Water level prediction device, water level prediction method, and computer program

Abstract

To provide a water level prediction device for predicting the water level in a sewer with high accuracy.SOLUTION: A water level prediction device includes: means 12 for generating a water level prediction model and a water level feature amount model using input data including time series data on an actual measured value of water level, weather information, and rainfall, data on structure information of a sewer, and data on information about topography, and calculating time series data on a prediction value of water level and a water level feature amount of input data; means 14 for generating a plurality of water level state clusters for clustering the input data and outputting a value indicating the water level state cluster of the input data; means 15 for generating a model for calculating a prediction width of the prediction value of water level and outputting the prediction width as time series data; means 11 for calculating a precipitation risk using a prediction value of an increment of water content in the sky; and means 13 for calculating a maximum value and a minimum value of the water level and outputting together with the prediction value of water level.SELECTED DRAWING: Figure 1

Description

TECHNICAL FIELD

[0001] Embodiments of the present invention relate to a water level prediction device, a water level prediction method, and a computer program. BACKGROUND ART

[0002] In recent years, localized heavy rainfall that falls locally and in a short period of time has increased, and inland flooding where water overflows inside urban areas has occurred frequently. In fact, when looking at flood damage (external water flooding, inland flooding) in terms of damage amount, the damage amount of inland flooding accounts for about half of the total national damage amount, exceeds 90% of the total damage amount in Tokyo Metropolis, and in urban areas where embankment maintenance has progressed relatively, inland flooding has become a new issue. To address this issue, research on real-time water level prediction based on meteorological data is being advanced.

[0003] For example, a flood prediction system is known that predicts the possibility of flooding and inundation depth in an area based on pre-stored surface data such as elevation in the prediction target area, the actual rainfall that fell in the several hours past from the current time, and rainfall data predicted to fall in the next several hours.

[0004] Furthermore, techniques have been proposed that use machine learning (multiple regression analysis, Kalman filter, neural network, random forest, support vector machine, etc.) to predict the water level rise rate from rainfall intensity and calculate the predicted time until inland flooding occurs. It has also been proposed to select a future inundation depth distribution map from a plurality of pre-calculated inundation depth distribution maps based on observed rainfall and predicted rainfall. PRIOR ART DOCUMENTS PATENT DOCUMENTS

[0005] PATENT DOCUMENT 1 Japanese Unexamined Patent Application Publication No. 2020-201704 PATENT DOCUMENT 2 Japanese Patent No. 6867638 SUMMARY OF THE INVENTION [Problems that the invention aims to solve]

[0006] Given the complexity of the phenomena leading up to flooding and the continuous changes in precipitation patterns due to climate change, there is a need for more accurate water level predictions to reduce damage from inland flooding. Furthermore, if water level predictions are unreliable, users may not trust the prediction results, hindering them from choosing appropriate disaster prevention actions. Therefore, it is necessary to provide supplementary information that allows users to understand and accept the water level prediction results, or to make their own informed decisions.

[0007] Embodiments of the present invention have been made in view of the above circumstances, and aim to provide a water level prediction device, a water level prediction method, and a computer program that can perform highly accurate predictions of water levels inside sewer pipes and enable users to take appropriate disaster prevention actions. [Means for solving the problem]

[0008] The water level prediction device according to this embodiment includes a water level learning and prediction unit that generates a water level prediction model and a water level feature model using input data including actual water level values ​​measured by sensors installed in sewer pipes, weather information and rainfall time series data, structural information data of the sewer pipes and topographic information data, calculates time series data of predicted water level values ​​using the water level prediction model, and calculates at least one water level feature of the input data using the water level feature model; a water level state cluster calculation unit that generates a plurality of water level state clusters by clustering the input data using the input data and the water level feature, and outputs a value indicating the water level state cluster of the input data; a prediction width learning and prediction unit that generates a prediction width model that calculates the prediction width of the predicted water level values ​​using the water level state cluster and the input data, and outputs the prediction width as time series data; and a VIL (Vertically Integrated The system includes a precipitation risk calculation unit that calculates precipitation risk using predicted values ​​of the increase in upper-air moisture content based on weather radar information such as Liquid, and a prediction result synthesis unit that calculates the maximum and minimum values ​​of the water level using time-series data of the predicted water level, the water level state cluster, the prediction range, and the precipitation risk, and outputs them together with the predicted water level. [Brief explanation of the drawing]

[0009] [Figure 1] Figure 1 is a schematic block diagram showing one example configuration of a water level prediction device according to the first embodiment. [Figure 2] Figure 2 is a schematic diagram showing an example of a predicted water level value obtained by a water level prediction model. [Figure 3] Figure 3 illustrates an example of a water level prediction model generated by machine learning. [Figure 4] Figure 4 is a schematic diagram showing an example of a water level state cluster obtained by clustering multiple water level features. [Figure 5] Figure 5 is a diagram illustrating an example of a prediction width model used in the prediction width learning prediction unit shown in Figure 1. [Figure 6]Figure 6 illustrates another example of the prediction width model used in the prediction width learning prediction unit shown in Figure 1. [Figure 7] Figure 7 illustrates an example of precipitation risk calculation using observational data from a three-dimensional weather radar. [Figure 8] Figure 8 illustrates an example of precipitation risk calculation using observational data from a three-dimensional weather radar. [Figure 9] Figure 9 is a diagram illustrating an example of a graph generated by the prediction result synthesis unit shown in Figure 1. [Figure 10] Figure 10 is a diagram illustrating an example of how the prediction result synthesis unit shown in Figure 1 calculates prediction results while considering precipitation risk. [Figure 11] Figure 11 is a schematic block diagram showing one example configuration of a water level prediction device according to the second embodiment. [Figure 12] Figure 12 is a diagram illustrating an example of the operation of the prediction result synthesis and update unit of the water level prediction device according to the second embodiment. [Figure 13] Figure 13 is a diagram illustrating another example of the operation of the prediction result synthesis and update unit of the water level prediction device according to the second embodiment. [Modes for carrying out the invention]

[0010] The water level prediction device, water level prediction method, and computer program of the embodiment will be described below with reference to the drawings. Figure 1 is a schematic block diagram showing one example configuration of a water level prediction device according to the first embodiment. The water level prediction device 10 of the first embodiment outputs the predicted water level result calculated using data input from the weather radar database DB1, rainfall database DB2, water level database DB3, sewer pipe structure information database DB4, and topographic information database DB5 to the display unit 20.

[0011] The weather radar database DB1 stores, for example, time-series data of weather information related to rain, snow, graupel, hail, and sleet observed by weather radar (for example, time-series data indicating position information of rain, snow, graupel, hail, and sleet, movement (wind direction) of rain, snow, graupel, hail, and sleet, intensity of rain, snow, graupel, hail, and sleet, etc.). The rainfall database DB2 stores time-series data of precipitation distribution per unit area (for example, a 1 km square) for a predetermined period (1 hour, 30 minutes, etc.), which is analyzed by combining, for example, observation data from weather radar and data measured by rain gauges.

[0012] The water level database DB3 stores, for example, measured value data of water levels measured by water level sensors installed in sewer conduits, in association with measurement times and measurement positions. The sewer conduit structure information database DB4 stores information on sewer conduit structures, such as the type of sewer pipes of the conduits, laying routes, gradients, lengths, depths, ages, and positions of manholes. The terrain information database DB5 stores, for example, information such as latitude, longitude, and altitude representing the position (undulation) of the ground surface, as well as information such as area and gradient.

[0013] The water level prediction device 10 includes, for example, at least one processor and a memory storing a program executed by the processor, and can realize various functions described below by software or by a combination of software and hardware.

[0014] The water level prediction device 10 of the first embodiment includes a precipitation risk calculation unit 11, a water level learning prediction unit 12, a prediction result synthesis unit 13, a water level state cluster calculation unit 14, and a prediction range learning prediction unit 15.

[0015] The water level learning and prediction unit 12 generates a water level prediction model that outputs predicted water level values ​​using input data that includes weather information, rainfall information, and time-series data (time-synchronized data with time stamps) of water level data in sewer pipes, supplied from the weather radar database DB1, rainfall database DB2, water level database DB3, sewer pipe structure information database DB4, and topographic information database DB5, as well as sewer pipe structure information and topographic information data.

[0016] Figure 2 is a schematic diagram showing an example of a predicted water level value obtained by a water level prediction model. A water level prediction model is a model that outputs predicted water levels as time series data from the time of prediction (the time of the latest data in the time series data) up to a predetermined time later. A water level prediction model may use a physical model or a machine learning model such as a neural network.

[0017] Figure 3 illustrates an example of a water level prediction model generated by machine learning. This section outlines an example of generating a water level prediction model using a neural network. The water level prediction model is generated by taking weather radar information, rainfall information, time-series data of water levels in sewer pipes (time-synchronized data with timestamps), sewer pipe structure information, and topographic information as input training data, and learning the features of the intermediate layer to output the measured water level.

[0018] Furthermore, the water level learning and prediction unit 12 calculates at least one water level feature by reducing the dimensionality of the input data using a water level feature model and outputs it to the water level state cluster calculation unit. The water level feature represents the characteristics of the input data and corresponds to at least one of several water level state clusters (such as a state where there is no precipitation and the water level is stable, a state where there is precipitation and the water level is rising rapidly, a state where there is a certain amount of precipitation and the water level is stable, or a state where precipitation has ended and the water level is falling) classified by the water level state cluster calculation unit 14 described later. Considering the sewer pipe structure, etc., it is thought that the prediction model and its prediction accuracy will differ depending on the water level state, so this is considered effective for switching models in the subsequent prediction range calculation.

[0019] The water level feature model may be a model that calculates at least one water level feature using multivariate analysis such as principal component analysis, or it may be a model that calculates water level features using one or more hidden layers of a neural network (for example, as shown in Figure 3).

[0020] When training the neural network, the water level learning and prediction unit 12 trains the water level prediction model using the same data (training data) as the data input to the model when calculating the predicted value, and the actual value data that represents the correct answer. Similarly, when generating a water level feature model using machine learning, the water level learning and prediction unit 12 trains the water level feature model using the same data as the data used when calculating the water level features, and the correct water level features.

[0021] The water level learning and prediction unit 12 outputs the calculated water level prediction value to the prediction result synthesis unit 13 and the prediction width learning and prediction unit 15, and also outputs the calculated water level feature quantity to the water level state cluster calculation unit 14.

[0022] The water level state cluster calculation unit 14 acquires multiple water level features, for example, that are output by the water level learning and prediction unit 12 when it performs machine learning. The water level state cluster calculation unit 14 generates multiple clusters by clustering the acquired multiple water level features, which are then designated as water level state clusters. If the water level learning and prediction unit 12 calculates water level features using multivariate analysis, it may generate multiple water level state clusters in advance by clustering multiple water level features calculated by multivariate analysis.

[0023] Figure 4 schematically shows an example of water level state clusters clustered using multiple water level features. Figure 4 shows an example in which multiple training data (at least a portion of the input data) are divided into water level state clusters A, B, and C, which are classified, for example, using a first water level feature and a second water level feature.

[0024] The water level state cluster calculation unit 14 acquires the water level features output by the water level learning and prediction unit 12 when calculating the predicted water level value, generates water level state clusters using the acquired water level features, and determines which water level state cluster the corresponding input data belongs to. Each cluster is expected to represent a state of water level, for example, a state where there is no precipitation and the water level is stable, a state where there is precipitation and the water level is rising rapidly, a state where there is a certain amount of precipitation and the water level is stable, a state where precipitation has ended and the water level is falling, etc. The water level state cluster calculation unit 14 outputs a value indicating the water level state cluster to which the input data belongs to the prediction width learning and prediction unit 15 and the prediction result synthesis unit 13.

[0025] The prediction range learning and prediction unit 15 uses the input data from the water level learning and prediction unit 12, the predicted water level, and the water level state cluster to calculate the variability or error of the prediction results output by the water level learning and prediction unit 12 as the prediction range.

[0026] The prediction width learning and prediction unit 15 may calculate the prediction width using a prediction width model created in advance by machine learning. The prediction width learning and prediction unit 15 may have one prediction width model or multiple prediction width models. For example, the prediction width learning and prediction unit 15 may generate prediction width models corresponding to each of multiple water level state clusters in advance by machine learning. The prediction width learning and prediction unit 15 can calculate the prediction width using the prediction width model corresponding to the value indicating the water level state cluster input from the water level state cluster calculation unit 14. Alternatively, the prediction width learning and prediction unit 15 may use the value indicating the water level state cluster as input data to generate a prediction width model that outputs a prediction width that takes the water level state cluster into consideration.

[0027] Figure 5 is a diagram illustrating an example of a prediction width model used in the prediction width learning prediction unit shown in Figure 1. The prediction range model may be generated by machine learning using the least-squares error value calculated during machine learning of the water level prediction model in the water level learning and prediction unit 12, the input data of the water level learning and prediction unit 12, the predicted water level value, and the water level state cluster. In this case, the output value of the prediction range model will be the least-squares error value.

[0028] Figure 6 illustrates another example of the prediction width model used in the prediction width learning prediction unit shown in Figure 1. The prediction range model may be generated by machine learning using the calculated value, the input data of the water level learning and prediction unit 12, the predicted water level, and the water level state cluster, after calculating how much the water level varies (e.g., the variance value) in training data of similar situations extracted using the nearest neighbor method or the like (e.g., situations where multiple input data values ​​are within a predetermined range set for each item). Alternatively, the prediction range learning and prediction unit 15 may generate a prediction range model that outputs a signed prediction range by machine learning using the error (signed) between the predicted water level and the measured value, the input data of the water level learning and prediction unit 12, the predicted water level, and the water level state cluster.

[0029] The precipitation risk calculation unit 11 acquires weather radar observation data from, for example, the weather radar database DB1. The weather radar observation data may be from a two-dimensional weather radar or from a three-dimensional weather radar. By using observation data from a three-dimensional weather radar, the precipitation risk calculation unit 11 can predict that even if there were no rain clouds before, rain clouds will rapidly develop and rain will suddenly begin.

[0030] Figures 7 and 8 illustrate an example of precipitation risk calculation using observational data from a three-dimensional weather radar. The precipitation risk calculation unit 11 calculates the amount of moisture in the upper atmosphere from observation data obtained from a three-dimensional weather radar. The precipitation risk calculation unit 11 calculates the amount of moisture in a cylindrical region at a predetermined altitude h having a base of unit area, for example, as shown in Figure 7. Using the calculated amount of moisture in the upper atmosphere, the precipitation risk calculation unit 11 generates a moisture content prediction model that outputs a predicted value for the increase in moisture content. The moisture content prediction model may be a physical model or a model generated by machine learning.

[0031] The precipitation risk calculation unit 11 converts the amount of moisture in the upper atmosphere and the predicted value calculated by the moisture amount prediction model into a risk amount for whether precipitation will occur during the predicted time. The precipitation risk calculation unit 11 may also use the value obtained by converting the predicted moisture amount to rainfall intensity (mm / h) as the risk amount. However, if precipitation is already predicted in the precipitation forecast, for example, the precipitation risk calculation unit 11 will use the value obtained by subtracting the amount of precipitation from the precipitation forecast from the rainfall intensity based on the predicted moisture amount, or 0, as the risk amount. The precipitation risk calculation unit 11 calculates the risk amount for sudden precipitation that is not predicted in the precipitation forecast. The precipitation risk calculation unit 11 outputs the calculated risk amount to the prediction result synthesis unit 13.

[0032] The prediction result synthesis unit 13 generates a graph using the time-series data of predicted water levels from the water level learning and prediction unit 12, the values ​​corresponding to water level state clusters from the water level state cluster calculation unit 14, the prediction range from the prediction range learning and prediction unit 15, and the risk amount from the precipitation risk calculation unit 11. The prediction result synthesis unit 13 outputs the generated prediction result graph to the display unit 20.

[0033] Figure 9 is a diagram illustrating an example of a graph generated by the prediction result synthesis unit shown in Figure 1. The prediction result synthesis unit 13 calculates the maximum and minimum values ​​of the future water level using, for example, the prediction range, and generates a single graph that shows the predicted water level, the predicted maximum water level, and the predicted minimum water level.

[0034] For example, if the output from the prediction unit that learns and predicts based on water level features is the least squares error, the prediction result synthesis unit 13 calculates the maximum value by adding the prediction range to the predicted water level value, and calculates the minimum value by subtracting the prediction range from the predicted water level value.

[0035] Furthermore, if, for example, the prediction range is the variance value, the prediction result synthesis unit 13 can generate a probability distribution using the average value of the prediction values ​​at each time point, and calculate the maximum and minimum values ​​that represent the desired probability.

[0036] Furthermore, the prediction result synthesis unit 13 may change the method of synthesizing the prediction results and the method of calculating the maximum and minimum water levels according to the water level state cluster. For example, when calculating prediction results using data classified as a water level state cluster with low precipitation, the prediction result synthesis unit 13 can change the values ​​of the probabilities used to determine the maximum and minimum values, which are the desired probabilities, from the generated probability distribution, depending on the water level state cluster. Furthermore, the prediction result synthesis unit 13 may generate a graph of the prediction results by further considering the precipitation risk (risk amount).

[0037] Figure 10 is a diagram illustrating an example of how the prediction result synthesis unit shown in Figure 1 calculates prediction results while considering precipitation risk. Here, for example, when the predicted water level is a constant value, an example of the maximum water level, minimum water level, and time function of the risk quantity is shown. The risk quantity function may be linear, nonlinear, or a function generated by machine learning. The prediction result synthesis unit 13 adds the value calculated by the risk quantity function to the maximum value.

[0038] The prediction result synthesis unit 13 only needs to generate at least one graph. The prediction result synthesis unit 13 does not need to generate a single graph that synthesizes all the calculated values ​​(maximum value, predicted value, minimum value, etc.). It may generate separate graphs for each of the multiple calculated values, or it may generate a graph that combines some of the multiple calculated values.

[0039] As described above, the water level prediction device 10 of this embodiment can not only calculate a predicted value for the water level in sewer pipes according to the precipitation conditions, but can also output the error range of the predicted value. Precipitation conditions can be defined by calculating a dimensionally reduced water level feature based on the input data for water level prediction. The way in which the water level in sewer pipes changes varies greatly depending on the situation, such as when the water level rises, or the operating status of the pumps at the sewage treatment plant, so the error trend of the water level prediction also varies greatly. Therefore, by automatically extracting the precipitation conditions and outputting the predicted value and prediction range (error range) accordingly, not only is the prediction accuracy improved, but it also becomes possible to provide the user with information to make their own judgments and encourage the user to take appropriate disaster prevention actions.

[0040] In other words, the water level prediction device 10 of this embodiment not only improves the accuracy of water level prediction, but also outputs a prediction error range according to the precipitation conditions, thereby showing the user how much the water level prediction results may deviate and encouraging the user to take appropriate disaster prevention actions.

[0041] Next, an example of the operation of the water level prediction device of this embodiment will be described. In the water level prediction device 10 of this embodiment, first, a water level prediction model, a water level feature model, multiple water level state clusters, and a prediction width model are generated. The water level prediction device 10 of this embodiment periodically acquires input data that combines time-synchronized time-series data with timestamps (such as weather radar, rainfall, and water level data in sewer pipes) with sewer pipe structure and topographic information. The water level prediction device 10 may acquire sewer pipe structure data and topographic information only when they are updated.

[0042] The water level learning and prediction unit 12 generates a water level prediction model using machine learning based on the input data. The water level learning and prediction unit 12 also generates a water level feature model that calculates water level features by reducing the dimensionality of the input data. The water level learning and prediction unit 12 may, for example, output the values ​​of the intermediate layer (features) obtained during the training of the water level prediction model as water level features.

[0043] The water level state cluster calculation unit 14 performs clustering using multiple water level features output from the water level learning and prediction unit 12 to generate multiple water level state clusters. The prediction range learning and prediction unit 15 generates at least one prediction range model by machine learning using the input data, the predicted water level values ​​output from the water level prediction model, the actual water level values, and the corresponding water level state clusters. The prediction range learning and prediction unit 15 may generate multiple prediction range models corresponding to each of the multiple water level state clusters.

[0044] Next, the water level prediction device 10 performs water level prediction using the generated water level prediction model, water level feature model, water level state cluster, and prediction width model. The water level learning and prediction unit 12 inputs the input data into the water level prediction model and calculates a predicted water level value. The water level learning and prediction unit 12 also uses a water level feature model to calculate water level features corresponding to the input data.

[0045] The water level state cluster calculation unit 14 determines which water level state cluster the water level features calculated by the water level learning and prediction unit 12 belong to, and outputs a value indicating the determined water level state cluster. The prediction width learning and prediction unit 15 calculates the prediction width using a prediction width model, based on the input data, the predicted water level calculated by the water level learning and prediction unit 12, and the values ​​indicating the water level state cluster.

[0046] The precipitation risk calculation unit 11 calculates a predicted value for the amount of moisture in the upper atmosphere above the target area, for example, using observation data from a three-dimensional weather radar, and calculates the risk of sudden precipitation (risk amount) that could not be predicted by the precipitation forecast, based on the predicted value of the amount of moisture.

[0047] The prediction result synthesis unit 13 uses the predicted water level, water level state cluster, prediction range (error), and precipitation risk (risk amount) to generate a graph showing the predicted water level, maximum value, and minimum value in time series. According to the water level prediction device 10 of this embodiment, future precipitation conditions can be estimated using a model generated from input data, and predicted water levels corresponding to these conditions can be output, thereby improving prediction accuracy. Furthermore, according to the water level prediction device 10 of this embodiment, by presenting the maximum and minimum water levels based on the prediction range of the predicted values ​​to the user, it is possible to show the user how much the prediction results may deviate and encourage the user to take appropriate disaster prevention actions.

[0048] In other words, according to this embodiment, it is possible to provide a water level prediction device, a water level prediction method, and a computer program that can perform highly accurate predictions of water levels inside sewer pipes and enable users to take appropriate disaster prevention actions.

[0049] Next, the water level prediction device, water level prediction method, and computer program of the second embodiment will be described in detail with reference to the drawings. In the following description, components similar to those in the first embodiment described above will be denoted by the same reference numerals and their descriptions will be omitted.

[0050] Figure 11 is a schematic block diagram showing one example configuration of a water level prediction device according to the second embodiment. The water level prediction device 10 of this embodiment differs from the water level prediction device 10 of the first embodiment described above in that it includes a prediction result synthesis update unit 16 instead of a prediction result synthesis unit 13, and is configured to correct the prediction value by comparing the predicted value and the actual value of the water level at the same time (or the same period).

[0051] The prediction result synthesis and update unit 16 obtains actual water level values ​​from the input data and corrects the prediction values ​​by comparing them with the actual values. For example, if the difference between the average value of the prediction values ​​and the average value of the actual values ​​over a predetermined period exceeds a first threshold, the prediction result synthesis and update unit 16 may correct the prediction values ​​so that the difference becomes smaller.

[0052] Figure 12 is a diagram illustrating an example of the operation of the prediction result synthesis and update unit of the water level prediction device according to the second embodiment. The prediction result synthesis and update unit 16 calculates, for example, the difference between the average value of the predicted values ​​over a predetermined period and the average value of the actual values ​​over the same period. When the calculated difference is greater than the first threshold, the prediction result synthesis and update unit 16 determines the slope of a linear function whose integral value (area) is equal to the difference between the average value of the predicted values ​​and the average value of the actual values, and adds the value calculated by this linear function to the prediction result to perform a correction.

[0053] Furthermore, the prediction result synthesis and update unit 16 does not need to update the correction amount for each input data. For example, the prediction result synthesis and update unit 16 may calculate a correction value for each water level state cluster, store the calculated correction value, and use the same correction value to correct the prediction value at multiple time points. By using a common correction value for multiple prediction values, stable prediction results can be provided to the user. Alternatively, the prediction result synthesis and update unit 16 may use multiple correction values ​​calculated for each input data and correct the prediction value with a weighted value obtained by combining the multiple correction values. The prediction result synthesis and update unit 16 may also compare the prediction error with the actual value and correct the predicted value.

[0054] Figure 13 is a diagram illustrating another example of the operation of the prediction result synthesis and update unit of the water level prediction device according to the second embodiment. For example, the prediction result synthesis and update unit 16 may multiply the pre-correction maximum value and pre-correction minimum value by a coefficient so that all actual values ​​are greater than or equal to the minimum value and less than or equal to the maximum value. However, correcting the prediction error corresponding to each predicted value may make the prediction result unstable. Therefore, the prediction result synthesis and update unit 16 may store a correction coefficient for each water level state cluster and use the same correction coefficient for multiple prediction errors. In this case, the prediction result synthesis and update unit 16 may calculate the correction coefficient at predetermined time intervals and update the value. Alternatively, the prediction result synthesis and update unit 16 may use multiple coefficients calculated for each prediction error and correct the prediction error with a weighted value obtained by assigning weights to multiple coefficients. The prediction result synthesis and update unit 16 may also use correction and update methods using reinforcement learning or the like.

[0055] Furthermore, as the prediction result synthesis update unit 16 continues learning, if at least one of the correction amount or correction coefficient exceeds its respective threshold, it determines that the output value of the prediction model and the actual value have diverged significantly, and outputs an update instruction (retraining instruction) for the water level prediction model, water level feature model, and prediction width model. This enables predictions that respond to climate change, and an improvement in prediction accuracy can be expected.

[0056] Furthermore, the water level prediction device 10 of this embodiment learns how to respond to climate change by comparing predicted water levels with actual measured values ​​and correcting the predicted values. For example, the water level prediction device 10 may learn the amount of bias (correction value or correction coefficient) to be added to the prediction result.

[0057] Next, an example of the operation of the water level prediction device 10 of this embodiment will be described. Here, we will describe the operation that differs from the first embodiment described above, and omit the description of the common operation. The prediction result synthesis and update unit 16 calculates, for example, the difference between the average value of the predicted water level and the average value of the actual water level over a predetermined period of time, determines the slope of a linear function whose integral value (area) is equal to the calculated difference, and uses the value calculated by this linear function as the correction value. The prediction result synthesis and update unit 16 stores the correction value and uses the predicted value corrected by the correction value in the prediction result displayed next.

[0058] Furthermore, the prediction result synthesis and update unit 16 corrects the pre-correction maximum value and pre-correction minimum value by multiplying them by a correction coefficient so that all actual values ​​are greater than or equal to the minimum value and less than or equal to the maximum value. The prediction result synthesis and update unit 16 stores the correction coefficient and uses the prediction error corrected by the correction coefficient in the prediction result displayed next.

[0059] The prediction result synthesis and update unit 16 updates the correction value and correction coefficient at a predetermined period. The prediction result synthesis and update unit 16 determines whether the correction value exceeds the first threshold, and if so, outputs an update instruction (retraining instruction) to the water level learning and prediction unit 12 to update (retrain) the prediction model. The prediction result synthesis and update unit 16 also determines whether the correction coefficient exceeds the second threshold, and if so, outputs an update instruction (retraining instruction) to the prediction width learning and prediction unit 15 to update (retrain) the prediction width model. The first and second thresholds may be pre-set values ​​or dynamically determined values.

[0060] When the water level learning and prediction unit 12 receives an update instruction (retraining instruction), it updates (retrains) the water level prediction model and the water level feature model. When the prediction width learning and prediction unit 15 receives an update instruction (retraining instruction), it updates (retrains) the prediction width model.

[0061] According to the water level prediction device 10 of the second embodiment described above, climate change can be addressed and prediction accuracy can be improved by correcting the water level prediction results. Furthermore, the water level prediction device 10 of this embodiment can be expected to achieve stable improvement in prediction accuracy by updating (retraining) the prediction model at an appropriate timing based on fluctuations in the correction amount. In other words, according to this embodiment, it is possible to provide a water level prediction device, a water level prediction method, and a computer program that can perform highly accurate predictions of water levels inside sewer pipes and enable users to take appropriate disaster prevention actions.

[0062] The program according to this embodiment may be transferred while stored on an electronic device, or it may be transferred while not stored on an electronic device. In the latter case, the program may be transferred via a network, or it may be transferred while stored on a storage medium. The storage medium is a non-temporary tangible medium. The storage medium is a computer-readable medium. The storage medium can be any medium that is capable of storing a program and is readable by a computer, such as a CD-ROM or memory card, and its form is not limited.

[0063] While several embodiments of the present invention have been described, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These novel embodiments can be carried out in a variety of other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims of the invention and its equivalents. [Explanation of Symbols]

[0064] 10...Water level prediction device, 11...Precipitation risk calculation unit, 12...Water level learning prediction unit, 13...Prediction result synthesis unit, 14...Water level state cluster calculation unit, 15...Prediction range learning prediction unit, 16...Prediction result synthesis update unit, 20...Display unit, DB1...Weather radar database, DB2...Rainfall database, DB3...Water level database, DB4...Sewer pipe structure information database, DB5...Topographic information database

Claims

1. A water level learning and prediction unit generates a water level prediction model and a water level feature model using input data including actual water level measurements taken by sensors installed in sewer pipes, time-series data of weather information and rainfall, data of structural information of the sewer pipes, and data of topographic information; calculates time-series data of predicted water level values ​​using the water level prediction model; and calculates at least one water level feature of the input data using the water level feature model. A water level state cluster calculation unit generates a plurality of water level state clusters by clustering the input data using the input data and the water level feature quantities, and outputs a value indicating the water level state cluster of the input data. A prediction width learning and prediction unit generates a prediction width model that calculates the prediction width of the predicted water level using the water level state cluster and the input data, and outputs the prediction width as time series data. A precipitation risk calculation unit that calculates precipitation risk using predicted values ​​of the increase in upper-air moisture content based on weather radar information, A water level prediction device having a prediction result synthesis unit that calculates the maximum and minimum values ​​of the water level using the time-series data of the predicted water level, the water level state cluster, the prediction range, and the precipitation risk, and outputs them together with the predicted water level.

2. The water level prediction device according to claim 1, wherein the prediction width learning prediction unit generates a plurality of prediction width models corresponding to each of the plurality of water level state clusters by machine learning using the input data, and calculates the prediction width by switching the prediction width model for each of the water level state clusters.

3. The water level prediction device according to claim 1, wherein the prediction range is the least squares error calculated when the water level prediction model is created in the water level learning and prediction unit, or the variance of the neighboring data extracted by the nearest neighbor method.

4. The precipitation risk calculation unit calculates the amount of moisture in the upper atmosphere from observation data of a three-dimensional weather radar, and calculates the precipitation risk based on the amount of moisture in the upper atmosphere, according to claim 1, water level prediction device.

5. The water level prediction device according to claim 3, wherein the prediction result synthesis unit calculates the maximum value of the water level by adding the least squares error to the predicted value of the water level when the prediction range is the least squares error, calculates the minimum value of the water level by subtracting the least squares error from the predicted value of the water level, and when the prediction range is the variance value, generates a probability distribution using the variance value with the predicted value of the water level at each time as the average value, and sets the value that gives the desired probability as the maximum value of the water level and the minimum value of the water level.

6. The water level prediction device according to claim 1, wherein the prediction result synthesis unit changes the method for calculating the maximum value of the water level and the method for calculating the minimum value of the water level according to the water level state cluster.

7. The water level prediction device according to claim 1, wherein instead of the prediction result synthesis unit, it comprises a prediction result synthesis update unit that calculates the maximum and minimum values ​​of the water level using the time-series data of the predicted water level, the water level state cluster, the prediction range, and the precipitation risk, and outputs them together with the predicted water level, and corrects the predicted water level by a correction amount based on the result of comparing the predicted water level with the actual water level at the same time.

8. The water level prediction device according to claim 7, wherein the prediction result synthesis and update unit calculates the correction amount using a function such that the integral value of the difference between the average value of the predicted water level and the average value of the actual water level over a predetermined period is equal, and corrects the predicted water level using the correction amount.

9. The water level prediction device according to claim 7, wherein the prediction result synthesis and update unit corrects the minimum water level and the maximum water level by multiplying them by a coefficient so that at all time points in the time series data to be output, the actual value is greater than or equal to the minimum water level and less than or equal to the maximum water level.

10. The water level prediction device according to claim 7, wherein the prediction result synthesis update unit outputs an update instruction for the water level prediction model to the water level learning prediction unit when the correction amount exceeds a threshold.

11. The water level prediction device according to claim 9, wherein the prediction result synthesis update unit outputs an update instruction for the prediction width model to the prediction width learning prediction unit when the coefficient exceeds a threshold.

12. Using input data including actual water level measurements taken by sensors installed in sewer pipes, time-series data of weather information and rainfall, structural information data of the sewer pipes, and topographic information data, a water level prediction model and a water level feature model are generated. The water level prediction model calculates time-series data of predicted water levels, and the water level feature model calculates at least one water level feature of the input data. Using the input data and the water level features, a plurality of water level state clusters are generated by clustering the input data, and a value representing the water level state cluster of the input data is output. A prediction range model is generated using the water level state cluster and the input data to calculate the prediction range of the predicted water level, and the prediction range is output as time series data. Precipitation risk is calculated using predicted values ​​for the increase in upper-air moisture content based on weather radar information. A water level prediction method that calculates the maximum and minimum values ​​of the water level using time-series data of predicted water levels, water level state clusters, prediction ranges, and precipitation risk, and outputs them together with the predicted water levels.

13. A computer program that causes a computer to perform the method described in claim 12.

Images