Water level prediction system, learning device, water level prediction method, water level prediction program, learning method, and learning program

Abstract

To obtain a water level prediction system that can realize increase in accuracy of water level prediction.SOLUTION: A water level prediction system 1 includes: a measurement data acquisition unit 31 which acquires measurement data 23 on a river including measurement values of water levels at a plurality of locations in the river; a surrounding environment data acquisition unit 41 which acquires surrounding environment data including the operating condition of a river facility 200 installed in a basin of the river; a measurement data processing unit 12 which removes abnormal data included in the measurement data 23 acquired by the measurement data acquisition unit 31; and a water level prediction unit 11 which performs water level prediction of the river on the basis of a learned model 25 generated by executing learning to predict a water level change amount based on the measurement data 23 and the surrounding environment data, processed measurement data 24 being measurement data 23 obtained by removing the abnormal data by the measurement data processing unit 12, and the surrounding environment data acquired by the surrounding environment data acquisition unit 41.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present disclosure relates to a water level prediction system, a learning device, a water level prediction method, a water level prediction program, a learning method, and a learning program that predict the water level of a river or the like for flood prevention. [Background technology]

[0002] In recent years, as the risk of flooding increases due to climate change, there is a need for rapid and appropriate flood prevention activities against sudden rises in river water levels. To achieve rapid and appropriate flood prevention activities, it is important to predict as accurately as possible the future state of the area subject to flood prevention activities.

[0003] For example, Patent Document 1 describes a technology that uses river information, such as water levels and rainfall measured in rivers, and facility operation information, such as the operating status of pumps at drainage pumping stations and the open / closed status of floodgates, to predict future water levels through modeling techniques or machine learning. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] JP 2020-133275 A Summary of the Invention [Problem to be solved by the invention]

[0005] If abnormal information is included in the information used for water level prediction, the accuracy of the water level prediction may decrease, which may hinder appropriate flood prevention activities. In water level prediction, the current or past actual water level is generally used, and the actual water level is measured using a water level gauge. However, for example, if aquatic plants approach the water level gauge, the aquatic plants may be mistaken for water, and the measurement value output from the water level gauge may have a large error. In addition, if there is fluctuation on the water surface, even if the water level is the same, the measurement value differs depending on the timing of the measurement, making it difficult to measure the water level accurately. When water level prediction is more important for flood prevention activities, specifically, when there is a very large amount of rainfall and the water level of the river tends to rise rapidly, it is expected that the water surface fluctuation will become large and the error between the actual water level and the measurement value will become large. In particular, when strong winds occur in addition to increased rainfall due to a typhoon, the water surface fluctuation may become even larger, and the error may increase.

[0006] The present disclosure has been made in consideration of the above, and aims to provide a water level prediction system that can achieve high accuracy in water level prediction. [Means for solving the problem]

[0007] In order to solve the above-mentioned problems and achieve the objectives, the water level prediction system disclosed herein is characterized by comprising a measurement data acquisition unit that acquires river measurement data including water level measurement values ​​at multiple points on the river, an ambient environment data acquisition unit that acquires ambient environment data including the operating status of river facilities installed in the river basin, a measurement data processing unit that removes abnormal data included in the measurement data acquired by the measurement data acquisition unit, and a water level prediction unit that predicts the water level of the river based on a learned model generated by performing learning to predict the amount of water level change based on the measurement data and the ambient environment data, processed measurement data, which is the measurement data from which abnormal data has been removed by the measurement data processing unit, and the ambient environment data acquired by the ambient environment data acquisition unit. Effect of the Invention

[0008] According to the present disclosure, it is possible to obtain a water level prediction system capable of achieving high accuracy in water level prediction. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram showing a configuration example of a flood defense system including a water level prediction system according to a first embodiment. [Diagram 2] FIG. 1 is a diagram showing an example of a functional block configuration of a water level prediction system according to a first embodiment. [Diagram 3] FIG. 1 is a diagram showing the operation of the water level prediction system according to the first embodiment. [Figure 4] 1 is a flowchart showing an example of an operation of a measurement data processing unit of the water level prediction system according to the first embodiment. [Diagram 5] The first figure explains how to select learning data for each prediction target location. [Figure 6] The second figure explains how to select learning data for each prediction target location. [Figure 7] The third figure explains how to select learning data for each prediction target location. [Figure 8] FIG. 1 is a diagram illustrating an example of a learning device that generates a trained model used by a water level prediction unit of the water level prediction system according to the first embodiment. [Figure 9] Diagram to explain neural networks [Figure 10] 1 is a flowchart showing an example of an operation of a learning device. [Figure 11] FIG. 1 is a diagram showing a configuration example of an inference device that realizes a water level prediction unit of the water level prediction system according to the first embodiment. [Figure 12] 1 is a flowchart showing an example of an operation of an inference device. [Figure 13] 1 is a flowchart showing an example of an operation of the water level prediction system according to the first embodiment. [Figure 14] A diagram showing an example of the hardware configuration of a server that realizes the water level prediction unit and the measurement data processing unit of the water level prediction system. [Figure 15]FIG. 13 is a diagram showing an example of a functional block configuration of a water level prediction system according to a second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Below, a water level prediction system, a learning device, a water level prediction method, a water level prediction program, a learning method, and a learning program according to embodiments of the present disclosure are described in detail with reference to the drawings.

[0011] Embodiment 1 FIG. 1 is a diagram showing a configuration example of a flood defense system including a water level prediction system 1 according to the first embodiment. The flood defense system shown in FIG. 1 includes a water level prediction system 1 including a server 10, a database 20, a gateway device 30, and a linkage device 40 provided on a cloud, a gateway device 110, a slave station 120, a machine-side operation panel 130, an operation target 140, and a measurement device 150 provided in each of a plurality of fields 100 such as an observation station, a river facility 200 such as a water gate, a drainage pumping station, a regulation pond, and a dam provided in a river basin, and another system 300 that predicts future rainfall. The operation target 140 is a gate, a valve, and the like provided in the field 100. The measurement device 150 is a water level gauge, a rain gauge, a turbidity meter, and the like provided in a river. Each measurement device 150 repeatedly performs measurement at a predetermined period. The measurement results by each measuring device 150 include identification information of the measuring device 150 that performed the measurement, information on the time when the measurement was performed (hereinafter may be referred to as the measurement time), the measurement value, etc. For example, if the measuring device 150 is a water level gauge, the measurement results by the measuring device 150 include identification information of the water level gauge, the time when the water level was measured, the measurement value of the water level, etc.

[0012] The gateway device 30 of the water level prediction system 1 is connected to a telecommunications carrier network having wireless access lines, and acquires the measurement results from the measuring device 150 via the gateway device 110 and the slave station 120 in the field 100. In the following description, the measurement results from the measuring device 150 may be referred to as measurement data. The gateway device 30 may acquire information on the state of the operation target object 140 from the machine-side operation panel 130. The information acquired by the gateway device 30 is registered in the database 20. Although the illustration of the master station paired with the slave station 120 is omitted in FIG. 1, the master station is installed, for example, in a management room that remotely monitors the overall state of the river.

[0013] The linking device 40 of the water level prediction system 1 performs processing for operating in cooperation with the other system 300 and the river facility 200. The linking device 40 is connected to an Internet line, and acquires, for example, river facility operation information indicating the operation status of the river facility 200 from the river facility 200 and the other system 300, other system information which is information generated by the other system 300, and the like. Examples of the other system information are a predicted value of future rainfall, a predicted value of tide level, and the like. When the water level prediction of a point different from the point where the water level prediction system 1 predicts the water level is performed in the other system 300, the water level prediction value in the other system 300 may be included in the other system information. In the following description, the river facility operation information acquired by the linking device 40 from the river facility 200 and the other system information acquired from the other system 300 may be collectively referred to as surrounding environment data. The information acquired by the linking device 40 is registered in the database 20.

[0014] A functional block configuration of the water level prediction system 1 is shown in Fig. 2. Fig. 2 is a diagram showing an example of a functional block configuration of the water level prediction system 1 according to the first embodiment.

[0015] As shown in FIG. 2, the water level prediction system 1 includes a water level prediction unit 11 and a measurement data processing unit 12 in a server 10, a measurement data acquisition unit 31 that acquires measurement data from a measurement device 150, and an ambient environment data acquisition unit 41 that acquires ambient environment data from a river facility 200 and another system 300. The water level prediction unit 11 predicts the water level at a specified point in the river based on various data registered in a database 20. There are multiple points at which the water level prediction unit 11 predicts the water level. The measurement data processing unit 12 executes a process of removing abnormal values ​​from the measurement values ​​included in the measurement data acquired by the measurement data acquisition unit 31 to generate processed measurement data. The measurement data acquisition unit 31 is realized by the gateway device 30 shown in FIG. 1, and the ambient environment data acquisition unit 41 is realized by the linkage device 40 shown in FIG. 1.

[0016] The database 20 holds other system information 21 and river facility operation information 22 acquired by the surrounding environment data acquisition unit 41, measurement data 23 acquired by the measurement data acquisition unit 31, processed measurement data 24 generated by the measurement data processing unit 12, a trained model 25 used by the water level prediction unit 11 when predicting the water level, and a water level prediction value 26 indicating the water level prediction result by the water level prediction unit 11. Note that there are multiple pieces of each of the measurement data 23, the processed measurement data 24, the trained model 25, and the water level prediction value 26.

[0017] Fig. 3 is a diagram showing the operation of the water level prediction system 1 according to the embodiment 1. As shown in Fig. 3, other system information 21 and river facility operation information 22 constituting the surrounding environment data 28, processed measurement data 24, and a trained model 25 are input to the water level prediction unit 11.

[0018] The trained model 25 is a model for inferring the amount of change in water level over a certain period of time from the other system information 21, the river facility operation information 22, and the processed measurement data 24. The trained model 25 is generated, for example, by an external learning device. Details of the trained model 25 will be described later.

[0019] The water level prediction unit 11 inputs other system information 21, river facility operation information 22, and processed measurement data 24 to the trained model 25 as inference data, and calculates a predicted value of the water level after a certain time has passed based on the predicted value of the water level change output from the trained model 25. The water level prediction unit 11 outputs the calculated predicted value of the water level as a water level prediction value 26.

[0020] As described above, the processed measurement data 24 is generated by the measurement data processing unit 12. The operation of the measurement data processing unit 12 will be described. FIG. 4 is a flowchart showing an example of the operation of the measurement data processing unit 12 of the water level prediction system 1 according to the first embodiment. The measurement data processing unit 12 generates the processed measurement data 24 by repeatedly executing the process according to the flowchart shown in FIG. 4 a predetermined number of times. The number of times that the measurement data processing unit 12 repeats the process is determined by the number of measurement data 23 registered in the database 20.

[0021] As shown in FIG. 4, the measurement data processing unit 12 first acquires the measurement data 23 from the database 20 (step S11). In this step S11, the measurement data processing unit 12 acquires two pieces of measurement data 23 with consecutive measurement times. For example, when the measurement by the measuring device 150 is performed at a 5-minute cycle, the measurement data processing unit 12 acquires the measurement data 23 at a certain measurement time (hereinafter referred to as the first measurement data) and the measurement data 23 with a measurement time 5 minutes after the measurement time of the first measurement data (hereinafter referred to as the second measurement data). For example, the measurement data processing unit 12 acquires the measurement data 23 with a measurement time of 10:00 as the first measurement data, and acquires the measurement data 23 with a measurement time of 10:05 as the second measurement data. The measurement data processing unit 12 then calculates the change width of the measurement data (step S12). In step S12, the measurement data processing unit 12 calculates the difference between the measurement value included in the first measurement data and the measurement value included in the second measurement data, and sets this as the variation range. The measurement data processing unit 12 then compares the calculated variation range with a predetermined abnormality determination threshold (step S13). The abnormality determination threshold is determined by a flood prevention work staff member based on experience, for example, with reference to the period at which the measuring device 150 performs measurements, past measurement results, etc.

[0022] If the change range is greater than the abnormality determination threshold (step S13: Yes), the measurement data processing unit 12 determines that the measurement data 23 taken as the second measurement data is abnormal data and discards this measurement data 23 (step S14).If the change range is equal to or less than the abnormality determination threshold (step S13: No), the measurement data processing unit 12 stores the measurement data 23 taken as the second measurement data in the database 20 as processed measurement data 24 (step S15).

[0023] The measurement data processing unit 12 executes the above-mentioned steps S11 to S15 (hereinafter referred to as data processing) on ​​all the measurement data 23 registered in the database 20, and discards the measurement data 23 that is deemed to be abnormal. The data processing may be executed every time the measurement data acquisition unit 31 acquires new measurement data 23, or may be executed periodically. For example, the measurement data processing unit 12 may execute the data processing at 30-minute intervals. In this case, the measurement data processing unit 12 executes the data processing on the measurement data 23 acquired by the measurement data acquisition unit 31 after the previous data processing is executed, that is, the measurement data 23 acquired within the past 30 minutes. Furthermore, when the measurement data processing unit 12 executes step S14 to discard the measurement data 23, it changes the abnormality determination threshold value used in step S13 of the data processing to be executed next. For example, if measurements are performed at five-minute intervals, and the measurement data 23 at the measurement time t is determined to be normal, and the measurement data 23 at the measurement time 5 minutes after the measurement time t is determined to be abnormal, in data processing processing targeting the measurement data at the measurement time 10 minutes after the measurement time t, the measurement data processing unit 12 calculates the change width between the measurement data 23 at the measurement time t and the measurement data 23 at the measurement time 10 minutes after the measurement time t in step S12, and performs judgment in step S13 using a threshold value that is twice the abnormality judgment threshold value used in judging the measurement data 23 at the measurement time 5 minutes after the measurement time t. Alternatively, the abnormality judgment threshold value may not be changed, and half the change width calculated in step S12 may be compared with the abnormality judgment threshold value in step S13.

[0024] By performing the above-mentioned data processing process, the measurement data processing unit 12 removes measurement data 23 that contains abnormal measurement values. Even if, for example, water plants or the like approach the water level meter and increase the error in the measurement value by the water level meter, or if large fluctuations occur on the water surface and increase the error in the measurement value, it is possible to predict the water level using only measurement values ​​with small errors, thereby preventing a decrease in prediction accuracy.

[0025] Generally, the period in which the measuring device 150 performs measurements (hereinafter referred to as the measurement period) is different from the period in which the water level prediction unit 11 predicts the water level (hereinafter referred to as the prediction period). For this reason, when the prediction period is longer than the measurement period, the water level prediction unit 11 uses the average value of the measurement values ​​in the prediction period to predict the water level. For example, when the measurement period of the measuring device 150 is 1 minute and the prediction period of the water level prediction unit 11 is 10 minutes, the water level prediction unit 11 predicts the water level using the 10-minute average value of the measurement values.

[0026] In addition, the measurement data acquisition unit 31 acquires measurement data from a large number of measuring instruments 150, but there may be cases where the format of the measurement values ​​included in the measurement data is not unified. For example, there may be cases where the unit of the measurement value acquired from one water level gauge is "m" and the unit of the measurement value acquired from another water level gauge is "cm". There may also be cases where the measurement value acquired from one water level gauge indicates the height of the water surface from the river bottom, and the measurement value acquired from the other water level gauge indicates the elevation of the water surface. If the formats of the measurement values ​​are different, it becomes difficult to handle them in subsequent processing and in the work of flood prevention personnel comparing the measurement values. For this reason, the measurement data processing unit 12 executes a process of processing the measurement values ​​to unify the format when the format of the measurement values ​​is not unified. For example, the format is unified before calculating the change range in the above step S12. Although an example of the measurement value being the water level of a river has been described, the same applies when the measurement value is the amount of rainfall or the turbidity of a river. In addition, the measurement data processing unit 12 also unifies the types of information included in the measurement data 23 when the types of information are different or when the order of the information is different. This unifies the format of the processed measurement data 24 generated by the measurement data processing unit 12, making it easier to handle the data. For example, it becomes easier for a person in charge of operations to set the anomaly determination threshold used in step S13 above, and the setting work can be performed efficiently.

[0027] Returning to the explanation of the operation of the water level prediction system 1, in the water level prediction system 1, the water level prediction unit 11 repeatedly predicts the water level after a certain period of time has elapsed while changing the reference time.

[0028] For example, if the trained model 25 is configured to output a predicted value of the amount of water level change over the course of 30 minutes after inference data is input, the water level prediction unit 11 first sets the current time as a reference time and calculates a predicted value of the water level 30 minutes after the reference time, i.e., a predicted value of the water level 30 minutes after the current time. Specifically, the water level prediction unit 11 obtains a predicted value of the amount of water level change over the course of 30 minutes from the reference time using the inference data at the current time and the trained model 25, and calculates a predicted value of the water level 30 minutes after the current time by adding the obtained predicted value of the amount of water level change to the water level at the current time.

[0029] Next, the water level prediction unit 11 sets 30 minutes after the current time as the reference time, and calculates the predicted water level value 30 minutes after the reference time, i.e., the predicted water level value 60 minutes after the current time. The inference data to be input to the trained model 25 at this time is the inference data 30 minutes after the current time. Specifically, the inference data 30 minutes after the current time is other system information 21 corresponding to 30 minutes after the current time (predicted value of rainfall 30 minutes after the current time, predicted value of tide level, etc.), river facility operation information 22 for the current time, and processed measurement data 24. Since river facility operation information 22 and processed measurement data 24 for 30 minutes after the current time do not exist, the latest data, that is, the data for the current time, is used. The water level prediction unit 11 calculates a predicted water level value 60 minutes after the current time by adding the predicted water level change amount during the 30 minutes from the reference time (30 minutes after the current time) obtained using the trained model 25 to the water level at the reference time (predicted water level value 30 minutes after the current time). Thereafter, the water level prediction unit 11 repeats the same process a certain number of times to sequentially calculate the predicted water level value 30 minutes after the current time, the predicted water level value 60 minutes after, the predicted water level value 90 minutes after, and so on. For example, the water level prediction unit 11 calculates the predicted water level value every 30 minutes up to three hours ahead. Note that the interval for calculating the predicted water level value is not limited to every 30 minutes.

[0030] Next, the trained model 25 used by the water level prediction unit 11 when predicting the water level will be described. Here, the water level prediction unit 11 predicts the water level for a plurality of points, but since the width and gradient of the river are different for each point, it is difficult to perform accurate prediction using the same model. In addition, it is considered that information that is strongly correlated with the water level change differs for each point. For example, it is considered that the change in the tide level affects the water level at a point close to the sea level (a point close to the river mouth), but does not affect the water level at a point far from the river mouth. On the other hand, in the upstream where the water flow is fast, it is considered that the correlation between the change in turbidity and the change in water level is stronger than in the midstream or downstream. For example, it is considered that the turbidity changes more greatly when the flow rate of the river increases and the water level rises in the upstream. In addition, there is a correlation between the water level changes at a plurality of points provided in a certain river and its tributaries. For example, when the water level rises at a certain point of the river, the water level at other points downstream also rises after a predetermined time has passed. In addition, the operating status of the river facility 200 affects the surrounding water level. For example, when a floodgate is closed, the water level rises upstream of the floodgate and falls downstream. However, the type and number of river facilities 200 around each point differs from point to point. For this reason, the water level prediction system 1 prepares a different trained model 25 for each prediction target point, and the water level prediction unit 11 predicts the water level at each prediction target point using the trained model 25 associated with the prediction target point. The learning device that generates the trained model 25 generates the trained model 25 by performing learning using learning data with a different configuration for each prediction target point.

[0031] Here, an example of a method for selecting learning data for each prediction target point will be described with reference to Fig. 5 to Fig. 7. Fig. 5 is a first diagram for explaining a method for selecting learning data for each prediction target point, Fig. 6 is a second diagram for explaining a method for selecting learning data for each prediction target point, and Fig. 7 is a third diagram for explaining a method for selecting learning data for each prediction target point. Water level observation stations A1 to A9 shown in Fig. 5 correspond to the water level prediction target points.

[0032] In selecting learning data for each prediction target point, first, for each water level observation station, other water level observation stations related to the water level change are selected. In other words, for each of the water level observation stations A1 to A9, other water level observation stations located upstream of the observation station are selected. For example, if the water level observed at other water level observation stations located upstream of the observation station rises, the water level observed by the observation station will also rise in the future, so it can be said that there is a correlation with the water level change. For this reason, the water level observed at other water level observation stations located upstream is selected as learning data. A model capable of estimating the amount of water level change can be generated by learning how much the water level observed by the observation station will rise in the future when a water level rise occurs at other water level observation stations located upstream. In the example shown in FIG. 5, in the case of the water level observation station A1 located at the most downstream, all of the other water level observation stations A2 to A9 are located upstream, so the water level observation stations A2 to A9 are selected. For water level observation station A6 located in the midstream, water level observation stations A7 to A9 are selected. For water level observation stations A3, A5, and A9 located in the upstream, no water level observation station is selected because there are no other water level observation stations upstream.

[0033] Next, for each water level observation station, a rain gauge station that is related to the water level change is selected. For example, a rain gauge station that exists within a specified range centered on the water level observation station is selected. For example, when it rains, the water level rises with an amount of change corresponding to the amount of rainfall, so it can be said that there is a correlation with the water level change. For this reason, the rainfall observed at the rain gauge stations existing around the water level observation station is selected as the learning data. By learning how much the water level observed by the observation station will rise in the future when a certain amount of rainfall is observed at the surrounding rain gauge stations, a model that can infer the amount of water level change can be generated. In the example shown in FIG. 6, rain gauge station B1 exists within a specified range from water level observation station A1, and rain gauge station B2 exists within a specified range from water level observation station A7 and within a specified range from water level observation station A8. For this reason, rain gauge station B1 is selected as the rain gauge station related to the water level changes of water level observation station A1, and rain gauge station B2 is selected as the rain gauge station related to the water level changes of water level observation stations A7 and A8. Note that the installation location of the rain gauge stations is not limited to rivers. Rain gauge stations are installed at any location on or around rivers. Also, while rain gauge stations B1 and B2 are shown as examples in Figure 6, if tide observation stations and turbidity observation stations exist, these stations are also selected in the same way.

[0034] Next, for each water level observation station, river facilities related to water level changes are selected. For example, river facilities installed within a specified range centered on the water level observation station are selected. For example, when an operation such as opening or closing a water gate is performed, the water level around the water gate changes, so it can be said that there is a relationship to water level changes. For this reason, the operation status of river facilities existing around the water level observation station is selected as learning data. By learning how much the water level observed by the observation station will rise in the future when the operation status of the surrounding river facilities is such that a model capable of inferring the amount of water level change can be generated. In the example shown in FIG. 7, since the iron door D1 is located within a specified range from the water level observation station A1, the iron door D1 is selected as the river facility related to the water level change of the water level observation station A1. Also, since the water gate C1 is located within a specified range from the water level observation station A7, the water gate C1 is selected as the river facility related to the water level change of the water level observation station A7. In Figure 7, iron gate D1 and water gate C1 are shown as examples of river facilities, but river facilities are not limited to iron gates and water gates. Other river facilities include pumping stations, culverts, regulating ponds, flood control green spaces, underground rivers, dams, and sewage pumps. These river facilities are also selected as river facilities related to the water level changes at each water level observation station.

[0035] The selection of observation stations (water level observation stations, rainfall observation stations, tide observation stations, turbidity observation stations, etc.) and river facilities (iron gates, sluice gates, pumping stations, culverts, regulating ponds, flood control green spaces, underground rivers, dams, sewage pumps, etc.) that are relevant to water level changes at each point is carried out, for example, by a person in charge of flood prevention work.

[0036] When the above three stages of selection are completed, the configuration of the learning data used to generate the trained model 25 for each prediction target point is determined. In the example described with reference to Figs. 5 to 7, for example, for the water level observation station A1, the water level observed at each of the water level observation stations A2 to A9, the rainfall observed at the rainfall observation station B1, and the operating status of the iron door D1 are selected as the learning data. In addition to the rainfall observation station B1 and the iron door D1, if there are other observation stations and river facilities not shown in the figure in the vicinity of the water level observation station A1, the measurement data obtained at those observation stations and the operating status of the river facilities are also selected as the learning data. For the water level observation station A2, the water level observed at the water level observation station A3 is selected as the learning data. For the water level observation station A8, the water level observed at the water level observation station A9 and the rainfall observed at the rainfall observation station B2 are selected as the learning data. The above-mentioned other system information 21, such as the predicted value of future rainfall, is selected as learning data for all water level observation stations.

[0037] In this way, when generating a trained model 25 for each water level observation station corresponding to the water level prediction target point, information that is correlated with the water level changes at the prediction target point using the trained model 25 is selected as learning data.

[0038] Fig. 8 is a diagram showing an example of a learning device 60 that generates a trained model 25 used by the water level prediction unit 11 of the water level prediction system 1 according to the first embodiment. The learning device 60 shown in Fig. 8 generates a trained model 25 used in the process of predicting the water level by the water level prediction unit 11. The learning device 60 generates a different trained model 25 for each of a plurality of prediction target points for which the water level prediction unit 11 predicts the water level.

[0039] The learning device 60 includes a data acquisition unit 61 that acquires learning data, and a model generation unit 62 that generates a trained model 25. The trained model 25 generated by the model generation unit 62 is held in a trained model storage unit 70. Although FIG. 8 shows an example in which the trained model storage unit 70 is provided outside the learning device 60, the trained model storage unit 70 may be provided inside the learning device 60. The trained model storage unit 70 may also be provided in the database 20 of the water level prediction system 1. The data acquisition unit 61 and the model generation unit 62 may be provided in the server 10 of the water level prediction system 1.

[0040] The data acquisition unit 61 acquires the measurement data, the surrounding environment data, and the amount of water level change over a certain period as learning data. The amount of water level change over a certain period corresponds to the teacher data for machine learning. The amount of water level change over a certain period is the amount of water level change over a certain period from the measurement time included in the measurement data until a certain time has elapsed. For example, if the certain time is 30 minutes, the amount of water level change over a certain period from the measurement time until 30 minutes has elapsed is the amount of water level change over a certain period. As described above, the measurement data is the measurement result by the measuring device 150 such as a water level gauge, a rain gauge, a turbidity meter, etc. The surrounding environment data is other system information such as a predicted value of future rainfall, a predicted value of tide level, etc., and river facility operation information indicating the operation status of the river facility. The configuration of the measurement data and the configuration of the surrounding environment data are different for each point where water level prediction is performed using the trained model 25 to be generated. For example, when generating a trained model 25 used to predict the amount of water level change at the water level observation station A1 described with reference to Figures 5 to 7, the measured water levels at each of the water level observation stations A2 to A9 and the measured rainfall at the rainfall observation station B1 are acquired as measurement data by the data acquisition unit 61. The operating status of the iron door D1 is acquired as river facility operation information of the surrounding environment data by the data acquisition unit 61. In addition, when generating a trained model 25 used to predict the amount of water level change at the water level observation station A8, the measured water level at the water level observation station A9 and the measured rainfall at the rainfall observation station B2 are acquired as measurement data by the data acquisition unit 61.

[0041] The amount of water level change for a certain period of time used as training data is created based on the actual water level that has already been measured. Note that the measurement data used as learning data may be measurement data obtained by executing a process similar to the process of removing abnormal measurement data by the measurement data processing unit 12 described above, that is, measurement data from which abnormal measurement data has been removed. By using the measurement data from which abnormal measurement data has been removed as learning data, the accuracy of inference by the trained model 25 to be generated can be improved.

[0042] The model generation unit 62 learns the relationship between the measurement data and the surrounding environment data and the amount of water level change over a certain period of time based on the learning data created based on a combination of the measurement data and the surrounding environment data output from the data acquisition unit 61 and the amount of water level change over a certain period of time. That is, the model generation unit 62 generates a trained model 25 that outputs a predicted value of the amount of water level change over a certain period of time when the measurement data and the surrounding environment data are input. Here, the learning data is data that associates the measurement data and the surrounding environment data with the amount of water level change over a certain period of time.

[0043] The model generation unit 62 is realized by applying a neural network, for example. The model generation unit 62 learns the amount of water level change over a certain period of time corresponding to the measurement data and the surrounding environment data by so-called supervised learning, for example, according to a neural network model. Here, supervised learning refers to a method in which a set of data of input and result (label) is given to the learning device 60, and the features of the learning data are learned and a result is inferred from the input.

[0044] A neural network is composed of an input layer made up of multiple neurons, an intermediate layer (hidden layer) made up of multiple neurons, and an output layer made up of multiple neurons. The intermediate layer may be one layer, or two or more layers.

[0045] Fig. 9 is a diagram for explaining a neural network. For example, in a three-layered neural network as shown in Fig. 9, when multiple inputs are input to the input layer X (X1-X3), the values ​​are multiplied by weight W1 (w11-w16) and input to the intermediate layer Y (Y1-Y2), and the result is further multiplied by weight W2 (w21-w26) and output from the output layer Z (Z1-Z3). This output result changes depending on the values ​​of weight W1 and weight W2.

[0046] The neural network used in the model generation unit 62 of the learning device 60 in this embodiment learns the amount of water level change over a certain period of time corresponding to the measurement data and the surrounding environmental data through so-called supervised learning, in accordance with learning data created based on a combination of the measurement data, the surrounding environmental data, and the amount of water level change over a certain period of time acquired by the data acquisition unit 61.

[0047] In other words, the neural network learns by inputting measurement data and surrounding environmental data into the input layer and adjusting the weights W1 and W2 so that the results output from the output layer approach the amount of water level change over a certain period of time.

[0048] The model generation unit 62 executes the above-described learning to generate and output a trained model.

[0049] The trained model storage unit 70 stores the trained model output from the model generation unit 62.

[0050] Next, an operation of the learning device 60 for generating a trained model will be described with reference to Fig. 10. Fig. 10 is a flowchart showing an example of the operation of the learning device 60.

[0051] The learning device 60 first acquires learning data (step S21). Specifically, the data acquisition unit 61 acquires the measurement data, the surrounding environment data, and the amount of water level change over a certain period of time, and creates learning data by associating these. Note that, although the measurement data, the surrounding environment data, and the amount of water level change over a certain period of time are acquired simultaneously, it is sufficient that the measurement data, the surrounding environment data, and the amount of water level change over a certain period of time are acquired in an associated manner, and the measurement data, the surrounding environment data, and the amount of water level change over a certain period of time may be acquired at different times.

[0052] The learning device 60 then performs a learning process (step S22). Specifically, the model generating unit 62 learns the amount of water level change over a certain period of time corresponding to the measurement data and the surrounding environment data by so-called supervised learning according to learning data created based on a combination of the measurement data, the surrounding environment data, and the amount of water level change over a certain period of time acquired by the data acquiring unit 61, and generates a learned model.

[0053] The learning device 60 then stores the trained model in the trained model storage unit 70 (step S23). Specifically, the model generation unit 62 outputs the trained model generated in step S22, and the trained model storage unit 70 stores the trained model.

[0054] Next, an inference device for realizing the water level prediction unit 11 will be described with reference to Figs. 11 and 12. Fig. 11 is a diagram showing a configuration example of an inference device 80 for realizing the water level prediction unit 11 of the water level prediction system 1 according to the first embodiment. The inference device 80 includes a data acquisition unit 81 for acquiring measurement data and surrounding environment data as inference data, and an inference unit 82 for inferring the amount of water level change over a certain period of time. The trained model storage unit 70 shown in Fig. 11 corresponds to the trained model storage unit 70 shown in Fig. 8, and stores a trained model generated by the learning device 60. Note that, although Fig. 11 shows an example in which the trained model storage unit 70 is provided outside the inference device 80, the trained model storage unit 70 may be provided inside the inference device 80.

[0055] The inference unit 82 infers the amount of water level change over a certain period of time corresponding to the measurement data and the surrounding environment data, using the trained model stored in the trained model storage unit 70. That is, the inference unit 82 inputs the measurement data and the surrounding environment data acquired by the data acquisition unit 81 into the trained model stored in the trained model storage unit 70 to execute inference, and acquires the amount of water level change over a certain period of time as an inference result. The inference unit 82 outputs the amount of water level change over a certain period of time, which is the inference result.

[0056] The inference device 80 is used to infer the amount of water level change over a certain period of time corresponding to the measurement data and the surrounding environment data, but may be configured to be connected to the water level prediction system 1 via a network and operate as a device separate from the water level prediction system 1. The inference device 80 may also be built into the water level prediction system 1. Furthermore, the inference device 80 may exist on a cloud server external to the water level prediction system 1.

[0057] Next, the operation of the inference device 80 to infer the amount of water level change over a certain period of time will be described with reference to Fig. 12. Fig. 12 is a flowchart showing an example of the operation of the inference device 80, and shows an example of the operation of the water level prediction unit 11 of the water level prediction system 1 according to the first embodiment to infer the amount of water level change over a certain period of time using the inference device 80.

[0058] First, the inference device 80 acquires data to be used for inferring the amount of water level change over a certain period of time (step S31). Specifically, the data acquisition unit 81 acquires measurement data and surrounding environment data as data for inference.

[0059] The inference device 80 then inputs the inference data acquired in step S31 into the trained model stored in the trained model storage unit 70 (step S32) to obtain an inference result. That is, the inference unit 82 inputs the measurement data and the surrounding environment data acquired by the data acquisition unit 81 into the trained model stored in the trained model storage unit 70, and accordingly acquires the amount of water level change over a certain period of time output from the trained model.

[0060] The inference device 80 then outputs the amount of water level change over a certain period, which is the inference result (step S33).

[0061] This allows the amount of water level change over a certain period of time to be inferred from the measurement data and surrounding environmental data.

[0062] In this way, machine learning can be used to realize the process in which the water level prediction unit 11 of the water level prediction system 1 derives the amount of water level change for a certain period of time.

[0063] In this embodiment, the inference device 80 that infers the amount of water level change from the measurement data and the surrounding environment data has been described, but it is also possible to configure the inference device 80 to infer the water level itself from the measurement data and the surrounding environment data. That is, the inference device 80 can also be configured to perform inference using a trained model that has been trained to infer the water level at a time when a fixed time has elapsed from a certain time. In this case, the above-mentioned learning device 60 performs learning using data that associates the water level, the measurement data, and the surrounding environment data at a certain time with the water level at a time when a fixed time (e.g., 30 minutes) has elapsed from the certain time as learning data. In this case, the water level at the time when a fixed time has elapsed from the certain time becomes the teacher data.

[0064] Furthermore, the measurement data used as learning data and inference data may be average values, integrated values, amounts of change, etc., over a certain period of time in the past.

[0065] In addition, the learning algorithm used in the model generation unit 62 may be deep learning, which learns to extract the features themselves, or machine learning may be performed according to other known methods, such as genetic programming, functional logic programming, and support vector machines.

[0066] Next, a description will be given of the overall operation of the water level prediction system 1. Fig. 13 is a flowchart showing an example of the operation of the water level prediction system 1 according to the first embodiment.

[0067] The water level prediction system 1 first acquires measurement data and surrounding environment data (step S41). That is, the measurement data acquisition unit 31 acquires the measurement data 23 from the measurement equipment 150, and the surrounding environment data acquisition unit 41 acquires the river facility operation information 22 from the river facility 200 as the surrounding environment data 28, and acquires the other system information 21 from the other system 300. In addition, the measurement data processing unit 12 executes a data processing process to remove abnormal data from the measurement data 23, and generates processed measurement data 24.

[0068] The water level prediction system 1 then determines the predicted time (step S42). The predicted time is determined by the water level prediction unit 11. For example, the water level prediction unit 11 sets the measurement time indicated by the time information included in the measurement data acquired in step S41 as the reference time, and determines the time when a certain amount of time has passed since the reference time as the predicted time. For example, the water level prediction unit 11 sets the predicted time to be 10 minutes after the set reference time. Note that the time interval from the reference time to the predicted time (for example, 10 minutes) is fixed, and the predicted time is determined when the reference time is determined. Therefore, the process of determining the predicted time can also be said to be a process of determining the reference time.

[0069] The water level prediction system 1 then selects a water level observation station for water level prediction (step S43). The selection of the water level observation station for water level prediction is performed by the water level prediction unit 11. The water level prediction unit 11 selects one of the water level observation stations for which the water level prediction has not been completed at the prediction time determined in step S42.

[0070] The water level prediction system 1 next predicts the water level at the prediction time determined in step S42 for the water level observation station selected in step S43 (step S44). Specifically, the water level prediction unit 11 predicts the water level at the selected station at the prediction time determined in step S42 using the trained model 25 associated with the selected station, which is the water level observation station selected in step S43. The water level prediction unit 11 first inputs the river facility operation information 22 and the processed measurement data 24 at the reference time and the other system information 21 at the prediction time to the trained model 25, and accordingly obtains a predicted value of the water level change output from the trained model 25. The predicted value output from the trained model 25 is a predicted value of the water level change from the reference time to the prediction time. The water level prediction unit 11 then obtains a predicted value of the water level at the prediction time by adding the predicted value of the water level change output from the trained model 25 to the water level at the selected station at the reference time.

[0071] The water level prediction system 1 then checks whether predictions for all stations have been completed, i.e., whether predictions of the water level at the prediction time have been completed for all water level observation stations (step S45). This check is performed by the water level prediction unit 11. If predictions for all stations have not been completed (step S45: No), the process returns to step S43, where the water level prediction unit 11 selects a water level observation station for which the water level will be predicted (step S43), and predicts the water level of the selected water level observation station (step S44). The water level observation station for which the water level will be predicted is selected from among water level observation stations for which the water level at the prediction time has not yet been predicted.

[0072] On the other hand, if predictions for all stations are completed (step S45: Yes), the water level prediction unit 11 checks whether predictions for all times are completed, that is, whether water level predictions are completed over the entire predetermined prediction target period (step S46). For example, if the prediction target period is a period from the current time to three hours ahead, the water level prediction unit 11 checks whether water level predictions for the period from the current time to three hours ahead are completed. If predictions for all times are not completed (step S46: No), the water level prediction unit 11 returns to step S42 and determines the prediction time (step S42). In this case, the water level prediction unit 11 sets the prediction time determined in the previous execution of step S42 as a new reference time, and sets the time when a certain time is determined from the reference time as the prediction time. For example, if the reference time set in the previous execution of step S42 was 11:00 and the determined predicted time was 11:10, the water level prediction unit 11 sets the new reference time to 11:10 and determines the predicted time to be 11:20. Thereafter, the processes of steps S42 to S46 are repeated until it is determined in step S46 that predictions for all times have been completed. If predictions for all times have been completed (step S46: Yes), the operation of the water level prediction system 1 ends.

[0073] Next, we will explain the hardware configuration of the server 10 that realizes the water level prediction unit 11 and the measurement data processing unit 12 of the water level prediction system 1. Fig. 14 is a diagram showing an example of the hardware configuration of the server 10 that realizes the water level prediction unit 11 and the measurement data processing unit 12 of the water level prediction system 1. The server 10 is composed of, for example, a processor 201, a memory 202, and a communication circuit 203 shown in Fig. 14.

[0074] The processor 201 is a CPU (Central Processing Unit, also called a central processing unit, processing unit, arithmetic unit, microprocessor, microcomputer, or DSP (Digital Signal Processor)), a system LSI (Large Scale Integration), etc. The memory 202 is a non-volatile or volatile semiconductor memory such as a RAM (Random Access Memory), a ROM (Read Only Memory), or a flash memory, a magnetic disk, etc. The communication circuit 203 is an electronic circuit for connecting to a communication network and transmitting and receiving data, etc.

[0075] When the server 10 is composed of a processor 201, a memory 202, and a communication circuit 203, the above-mentioned water level prediction unit 11 and the measurement data processing unit 12 are realized by the processor 201 executing a program for operating as each of these units. The program for operating as the water level prediction unit 11 and the measurement data processing unit 12 is stored in advance in the memory 202. The processor 201 operates as the water level prediction unit 11 and the measurement data processing unit 12 by reading out the above program from the memory 202 and executing it.

[0076] In addition, the programs stored in memory 202 for operating as the water level prediction unit 11 and the measurement data processing unit 12 may be provided to users, etc. in a form written on a storage medium such as a CD (Compact Disc)-ROM or a DVD (Digital Versatile Disc)-ROM, or may be provided to users, etc. via a network.

[0077] As described above, the water level prediction system 1 according to the first embodiment includes a measurement data acquisition unit 31 that acquires measurement data 23 from a measurement device 150 such as a water level gauge installed in a river, a surrounding environment data acquisition unit 41 that acquires river facility operation information 22 and other system information 21, which are surrounding environment data 28, a measurement data processing unit 12 that generates processed measurement data 24 by removing abnormal data included in the measurement data 23, and a water level prediction unit 11 that predicts the future water level of the river based on the processed measurement data 24, the surrounding environment data 28, and a trained model 25 for inferring the amount of change in the water level of the river. The trained model 25 exists individually for each of the multiple water level prediction target points. In addition, each of the multiple trained models 25 is generated using learning data with a different configuration for each prediction target point. The learning data is composed of information that is correlated with the water level change at the prediction target point. According to the water level prediction system 1 of the first embodiment, the prediction process is performed using the processed measurement data 24, which is the measurement data 23 from which abnormal data has been removed, and also using a trained model generated using different training data for each point where the water level is predicted, so that it is possible to achieve high accuracy in water level prediction. Note that, although the embodiment has been described as predicting the water level at a point set in a river, the multiple points where the water level prediction system 1 predicts the water level may include a point set in a lake or marsh.

[0078] Embodiment 2 Fig. 15 is a diagram showing an example of a functional block configuration of a water level prediction system 1a according to embodiment 2. In Fig. 15, the same components as those in the water level prediction system 1 according to embodiment 1 shown in Fig. 2 are denoted by the same reference numerals. Descriptions of the components denoted by the same reference numerals as those in the water level prediction system 1 according to embodiment 1 will be omitted.

[0079] The water level prediction system 1a includes a server 10a, a database 20, a measurement data acquisition unit 31, and an ambient environment data acquisition unit 41. The server 10a includes a water level prediction unit 11, a measurement data processing unit 12, and a learning unit 13.

[0080] In the water level prediction system 1a according to this embodiment, the learning unit 13 performs re-learning at a predetermined timing to update the trained model 25. For example, when the measurement data processing unit 12 creates new processed measurement data 24, the learning unit 13 performs re-learning using the latest processed measurement data 24 to update the trained model 25. The learning unit 13 may perform re-learning periodically at a set cycle, such as performing re-learning once a day. The learning method is the same as the method in which the learning device 60 described in the first embodiment learns the relationship between the measurement data and the surrounding environment data and the amount of water level change for a certain period of time.

[0081] In this way, the water level prediction system 1a according to the present embodiment further includes a learning unit 13, which re-learns the trained model 25 used by the water level prediction unit 11 when predicting the water level. This makes it possible to appropriately update the trained model 25 during operation of the water level prediction system 1a, and to predict the water level using the optimized trained model 25.

[0082] The configurations shown in the above embodiments are merely examples, and may be combined with other known technologies, or the embodiments may be combined with each other. Also, parts of the configurations may be omitted or modified without departing from the spirit of the invention.

[0083] Various aspects of the present disclosure are summarized below as appendices.

[0084] (Appendix 1) a measurement data acquisition unit that acquires measurement data of the river including measurement values ​​of water levels at a plurality of points of the river; a surrounding environment data acquisition unit that acquires surrounding environment data including an operating status of river facilities provided in the river basin; a measurement data processing unit that removes abnormal data included in the measurement data acquired by the measurement data acquisition unit; a water level prediction unit that predicts the water level of the river based on a trained model generated by performing learning to predict a water level change based on the measurement data and the surrounding environment data, processed measurement data, which is the measurement data from which abnormal data has been removed by the measurement data processing unit, and the surrounding environment data acquired by the surrounding environment data acquisition unit; and A water level prediction system comprising: (Appendix 2) There are a plurality of target points for the water level prediction, The water level prediction unit performs the water level prediction using a different trained model for each of the target points, The trained model, which differs for each target location, is generated by performing training using training data individually selected for each target location. 2. The water level prediction system according to claim 1, (Appendix 3) The measurement data further includes a measurement value of turbidity of the river, a measurement value of rainfall, and a measurement value of tide level. 3. The water level prediction system according to claim 1 or 2. (Appendix 4) The surrounding environment data further includes other system information acquired from other external systems, The other system information includes all or a part of a rainfall prediction value, a tide level prediction value, and a water level prediction value at a point other than the target point of the water level prediction, 4. A water level prediction system according to any one of claims 1 to 3. (Appendix 5) the water level prediction unit predicts a water level change amount from a preset reference time until a certain time has elapsed using the learned model, the processed measurement data, and the surrounding environment data, and further adds the predicted water level change amount to the water level at the reference time to generate a predicted value of the water level at a prediction time, which is the time when the certain time has elapsed from the reference time. 5. A water level prediction system according to any one of claims 1 to 4. (Appendix 6) the water level prediction unit resets the prediction time to the new reference time, predicts a water level change from the reset reference time to the prediction time using the learned model, the processed measurement data, and the surrounding environment data, and further adds the predicted water level change to the water level at the reset reference time to generate a predicted value of the water level at the reset prediction time; and repeatedly executes a water level prediction process to estimate the water level every time the fixed time elapses over a specified future period. 6. The water level prediction system according to claim 5, (Appendix 7) When the certain time is long compared to a period during which the measurement values ​​included in the measurement data are obtained, the water level prediction unit uses an average value of the measurement values ​​included in the measurement data for the certain period of time in the process of predicting the water level change amount. 7. The water level prediction system according to claim 5 or 6. (Appendix 8) a learning unit that re-learns the learned model based on the processed measurement data and the surrounding environment data; 8. A water level prediction system according to any one of claims 1 to 7, comprising: (Appendix 9) a data acquisition unit that acquires measurement data of a river including measurement values ​​of water levels at multiple points of the river and surrounding environment data including operating conditions of river facilities installed in the river basin; a model generation unit that generates a trained model for inferring a water level change per unit time at a prediction target point, which is one of the plurality of points, from the measurement data and the surrounding environment data, using learning data created based on the measurement data and the surrounding environment data; and A learning device comprising: (Appendix 10) A water level prediction method for a water level prediction system to predict the water level of a river, comprising: a measurement data acquisition step of acquiring measurement data of the river including measurement values ​​of water levels at a plurality of points of the river; A surrounding environment data acquisition step of acquiring surrounding environment data including an operating status of a river facility provided in the river basin; a measurement data processing step of excluding abnormal data included in the measurement data acquired in the measurement data acquisition step; a water level prediction step of predicting the water level of the river based on a trained model generated by performing learning to predict a water level change based on the measurement data and the surrounding environment data, processed measurement data which is the measurement data from which abnormal data has been removed in the measurement data processing step, and the surrounding environment data acquired in the surrounding environment data acquisition step; A water level prediction method comprising: (Appendix 11) a measurement data acquisition step of acquiring measurement data of the river including measurement values ​​of water levels at a plurality of points of the river; A surrounding environment data acquisition step of acquiring surrounding environment data including an operating status of a river facility provided in the river basin; a measurement data processing step of excluding abnormal data included in the measurement data acquired in the measurement data acquisition step; a water level prediction step of predicting the water level of the river based on a trained model generated by performing learning to predict a water level change based on the measurement data and the surrounding environment data, processed measurement data which is the measurement data from which abnormal data has been removed in the measurement data processing step, and the surrounding environment data acquired in the surrounding environment data acquisition step; A water level prediction program characterized by causing a computer to execute the above. (Appendix 12) A learning method for a learning device to generate a trained model, comprising: A data acquisition step of acquiring measurement data of a river including measurement values ​​of water levels at a plurality of points of the river and surrounding environment data including operating conditions of river facilities installed in the river basin; a model generation step of generating a trained model for inferring a water level change per unit time at a prediction target point, which is one of the plurality of points, from the measurement data and the surrounding environment data using learning data generated based on the measurement data and the surrounding environment data; A learning method comprising: (Appendix 13) A data acquisition step of acquiring measurement data of a river including measurement values ​​of water levels at a plurality of points of the river and surrounding environment data including operating conditions of river facilities installed in the river basin; a model generation step of generating a trained model for inferring a water level change per unit time at a prediction target point, which is one of the plurality of points, from the measurement data and the surrounding environment data using learning data generated based on the measurement data and the surrounding environment data; A learning program characterized by causing a computer to execute the above. [Explanation of symbols]

[0085] 1,1a water level prediction system, 10,10a server, 11 water level prediction section, 12 measurement data processing section, 13 learning section, 20 database, 21 other system information, 22 river facility operation information, 23 measurement data, 24 processed measurement data, 25 trained model, 26 water level prediction value, 28 surrounding environment data, 30,110 gateway device, 31 measurement data acquisition section, 40 linkage device, 41 surrounding environment data acquisition section, 60 learning device, 61,81 data acquisition section, 62 model generation section, 70 trained model memory section, 80 inference device, 82 inference section, 100 field, 120 slave station, 130 machine side operation panel, 140 operation object, 150 measuring equipment, 200 river facility, 300 other systems.

Claims

1. a measurement data acquisition unit that acquires measurement data of the river including measurement values ​​of water levels at a plurality of points of the river; a surrounding environment data acquisition unit that acquires surrounding environment data including an operating status of river facilities provided in the river basin; a measurement data processing unit that removes abnormal data included in the measurement data acquired by the measurement data acquisition unit; a water level prediction unit that predicts the water level of the river based on a trained model generated by performing learning to predict a water level change based on the measurement data and the surrounding environment data, processed measurement data, which is the measurement data from which abnormal data has been removed by the measurement data processing unit, and the surrounding environment data acquired by the surrounding environment data acquisition unit; and A water level prediction system comprising:

2. There are a plurality of target points for the water level prediction, The water level prediction unit performs the water level prediction using a different trained model for each of the target points, The trained model, which differs for each target location, is generated by performing training using training data individually selected for each target location. The water level prediction system according to claim 1 .

3. The measurement data further includes a measurement value of turbidity of the river, a measurement value of rainfall, and a measurement value of tide level. The water level prediction system according to claim 1 .

4. The surrounding environment data further includes other system information acquired from other external systems, The other system information includes all or a part of a rainfall prediction value, a tide level prediction value, and a water level prediction value at a point other than the target point of the water level prediction, The water level prediction system according to claim 1 .

5. the water level prediction unit predicts a water level change amount from a preset reference time until a certain time has elapsed using the learned model, the processed measurement data, and the surrounding environment data, and further adds the predicted water level change amount to the water level at the reference time to generate a predicted value of the water level at a prediction time, which is the time when the certain time has elapsed from the reference time. The water level prediction system according to claim 1 .

6. the water level prediction unit resets the prediction time to the new reference time, predicts a water level change from the reset reference time to the prediction time using the learned model, the processed measurement data, and the surrounding environment data, and further adds the predicted water level change to the water level at the reset reference time to generate a predicted value of the water level at the reset prediction time; and repeatedly executes a water level prediction process to estimate the water level every time the fixed time elapses over a specified future period. The water level prediction system according to claim 5.

7. When the certain time is long compared to a period during which the measurement values ​​included in the measurement data are obtained, the water level prediction unit uses an average value of the measurement values ​​included in the measurement data for the certain period of time in the process of predicting the water level change amount. The water level prediction system according to claim 5.

8. a learning unit that re-learns the learned model based on the processed measurement data and the surrounding environment data; The water level prediction system according to any one of claims 1 to 7, further comprising:

9. a data acquisition unit that acquires measurement data of a river including measurement values ​​of water levels at multiple points of the river and surrounding environment data including operating conditions of river facilities installed in the river basin; a model generation unit that generates a trained model for inferring a water level change per unit time at a prediction target point, which is one of the plurality of points, from the measurement data and the surrounding environment data, using learning data created based on the measurement data and the surrounding environment data; and A learning device comprising:

10. A water level prediction method for a water level prediction system to predict the water level of a river, comprising: a measurement data acquisition step of acquiring measurement data of the river including measurement values ​​of water levels at a plurality of points of the river; A surrounding environment data acquisition step of acquiring surrounding environment data including an operating status of a river facility provided in the river basin; a measurement data processing step of excluding abnormal data included in the measurement data acquired in the measurement data acquisition step; a water level prediction step of predicting the water level of the river based on a trained model generated by performing learning to predict a water level change based on the measurement data and the surrounding environment data, processed measurement data which is the measurement data from which abnormal data has been removed in the measurement data processing step, and the surrounding environment data acquired in the surrounding environment data acquisition step; A water level prediction method comprising:

11. a measurement data acquisition step of acquiring measurement data of the river including measurement values ​​of water levels at a plurality of points of the river; A surrounding environment data acquisition step of acquiring surrounding environment data including an operating status of a river facility provided in the river basin; a measurement data processing step of excluding abnormal data included in the measurement data acquired in the measurement data acquisition step; a water level prediction step of predicting the water level of the river based on a trained model generated by performing learning to predict a water level change based on the measurement data and the surrounding environment data, processed measurement data which is the measurement data from which abnormal data has been removed in the measurement data processing step, and the surrounding environment data acquired in the surrounding environment data acquisition step; A water level prediction program characterized by causing a computer to execute the above steps.

12. A learning method for a learning device to generate a trained model, comprising: A data acquisition step of acquiring measurement data of a river including measurement values ​​of water levels at a plurality of points of the river and surrounding environment data including operating conditions of river facilities installed in the river basin; a model generation step of generating a trained model for inferring a water level change per unit time at a prediction target point, which is one of the plurality of points, from the measurement data and the surrounding environment data using learning data generated based on the measurement data and the surrounding environment data; A learning method comprising:

13. A data acquisition step of acquiring measurement data of a river including measurement values ​​of water levels at a plurality of points of the river and surrounding environment data including operating conditions of river facilities installed in the river basin; a model generation step of generating a trained model for inferring a water level change per unit time at a prediction target point, which is one of the plurality of points, from the measurement data and the surrounding environment data using learning data generated based on the measurement data and the surrounding environment data; A learning program characterized by causing a computer to execute the above.

Images