or 8, characterized in that:

11. 9. A test data interpolation program for causing a computer to execute each processing step of the test data interpolation method according to claim 7.

12. A verification data extraction step of extracting the tributary upstream data and the main stream upstream data, which indicate water levels, past observed rainfall, and future forecast rainfall from a verification target time, as verification data for verifying the water level prediction model constructed by the water level prediction model construction method according to claim 1; a water level prediction data output step of inputting, as the verification data, the tributary upstream data and the mainstream upstream data indicating the water level, past observed rainfall, and forecast rainfall for verifying the water level at the water level prediction point at multiple times into the water level prediction model, and outputting predicted data of the water level at the water level prediction point at the multiple times; a water level prediction accuracy calculation step of calculating a rate of prediction data of the water level at the water level prediction point at the multiple times that falls within an error band around the observed data of the water level at the water level prediction point at the multiple times as a prediction accuracy rate of the water level at the water level prediction point; A water level prediction accuracy calculation method comprising the steps of:

13. The verification data extraction step also extracts, as the verification data, the tributary upstream data and the mainstream upstream data indicating future observed precipitation amounts from the verification target time, a water level prediction error range setting step of inputting, as the verification data, the water level for verifying the water levels at the water level prediction point at the multiple times, the tributary upstream data and the main stream upstream data indicating past observed rainfall and future observed rainfall 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 rate calculation method according to claim 12 .

14. The water level prediction accuracy rate 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. The water level prediction accuracy rate calculation method according to claim 12 or 13,

15. 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 prediction target time at the water level prediction point when predicting the water level at the prediction target time at the water level prediction point based on the tributary upstream data and the main stream upstream data indicating the water level, past observed rainfall, and forecast rainfall. The water level prediction accuracy rate calculation method according to claim 12 or 13,

16. A water level prediction accuracy rate calculation program for causing a computer to execute each processing step of the water level prediction accuracy rate calculation method according to claim 12 or 13 in sequence.

Images