Learning device, prediction device, learning method, prediction method, and water treatment system

Abstract

To provide a learning device, a prediction device, a learning method, a prediction method, and a water treatment system that make it possible to improve a prediction accuracy of an inflow amount of water to be treated.SOLUTION: A learning device includes: a region specifying unit that specifies one or more regions in which a measurement state of a rainfall intensity satisfies a condition among a plurality of regions in which the rainfall intensity of rainwater can be measured; an intensity acquisition unit that acquires the rainfall intensity at a first timing for each of identified one or more regions; an inflow amount acquisition unit that acquires an inflow amount at a second timing that is a predetermined time after the first timing; a data generating unit that generates training data including the rainfall intensity in each of the one or more acquired regions and the acquired inflow amount; and a model generation unit that generates a learning model through machine learning of the generated training data.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a learning device, a prediction device, a learning method, a prediction method, and a water treatment system. [Background technology]

[0002] In a water treatment system (hereinafter simply referred to as a water treatment system) that treats water to be treated, for example, a prediction is made regarding the inflow amount of the water to be treated into the water treatment system. Then, an operator of the water treatment system (hereinafter simply referred to as an operator) performs necessary operation of the water treatment system based on, for example, the prediction result regarding the inflow amount of the water to be treated (see Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2020-170339 A Summary of the Invention [Problem to be solved by the invention]

[0004] In the water treatment system described above, it is required to perform an appropriate operation according to the situation, and therefore it is desired to improve the prediction accuracy of the inflow amount of the water to be treated, for example. [Means for solving the problem]

[0005] In order to improve the prediction accuracy of the inflow volume of treated water as described above, the learning device of the present invention is a learning device that generates a learning model that predicts the inflow volume of treated water flowing into a water treatment device that treats treated water including rainwater, and has an area identification unit that identifies one or more areas among a plurality of areas in which the rainfall intensity of the rainwater can be measured, where the measured state of the rainfall intensity satisfies a condition, an intensity acquisition unit that acquires the rainfall intensity at a first timing for each of the one or more identified areas, an inflow volume acquisition unit that acquires the inflow volume at a second timing a predetermined time after the first timing, a data generation unit that generates teacher data including the rainfall intensity in each of the one or more acquired areas and the acquired inflow volume, and a model generation unit that generates the learning model by machine learning of the generated teacher data. Effect of the Invention

[0006] According to the learning device, prediction device, learning method, prediction method, and water treatment system of the present invention, it is possible to improve the prediction accuracy of the inflow amount of water to be treated. [Brief description of the drawings]

[0007] [Figure 1] FIG. 1 is a diagram illustrating an example of a configuration of a water treatment system 100 according to the first embodiment. [Diagram 2] FIG. 2 is a diagram illustrating a hardware configuration of information processing device 10 in the first embodiment. [Diagram 3] FIG. 3 is a block diagram of information processing device 10 according to the first embodiment. [Figure 4] FIG. 4 is a diagram illustrating a specific example of the learning process in the first embodiment. [Diagram 5] FIG. 5 is a diagram illustrating a specific example of the prediction process according to the first embodiment. [Figure 6] FIG. 6 is a flowchart illustrating the learning process in the first embodiment. [Figure 7] FIG. 7 is a diagram illustrating a specific example of step S1. [Figure 8] FIG. 8 is a diagram illustrating a specific example of step S1. [Figure 9] FIG. 9 is a diagram illustrating a specific example of step S1. [Figure 10] FIG. 10 is a diagram illustrating a specific example of step S1. [Figure 11] FIG. 11 is a diagram illustrating a specific example of step S3. [Figure 12] FIG. 12 is a diagram illustrating a specific example of the teacher data DT4. [Figure 13] FIG. 13 is a diagram illustrating a specific example of the teacher data DT4. [Figure 14] FIG. 14 is a flowchart illustrating the prediction process according to the first embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0008] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings. However, such descriptions should not be interpreted in a limiting sense and do not limit the subject matter described in the claims. In addition, various changes, substitutions, and modifications can be made without departing from the spirit and scope of the present disclosure. In addition, different embodiments can be appropriately combined.

[0009] [Water treatment system 100 according to the first embodiment] First, a water treatment system 100 in the first embodiment will be described. Fig. 1 is a diagram illustrating a configuration example of the water treatment system 100 in the first embodiment.

[0010] The water treatment system 100 includes, for example, a pump P (hereinafter also referred to as a water treatment device P), an information processing device 10 (hereinafter also referred to as a learning device 10 or a prediction device 10), and a control device 20. Note that, although the following description will be given of a case in which the water treatment system 100 includes one pump P, the present invention is not limited to this. Specifically, the water treatment system 100 may include, for example, multiple pumps P having the same capacity, or multiple pumps P having different capacities.

[0011] The pump P is, for example, a pump that appropriately transfers the water to be treated that has flowed in through the pipe L to downstream equipment (not shown). The water to be treated may include, for example, at least one of rainwater and sewage. The pipe L may be, for example, made up of a plurality of pipes that are in communication with each other.

[0012] The information processing device 10 is, for example, a physical machine or a virtual machine, and performs a process of generating a learning model for predicting the inflow amount of the water to be treated flowing into the pump P (hereinafter, also referred to as a learning process).

[0013] Specifically, the information processing device 10 acquires, for example, the rainfall intensity of rainwater measured by a radar R and the water level in the pipe L measured by a water level gauge W (for example, the water level measured at the same timing as the rainfall intensity). In addition, the information processing device 10 acquires the inflow amount of the water to be treated (water to be treated including rainwater) that has flowed into the pump P at a timing (hereinafter also referred to as the second timing) after a predetermined time (hereinafter also referred to as the predetermined time) from the timing (hereinafter also referred to as the measurement timing or the first timing) at which the acquired rainfall intensity and water level are measured. The predetermined time may be, for example, several minutes or several tens of minutes. Then, the information processing device 10 generates teacher data (a plurality of teacher data) including, for example, the acquired rainfall intensity, water level, and inflow amount. After that, the information processing device 10 generates a learning model by, for example, machine learning of the generated teacher data. Note that the learning model generated by the information processing device 10 may be, for example, a Long Short Term Memory (LSTM).

[0014] Furthermore, the information processing device 10 performs a process of predicting the future inflow amount of the water to be treated into the pump P using the learning model generated in the learning process (hereinafter, also referred to as a prediction process).

[0015] Specifically, the information processing device 10 acquires, for example, rainfall intensity (hereinafter also referred to as new rainfall intensity) measured by a radar R and a water level (hereinafter also referred to as new water level) in the pipe L measured by a water level gauge W. Then, the information processing device 10 inputs, for example, the acquired new rainfall intensity and new water level into a learning model, and outputs a new inflow volume output from the learning model in association with the input of the new rainfall intensity and the new water level as a prediction result of the inflow volume of the water to be treated flowing into the pump P at a timing (hereinafter also referred to as fourth timing) a predetermined time after the timing (hereinafter also referred to as third timing) at which the new rainfall intensity and the new water level are measured.

[0016] The control device 20 automatically operates the pump P based on the inflow amount predicted in the prediction process by the information processing device 10, for example.

[0017] Specifically, for example, when the inflow volume predicted by the information processing device 10 is greater than the current inflow volume, the control device 20 performs control to increase the rotation speed of the pump P. Also, for example, when the inflow volume predicted by the information processing device 10 is less than the current inflow volume, the control device 20 performs control to decrease the rotation speed of the pump P.

[0018] Furthermore, for example, when the water treatment system 100 has multiple pumps P and the inflow volume predicted by the information processing device 10 is greater than the current inflow volume, the control device 20 performs control to increase the number of pumps P to be operated (the number of pumps P that transport the water to be treated). Furthermore, for example, when the water treatment system 100 has multiple pumps P and the inflow volume predicted by the information processing device 10 is less than the current inflow volume, the control device 20 performs control to decrease the number of pumps P to be operated.

[0019] In addition, for example, when the inflow volume predicted by the information processing device 10 exceeds a threshold value, the control device 20 performs control to limit the inflow volume of the treated water to the pump P by performing closing control of a gate (hereinafter also referred to as the inflow gate) provided upstream of the pump P.

[0020] Here, in the water treatment system 100, it is required to appropriately operate the pump P even when, for example, the inflow amount of the water to be treated into the pump P increases suddenly due to the occurrence of torrential rain or localized heavy rain. Specifically, in the water treatment system 100, it is required to appropriately drain the water to be treated into a river and prevent flood damage to land, buildings, etc., even when, for example, the inflow amount of the water to be treated into the pump P increases suddenly. Therefore, in the water treatment system 100, it is desired to improve the prediction accuracy of the inflow amount of the water to be treated into the pump P, for example.

[0021] Therefore, in the learning process, the information processing device 10 in this embodiment identifies, for example, one or more areas (hereinafter simply referred to as one or more areas) where the measurement state of rainfall intensity satisfies a condition, among multiple areas where the rainfall intensity of rainwater can be measured. The multiple areas where the rainfall intensity of rainwater can be measured are, for example, multiple areas that can be measured by radar R. Then, the information processing device 10 acquires, for example, the rainfall intensity measured in each of the identified one or more areas. Furthermore, the information processing device 10 acquires, for example, the inflow amount of water to be treated that has flowed into the pump P at a timing a predetermined time after the timing of measuring the acquired rainfall intensity.

[0022] After that, the information processing device 10 generates teacher data including at least the rainfall intensity in each of the one or more acquired regions and the acquired inflow amount. Then, the information processing device 10 generates a learning model by, for example, machine learning of the generated teacher data.

[0023] Specifically, the radar R repeatedly measures the rainfall intensity in each of a plurality of regions where the rainfall intensity can be measured, for example, by emitting radio waves and detecting reflected waves of the radio waves. Then, the information processing device 10 identifies, for example, each of the regions where the success probability of the measurement of the rainfall intensity by the radar R is equal to or higher than a threshold as one or more regions.

[0024] That is, the information processing device 10 in the present embodiment identifies each area where it is possible to determine that the success probability of measuring the rainfall intensity of rainwater is high, for example. Then, the information processing device 10 generates a learning model by using the rainfall intensity measured for each identified area, for example.

[0025] This allows the information processing device 10 in this embodiment to suppress the generation of teacher data that does not include an appropriate value for the rainfall intensity of rainwater (hereinafter also referred to as inappropriate teacher data). Therefore, the information processing device 10 can suppress the use of inappropriate teacher data in the generation of a learning model, for example, and can improve the prediction accuracy of the learning model generated in the learning process (prediction accuracy of the inflow amount of water to be treated into the pump P).

[0026] In the following, the water treatment system 100 will be described as having one information processing device 10, but the present invention is not limited to this. Specifically, the water treatment system 100 may have, for example, a plurality of information processing devices 10. The learning process and the prediction process may be performed, for example, in different information processing devices 10 included in the plurality of information processing devices 10. The learning process may be performed, for example, in a distributed manner in two or more information processing devices 10 included in the plurality of information processing devices 10. Similarly, the prediction process may be performed, for example, in a distributed manner in two or more information processing devices 10 included in the plurality of information processing devices 10.

[0027] [Hardware configuration of information processing device 10] Next, a description will be given of the hardware configuration of the information processing device 10. Fig. 2 is a diagram illustrating the hardware configuration of the information processing device 10 in the first embodiment.

[0028] The information processing device 10 is, for example, an electronic device having an electronic circuit. Specifically, as shown in Fig. 2, the information processing device 10 is, for example, a computer device having a CPU (Central Computing Unit) 101, which is a processor, a memory 102, a communication device 103, and a storage medium 104. Each part is connected to each other via, for example, a bus 105.

[0029] The storage medium 104 has a program storage area (not shown) that stores a program 110 for performing, for example, a learning process and a prediction process (hereinafter, these are also collectively referred to as a learning process, etc.).

[0030] The storage medium 104 also includes a storage unit 130 (hereinafter also referred to as an information storage area 130) that stores various information required for executing a learning process, etc. Specifically, the storage medium 104 may be, for example, a hard disk drive (HDD) or a solid state drive (SSD).

[0031] The CPU 101 executes, for example, the program 110 loaded from the storage medium 104 to the memory 102, thereby carrying out a learning process and the like.

[0032] The communication device 103 communicates with, for example, the control device 20. The communication device 103 also communicates with, for example, a radar R and a water level gauge W.

[0033] The electronic circuit of the information processing device 10 may be, for example, a field programmable gate array (FPGA) or an application specific integrated circuit (ASIC). In this case, the learning process or the like may be executed in, for example, the FPGA or the ASIC.

[0034] [Functions of information processing device 10] Next, the functions of the information processing device 10 will be described. Fig. 3 is a block diagram of the information processing device 10 in the first embodiment. Fig. 4 is a diagram for explaining a specific example of the learning process in the first embodiment. Fig. 5 is a diagram for explaining a specific example of the prediction process in the first embodiment.

[0035] As shown in FIG. 3, the information processing device 10 realizes various functions including an area identification unit 111, an intensity acquisition unit 112, a location identification unit 113, a water level acquisition unit 114, an inflow acquisition unit 115, a data generation unit 116, a model generation unit 117, an inflow estimation unit 118, and a result output unit 119, by organic cooperation between hardware such as a CPU 101 and a memory 102 and programs.

[0036] First, the functions in the learning process will be described.

[0037] The area identification unit 111, for example, identifies one or more areas (hereinafter also referred to as target areas) where the measurement state of rainfall intensity DT1 satisfies a condition from among multiple areas (hereinafter also referred to as candidate areas) where the rainfall intensity of rainwater can be measured.

[0038] Specifically, the radar R emits radio waves from an antenna (not shown) of the radar R into space at each measurement timing, and detects the reflected waves of the radio waves to measure the rainfall intensity in the space from which the radio waves are emitted. The measurement timing is, for example, once a minute. Then, the area identification unit 111 acquires, from the radar R, each of the rainfall intensities measured at each measurement timing for each of a plurality of candidate areas as rainfall intensities DT1. After that, the area identification unit 111 stores, for example, each of the acquired rainfall intensities DT1 in the information storage area 130.

[0039] Next, the area identification unit 111, for example, refers to the rainfall intensity DT1 acquired from the radar R (the rainfall intensity DT1 stored in the information storage area 130) and calculates the success probability of the measurement of the rainfall intensity DT1 by the radar R for each of the multiple candidate areas. Specifically, the area identification unit 111, for example, refers to the rainfall intensity DT1 for each measurement timing included in a predetermined period and calculates the success probability of the measurement of the rainfall intensity DT1 by the radar R for each of the multiple candidate areas. The predetermined period is, for example, several minutes or tens of minutes. Then, the area identification unit 111, for example, identifies one or more areas among the multiple candidate areas in which the calculated success probability is equal to or greater than a predetermined threshold as one or more target areas.

[0040] Here, the area specifying unit 111 may determine that the measurement of the rainfall intensity DT1 by the radar R has failed, for example, when the rainfall intensity DT1 acquired from the radar R is a negative value or a predetermined error value. Also, when the area specifying unit 111 compares the rainfall intensity DT1 acquired from the radar R with the rainfall intensity DT1 for another candidate area (for example, a neighboring candidate area) measured at the same measurement timing as the rainfall intensity DT1, and determines that the difference between the rainfall intensities DT1 is equal to or greater than a predetermined value (i.e., when it is determined that the rainfall intensities DT1 are deviated), it may determine that the rainfall intensity DT1 acquired from the radar R is an abnormal value and the measurement of the rainfall intensity DT1 by the radar R has failed. Furthermore, the area identification unit 111 may, for example, compare the rainfall intensity DT1 acquired from the radar R with the rainfall intensity DT1 for the same candidate area measured at a measurement timing different from the rainfall intensity DT1 (for example, the measurement timing immediately before or the measurement timing immediately after), and if it determines that the difference between the rainfall intensities DT1 is a predetermined value or more (i.e., if it determines that the rainfall intensities DT1 are deviating), it may determine that the rainfall intensity DT1 acquired from the radar R is an abnormal value and that the measurement of the rainfall intensity DT1 by the radar R has failed.

[0041] The causes of failure in the measurement of the rainfall intensity DT1 by the radar R are, for example, man-made objects (structures, buildings) and natural objects (mountains, tree leaves, especially leaves with raindrops on them). Measurement may also fail due to the type of tree and the season. For example, if the cause of failure is deciduous trees, the probability of measurement failure in a season with thick leaves may be higher than the probability of measurement failure in a season without leaves.

[0042] In this way, since the measurement fails due to various reasons, the accuracy of the learning model MD is improved by not using the rainfall intensity DT1 of the area where the measurement failed as training data. In particular, considering that the success or failure of the measurement may depend on the season, it is preferable to generate an optimal learning model MD according to the season. Therefore, as described above, the area identification unit 111 performs a process of identifying one or more target areas in advance when the learning model MD is generated, for example.

[0043] The intensity acquisition unit 112 acquires, for example, from the radar R, the rain intensity DT1 measured in each of the one or more target regions identified by the region identification unit 111.

[0044] Specifically, the intensity acquisition unit 112 acquires, for example, the rain intensity DT1 from the radar R for each measurement timing for one or more target regions.

[0045] In this case, the intensity acquisition unit 112 may, for example, acquire the rainfall intensities DT1 for all the candidate regions from the radar R, and then specify the rainfall intensities DT1 for one or more target regions from the acquired rainfall intensities DT1. Also, the intensity acquisition unit 112 may, for example, acquire the rainfall intensity DT1 acquired by an operator from the radar R and input to the information processing device 10.

[0046] The location identification unit 113, for example, identifies one or more locations (hereinafter also referred to as target locations) where the time required for the water to be treated flowing through each location to flow into the pump P (hereinafter also referred to as required arrival time) falls within a predetermined range (hereinafter also referred to as simply the predetermined range) among multiple locations (hereinafter also referred to simply as multiple candidate locations) where the water level DT2 in the pipe L of the water to be treated flowing toward the pump P can be measured. Each of the multiple candidate locations is, for example, a position in the pipe L where a water level meter W can be attached (for example, a position where a manhole, not shown, is installed).

[0047] Specifically, the location identifying unit 113 calculates, for example, each of a plurality of candidate locations, a required arrival time from the distance from each location to the pump P and the flow rate of the water to be treated in the pipe L. Then, for example, the location identifying unit 113 determines whether or not the required arrival time is within a predetermined range for each of a plurality of candidate locations. Thereafter, the location identifying unit 113 identifies, for example, one or more locations among the plurality of candidate locations for which it has determined that the required arrival time is within the predetermined range as one or more target locations. Note that the predetermined range may be, for example, a range previously determined by an operator.

[0048] The water level acquiring unit 114 acquires, for example, for each of the one or more target locations identified by the location identifying unit 113, the water level DT2 measured at each location from a water level gauge W installed at each location.

[0049] Specifically, the water level acquiring unit 114 acquires, for example, the water level DT2 for each measurement timing for one or more target locations from a water level gauge W installed at each location.

[0050] The water level acquiring unit 114 may acquire the water level DT2 acquired by an operator from a water level gauge W installed at each location and input to the information processing device 10, for example.

[0051] The inflow volume acquisition unit 115 acquires, from the pump P, for example, an inflow volume DT3 of the water to be treated that has flowed into the pump P.

[0052] Specifically, the inflow amount acquisition unit 115 acquires, for example, an inflow amount DT3 for each measurement timing.

[0053] The inflow volume acquiring unit 115 may acquire the inflow volume DT3 of the water to be treated from, for example, a flow meter (not shown) provided on the inlet side of the pump P in the pipe L. The inflow volume acquiring unit 115 may acquire the inflow volume DT3 acquired from the flow meter by an operator and input to the information processing device 10.

[0054] 4, the data generating unit 116 generates teacher data DT4 including, for example, the rainfall intensity DT1 (rainfall intensity DT1 for one or more target areas) acquired by the intensity acquiring unit 112, the water level DT2 (water level DT2 for one or more target locations) acquired by the water level acquiring unit 114, and the inflow amount DT3 acquired by the inflow amount acquiring unit 115. Then, the data generating unit 116 stores the generated teacher data DT4 in the information storage area 130, for example.

[0055] Specifically, the data generator 116 generates teacher data DT4 including, for example, rainfall intensity DT1 (rainfall intensity DT1 for one or more target areas) acquired during a target period (hereinafter simply referred to as the target period) designated in advance by an operator, water level DT2 (water level DT2 for one or more target locations) acquired during the target period, and inflow amount DT3 acquired at a timing after a predetermined time in the target period. The target period may be, for example, several hours.

[0056] In this case, the data generating unit 116 repeatedly generates the teacher data DT4 while gradually changing the target period, for example.

[0057] Specifically, the data generation unit 116 may generate teacher data DT4 having a target period of 180 minutes from 12:01 to 15:00, for example, and then generate teacher data DT4 having a target period of 180 minutes from 12:02 to 15:01, and further generate teacher data DT4 having a target period of 180 minutes from 12:03 to 15:02. Hereinafter, a plurality of teacher data DT4 generated while changing the target period will be collectively referred to as a teacher data group DG.

[0058] Furthermore, in this case, the data generator 116 repeatedly generates the teacher data group DG while changing the predetermined time, for example.

[0059] Specifically, the data generation unit 116 may, for example, generate a teacher data group DG when the specified time is 1 minute, then generate a teacher data group DG when the specified time is 2 minutes, and further generate a teacher data group DG when the specified time is 3 minutes.

[0060] The model generation unit 117 generates a learning model MD, for example, by machine learning of the teacher data DT4 (each teacher data DT4 included in the teacher data group DG) generated by the data generation unit 116. Then, the model generation unit 117 stores the generated learning model MD in the information storage area 130, for example.

[0061] Specifically, for example, the model generation unit 117 generates a learning model MD for each of the multiple teacher data groups DG generated by the data generation unit 116, using the teacher data DT4 included in each teacher data group DG.

[0062] That is, in this case, the model generation unit 117 performs machine learning on a teacher data group DG including, for example, teacher data DT4 having a target period of 180 minutes from 12:01 to 15:00 and a specified time of 1 minute, teacher data DT4 having a target period of 180 minutes from 12:02 to 15:01 and a specified time of 1 minute, and teacher data DT4 having a target period of 180 minutes from 12:03 to 15:02 and a specified time of 1 minute, and generates a learning model MD (first learning model MD) capable of predicting the inflow volume DT3 one minute later in the prediction process.

[0063] In this case, the model generation unit 117 performs machine learning on a teacher data group DG including, for example, teacher data DT4 having a target period of 180 minutes from 12:01 to 15:00 and a specified time of 2 minutes, teacher data DT4 having a target period of 180 minutes from 12:02 to 15:01 and a specified time of 2 minutes, and teacher data DT4 having a target period of 180 minutes from 12:03 to 15:02 and a specified time of 2 minutes, and generates a learning model MD (second learning model MD) capable of predicting the inflow volume DT3 two minutes later in the prediction process.

[0064] Furthermore, in this case, the model generation unit 117 performs machine learning on a teacher data group DG including, for example, teacher data DT4 having a target period of 180 minutes from 12:01 to 15:00 and a specified time of 60 minutes, teacher data DT4 having a target period of 180 minutes from 12:02 to 15:01 and a specified time of 60 minutes, and teacher data DT4 having a target period of 180 minutes from 12:03 to 15:02 and a specified time of 60 minutes, and so on, to generate a learning model MD (the 60th learning model MD) capable of predicting the inflow volume DT3 60 minutes later in the prediction process.

[0065] In the following, a case where both the rainfall intensity DT1 and the water level DT2 are included in the teacher data DT4 will be described, but this is not limited to the case. Specifically, the teacher data DT4 may include, for example, either the rainfall intensity DT1 or the water level DT2.

[0066] Next, the functions in the prediction process will be described.

[0067] The intensity acquisition unit 112 acquires, for example, from the radar R, the rain intensity DT11 measured in each of the one or more target regions identified by the region identification unit 111 in the learning process.

[0068] Specifically, for example, in response to an operator inputting information to the information processing device 10 that a prediction process will be performed, the intensity acquisition unit 112 acquires from the radar R the rainfall intensity DT1 for each measurement timing included in the most recent target period (for example, the target period from 180 minutes to 1 minute before the time when the operator inputs the information to perform prediction processing) for one or more target areas.

[0069] In this case, the intensity acquisition unit 112 may acquire the rain intensities DT11 for one or more target regions from the acquired rain intensities DT11 after acquiring the rain intensities DT11 for all the candidate regions from the radar R. Also, the intensity acquisition unit 112 may acquire the rain intensity DT11 acquired by an operator from the radar R and input to the information processing device 10.

[0070] For example, the water level acquisition unit 114 acquires the water level DT12 measured at each of the one or more target locations identified by the location identification unit 113 in the learning process from a water level gauge W installed at each location.

[0071] Specifically, the water level acquisition unit 114 acquires each water level DT12 for one or more target locations for each measurement timing included in the most recent target period (for example, a target period from 180 minutes to 1 minute before the timing when the worker inputs the information to the effect that the prediction process will be performed) in response to, for example, the worker inputting information to the information processing device 10 that the prediction process will be performed. That is, the water level acquisition unit 114 acquires each water level DT12 having the same measurement timing as each rainfall intensity DT11 acquired by the intensity acquisition unit 112, for example.

[0072] The water level acquiring unit 114 may acquire the water level DT12 acquired by an operator from a water level gauge W installed at each location and input to the information processing device 10, for example.

[0073] As shown in Figure 5, for example, the inflow estimation unit 118 inputs the rainfall intensity DT11 for one or more target areas acquired by the intensity acquisition unit 112 and the water level DT12 for one or more target locations acquired by the water level acquisition unit 114 into the learning model MD stored in the information storage area 130, and acquires the inflow volume DT13 output from the learning model MD in conjunction with the input of each rainfall intensity DT11 and each water level DT12.

[0074] Specifically, the operator inputs, for example, the time from the measurement timing of each rainfall intensity DT11 and each water level DT12 to the timing (hereinafter also referred to as the prediction timing) of predicting the inflow amount DT13. In this case, the inflow amount estimation unit 118 identifies, for example, a learning model MD (hereinafter also referred to as a specific learning model MD) that can make a prediction corresponding to the input time, among the learning models MD stored in the information storage area 130. More specifically, for example, when the time input by the operator is 2 minutes, the inflow amount estimation unit 118 identifies, as the specific learning model MD, a learning model MD that can make a prediction when the predetermined time is 2 minutes, among the learning models MD stored in the information storage area 130. Then, the inflow amount estimation unit 118 inputs, for example, the rainfall intensity DT11 for each of one or more target areas acquired by the intensity acquisition unit 112 and the water level DT12 for each of one or more target locations acquired by the water level acquisition unit 114, to the identified specific learning model MD.

[0075] The result output unit 119 outputs, for example, the inflow volume DT13 acquired by the inflow volume estimation unit 118 to an output device (not shown) of the information processing device 10 as a predicted result of the inflow volume DT3 at a timing a predetermined time after the measurement timing of each rainfall intensity DT11 and each water level DT12.

[0076] [Flowchart of learning process] Next, a description will be given of a flowchart of the learning process in the information processing device 10. Fig. 6 is a flowchart for explaining the learning process in the first embodiment. Figs. 7 to 11 are diagrams for explaining the learning process.

[0077] The region specifying unit 111 of the information processing device 10 specifies, for example, one or more target regions in which the measurement state of the rainfall intensity DT1 satisfies a condition from among a plurality of candidate regions (step S1 in FIG. 6).

[0078] Specifically, the area specifying unit 111 acquires, for example, the rainfall intensity DT1 for each of the plurality of candidate areas and for each measurement timing from the radar R. Then, the area specifying unit 111, for example, refers to the rainfall intensity DT1 acquired from the radar R and calculates the success probability of the measurement of the rainfall intensity DT1 by the radar R for each of the plurality of candidate areas. After that, the area specifying unit 111, for example, specifies, from among the plurality of candidate areas, one or more areas whose calculated success probability is equal to or greater than a predetermined threshold, as one or more target areas. A specific example of step S1 will be described below.

[0079] [Specific example of step S1] 7 to 10 are diagrams for explaining a specific example of step S1. Specifically, FIG. 7 and the like show a specific example of a map M including a pump P and a pipe L.

[0080] 7, the map M is divided into, for example, a plurality of rectangular regions M0. Each rectangular region M0 is, for example, a unit of area in which rainfall intensity DT1 can be measured by a radar R. The map M displays, for example, a pump P (a building in which the pump P is installed) and a pipe L extending to the pump P.

[0081] 8, the area specifying unit 111, for example, refers to information indicating the position of a branch pipe (not shown) communicating with the pipe L among the information stored in the information storage area 130, and specifies one or more areas (hereinafter also referred to as candidate areas M1) in which a branch pipe communicating with the pipe L is buried underground from among the multiple rectangular areas M0. That is, each of the one or more candidate areas M1 is, for example, an area where rainwater that has fallen on the ground flows into the pump P through a manhole and a branch pipe.

[0082] Next, the region identifying unit 111, for example, refers to the rainfall intensity DT1 (the rainfall intensity DT1 acquired from the radar R) stored in the information storage area 130, and calculates the success probability of measuring the rainfall intensity DT1 by the radar R for each of one or more candidate regions M1. Then, as shown in Fig. 9, for example, the region identifying unit 111 identifies one or more regions (hereinafter also referred to as excluded regions M2) among the multiple rectangular regions M0, in which the calculated success probability is smaller than a predetermined threshold.

[0083] Then, as shown in FIG. 10, the area identification unit 111 identifies, for example, each area obtained by excluding one or more exclusion areas M2 from one or more candidate areas M1 as one or more areas (hereinafter also referred to as target areas M3) in which the measurement state of rainfall intensity DT1 satisfies the conditions.

[0084] In this manner, the information processing device 10 in the present embodiment, for example, excludes the rainfall intensities DT1 measured for each excluded region M2 from the rainfall intensities DT1 used to generate the teacher data DT4 in advance. Then, the information processing device 10 generates the learning model MD by using the teacher data DT4 including the rainfall intensities DT1 measured for each region other than the excluded regions M2.

[0085] This allows the information processing device 10 in this embodiment to suppress the generation of inappropriate teacher data DT4, for example. Therefore, the information processing device 10 can suppress the use of inappropriate teacher data DT4 in the generation of the learning model MD, for example, and can improve the prediction accuracy of the learning model MD generated in the learning process (the prediction accuracy of the inflow amount of the water to be treated into the pump P).

[0086] Returning to FIG. 6, the intensity acquisition unit 112 of the information processing device 10 acquires, for example, the rainfall intensity DT1 measured in each of the one or more target regions identified in step S1 (step S2 in FIG. 6).

[0087] Specifically, the intensity acquisition unit 112 continuously acquires the rain intensity DT1 measured by the radar R for each measurement timing and for each of one or more target regions, for example.

[0088] Next, the location identification unit 113 of the information processing device 10 identifies, for example, one or more target locations from among multiple candidate locations, where the time required for rainwater flowing through each location to flow into the pump P is within a predetermined range (step S3 in Figure 6).

[0089] Specifically, the location identifying unit 113 calculates, for example, for each of a plurality of candidate locations, a required arrival time from the distance from each location to the pump P and the flow rate of rainwater in the pipe L. Then, the location identifying unit 113 determines, for example, for each of a plurality of candidate locations, whether or not the required arrival time is within a predetermined range. After that, the location identifying unit 113 identifies, for example, one or more locations among the plurality of candidate locations whose required arrival time falls within the predetermined range as one or more target locations. A specific example of step S3 will be described below.

[0090] [Specific example of step S3] 11 is a diagram for explaining a specific example of step S3. Specifically, FIG. 11 is a specific example of a map M including a pump P and a pipe L.

[0091] The operator, for example, determines the range of time (the time from the measurement timing of each rainfall intensity DT11 and each water level DT12 to the prediction timing of the inflow amount DT13) for which the inflow amount of the water to be treated to the pump P needs to be predicted with the highest accuracy, and inputs the determined range to the information processing device 10. Then, the location identification unit 113, for example, calculates the required arrival time for each of the multiple candidate locations by dividing the distance from each location to the pump P by the flow rate of the rainwater in the pipe L. The flow rate of the rainwater in the pipe L may be, for example, the flow rate when a torrential rain or localized heavy rain occurs in at least one of the candidate areas M1 described in FIG. 18. Then, the location identification unit 113, for example, determines the range of time input by the operator as a predetermined range, and identifies one or more locations among the multiple candidate locations where the required arrival time of the rainwater to the pump P is included within the predetermined range as one or more target locations.

[0092] Specifically, the location specifying unit 113 attaches a water level gauge W1 and a water level gauge W2 at two locations in the pipe L, for example, as shown in FIG.

[0093] Returning to FIG. 6, the water level acquisition unit 114 of the information processing device 10 acquires, for example, the water level DT2 measured at each of the one or more target locations identified in step S3 (step S4 in FIG. 6).

[0094] Specifically, the water level acquiring unit 114 continuously acquires the water level DT2 measured at each location for each measurement timing and for one or more target locations, for example.

[0095] Next, the inflow amount acquisition unit 115 of the information processing device 10 acquires, for example, an inflow amount DT3 of the water to be treated that has flowed into the pump P (step S5 in FIG. 6).

[0096] Specifically, the inflow volume acquisition unit 115 continuously acquires the inflow volume DT3 of the water to be treated that has flowed into the pump P at each measurement timing, for example.

[0097] Thereafter, the data generating unit 116 of the information processing device 10 generates teacher data DT4 including, for example, the rainfall intensity DT1 (rainfall intensity DT1 for one or more target areas) acquired in step S2, the water level DT2 (water level DT2 for one or more target locations) acquired in step S4, and the inflow amount DT3 acquired in step S5 (step S6 in FIG. 6).Then, the data generating unit 116 stores the generated teacher data DT4 in the information storage area 130, for example.

[0098] Specifically, the data generation unit 116 generates teacher data DT4 that includes, for example, rainfall intensity DT1 obtained within the target period (rainfall intensity DT1 for one or more target areas), water level DT2 obtained within the target period (water level DT2 for one or more target locations), and inflow volume DT3 obtained at a predetermined time after the target period.

[0099] In this case, the data generator 116 generates the teacher data group DG including a plurality of teacher data DT4, for example, by repeatedly generating the teacher data DT4 while changing the target period.

[0100] Furthermore, the data generator 116 generates a plurality of teacher data groups DG by repeatedly generating the teacher data group DG while changing the predetermined time, for example. A specific example of the teacher data DT4 will be described below.

[0101] [Specific example of training data DT4] 12 and 13 are diagrams for explaining a specific example of the teacher data DT4.

[0102] Specifically, Fig. 12 is a diagram illustrating a specific example of teacher data DT (hereinafter also referred to as teacher data DT4a) included in teacher data group DG when the predetermined time is 1 minute, and Fig. 13 is a diagram illustrating a specific example of teacher data DT (hereinafter also referred to as teacher data DT4b) included in teacher data group DG when the predetermined time is 2 minutes.

[0103] In the following, a case where the target periods corresponding to the teacher data DT4a and the teacher data DT4b are the same (for example, 180 minutes from 12:01 to 15:00) will be described. In addition, a case where the predetermined time varies in 1-minute increments between 1 minute and 60 minutes will be described.

[0104] The teacher data DT4 shown in FIG. 12 and the like has, for example, items of "rainfall intensity (1)" in which the rainfall intensity DT1 measured for a part of the one or more target regions M3 described in FIG. 10 (hereinafter also referred to as the first region) is set, and "rainfall intensity (2)" in which the rainfall intensity DT1 measured for a part of the one or more target regions M3 described in FIG. 10 that is different from the first region (hereinafter also referred to as the second region) is set. The teacher data DT4 shown in FIG. 12 and the like has, for example, items of "rainfall intensity (3)" in which the rainfall intensity DT1 measured for a part of the one or more target regions M3 described in FIG. 10 that is different from the first region and the second region (hereinafter also referred to as the third region). Specifically, each of the first region, the second region, and the third region may be, for example, a region obtained by dividing each target region M3 described in FIG. 10 into three. Furthermore, "Rainfall Intensity (1)" may be set to, for example, the sum of rainfall intensities DT1 measured for one or more target regions M3 included in the first region. Furthermore, "Rainfall Intensity (2)" may be set to, for example, the sum of rainfall intensities DT1 measured for one or more target regions M3 included in the second region. Furthermore, "Rainfall Intensity (3)" may be set to, for example, the sum of rainfall intensities DT1 measured for one or more target regions M3 included in the third region.

[0105] In addition, the teacher data DT4 shown in Figure 12 etc. has items such as "water level (1)" in which the water level DT2 measured by the water level gauge W1 described in Figure 11 is set, and "water level (2)" in which the water level DT2 measured by the water level gauge W2 is set.

[0106] Furthermore, the teacher data DT4 shown in FIG. 12 and the like has, for example, an "inflow amount" field in which an inflow amount DT13 that flows into the pump P after a predetermined time period of the target period is set.

[0107] Specifically, in the teacher data DT4a shown in FIG. 12, for example, in the column corresponding to "1st minute" (the column in which data measured at the measurement timing of the 1st minute within the target period is set), "Rainfall Intensity (1)", "Rainfall Intensity (2)", "Rainfall Intensity (3)", "Water Level (1)", and "Water Level (2)" are set to "0 (mm / hr)", "0 (mm / hr)", "1 (mm / hr)", "0.1 (m)", and "0.05 (m)".

[0108] Furthermore, in the teacher data DT4a shown in FIG. 12, for example, in the column corresponding to the "2nd minute" (the column in which data measured at the measurement timing of the 2nd minute within the target period is set), "Rainfall Intensity (1)", "Rainfall Intensity (2)", "Rainfall Intensity (3)", "Water Level (1)", and "Water Level (2)" are set to "0 (mm / hour)", "0 (mm / hour)", "2 (mm / hour)", "0.1 (m)", and "0.05 (m)".

[0109] In addition, in the teacher data DT4a shown in FIG. 12, for example, in the column corresponding to the "180th minute" (the column in which the data measured at the measurement timing of the 180th minute within the target period is set), "Rainfall Intensity (1)", "Rainfall Intensity (2)", "Rainfall Intensity (3)", "Water Level (1)", and "Water Level (2)" are set to "6 (mm / hour)", "2 (mm / hour)", "0 (mm / hour)", "0.25 (m)", and "0.2 (m)".

[0110] In addition, in the teacher data DT4a shown in FIG. 12, for example, in the column corresponding to “1 minute later” (the column in which data measured at the measurement timing 1 minute later in the target period is set), “inflow amount” is set to “190 (m 3 12 ( / min)" is set. Explanation of other data included in the teacher data DT4a shown in FIG. 12 will be omitted.

[0111] In addition, in the teacher data DT4b shown in FIG. 13, “Rainfall Intensity (1)”, “Rainfall Intensity (2)”, “Rainfall Intensity (3)”, “Water Level (1)” and “Water Level (2)” are set to the same data as in the teacher data DT4a described in FIG. 12, for example.

[0112] In the teacher data DT4b shown in FIG. 13, for example, in the column corresponding to “2 minutes later” (the column in which the data measured at the measurement timing 2 minutes later in the target period is set), “210 (m 3 / min) is set.

[0113] Returning to Fig. 6, the model generation unit 117 of the information processing device 10 generates a learning model MD by machine learning of the teacher data DT4 generated in step S6 (step S7 in Fig. 6). That is, the model generation unit 117 trains the learning model MD on the teacher data DT4 generated in step S6, for example. Then, the model generation unit 117 stores the generated learning model MD in the information storage area 130, for example.

[0114] Specifically, the model generation unit 117 generates a learning model MD for each of the multiple teacher data groups DG generated in step S6 by machine learning of the teacher data DT4 included in each teacher data group DG.

[0115] More specifically, the model generation unit 117 generates, for example, a first learning model MD (a learning model MD capable of predicting the inflow volume DT13 one minute later in the prediction process) by machine learning of a teacher data group DG including the teacher data DT4a described in Figure 12, and generates a second learning model MD (a learning model MD capable of predicting the inflow volume DT13 two minutes later in the prediction process) by machine learning of a teacher data group DG including the teacher data DT4b described in Figure 13.

[0116] In the above example, the case where all the teacher data DT4 used to generate the learning model MD is generated, and then the learning model MD is generated using the generated teacher data DT4 has been described, but the present invention is not limited to this. Specifically, the information processing device 10 may, for example, gradually increase the judgment accuracy of the learning model MD by repeatedly performing the processes from step S1 to step S7. That is, the information processing device 10 may, for example, gradually increase the judgment accuracy of the learning model MD by repeatedly performing the generation and learning of a portion of the teacher data DT4 used to generate the learning model MD.

[0117] This allows the operator to, for example, proceed with the generation and learning of the teacher data DT4 little by little while referring to the judgment accuracy of the learning model MD, and prevents the teacher data DT4 from being generated more than necessary. Therefore, the operator can, for example, reduce the workload required for generating the teacher data DT4.

[0118] Furthermore, when the information processing device 10 repeatedly performs the processes from step S1 to step S7, for example, the information processing device 10 may not perform step S1 every time. Specifically, the information processing device 10 may perform step S1 once every time the information processing device 10 performs steps S2, S4, S5, S6, and S7 a predetermined number of times.

[0119] Similarly, when the information processing device 10 repeatedly performs the processes from step S1 to step S7, for example, the information processing device 10 may not perform step S3 every time. Specifically, the information processing device 10 may perform step S3 once every time the information processing device 10 performs steps S2, S4, S5, S6, and S7 a predetermined number of times.

[0120] Furthermore, when repeatedly performing the processes from step S1 to step S7, the information processing device 10 may, for example, specify an area that has been specified as a target area a predetermined number of times in succession in step S1 as an area to be fixed to the target area (hereinafter, also referred to as a fixed target area). Then, in step S1, the information processing device 10 may, for example, specify one or more target areas from areas that have not been specified as fixed target areas in each of the previous steps S1. After that, in step S2, the information processing device 10 may, for example, acquire rainfall intensity DT1 measured in each of the one or more target areas specified in the previous step S1 and the one or more fixed target areas specified in each of the previous steps S1.

[0121] Similarly, when repeatedly performing the processes from step S1 to step S7, the information processing device 10 may, for example, specify a location that has been specified as a target location a predetermined number of times in succession in step S3 as a location to be fixed to the target location (hereinafter also referred to as a fixed target location). Then, in step S3, the information processing device 10 may, for example, specify one or more target locations from locations that have not been specified as a fixed target location in each of the previous steps S3. After that, in step S4, the information processing device 10 may, for example, acquire the water level DT2 measured at each of the one or more target locations specified in the previous step S3 and the one or more fixed target locations specified in each of the previous steps S3.

[0122] This allows the information processing device 10 to, for example, reduce the amount of processing required to generate the learning model MD. Therefore, the information processing device 10 can, for example, shorten the time required to generate the learning model MD.

[0123] [Flowchart of prediction process] Next, a description will be given of a flowchart of the prediction process in the information processing device 10. Fig. 14 is a flowchart illustrating the prediction process in the first embodiment.

[0124] As shown in FIG. 14, the intensity acquisition unit 112 of the information processing device 10 acquires, for example, the rainfall intensity DT11 measured in each area for one or more target areas identified by the area identification unit 111 in step S1 of FIG. 6 (step S11 of FIG. 14).

[0125] Specifically, for example, in response to an operator inputting information to the information processing device 10 that a prediction process will be performed, the intensity acquisition unit 112 acquires from the radar R the rainfall intensity DT11 for each measurement timing included in the most recent target period (for example, the target period from 180 minutes to 1 minute before the time when the operator inputs the information to perform the prediction process) for one or more target areas.

[0126] In addition, the water level acquisition unit 114 of the information processing device 10 acquires the water level DT12 measured at each location, for example, for one or more target locations identified by the location identification unit 113 in step S3 of Figure 6 (step S12 of Figure 14).

[0127] Specifically, the water level acquisition unit 114 acquires each water level DT12 for one or more target locations for each measurement timing included in the most recent target period (for example, a target period from 180 minutes to 1 minute before the timing when the worker inputs the information to the effect that the prediction process will be performed) in response to, for example, the worker inputting information to the information processing device 10 that the prediction process will be performed. That is, the water level acquisition unit 114 acquires each water level DT12 having the same measurement timing as each rainfall intensity DT11 acquired by the intensity acquisition unit 112, for example.

[0128] Then, the inflow estimation unit 118 of the information processing device 10 inputs, for example, the rainfall intensity DT11 for each of the one or more target areas obtained in step S11 and the water level DT12 for each of the one or more target locations obtained in step S12 into the learning model MD stored in the information storage area 130, and obtains the inflow volume DT13 output from the learning model MD in response to the input of the rainfall intensity DT11 and the water level DT12 (step S13 in Figure 14).

[0129] Specifically, the operator inputs, for example, the time from the measurement timing of the rainfall intensity DT11 and the water level DT12 to the prediction timing of the inflow volume DT13. Then, the inflow volume estimation unit 118 identifies, for example, a specific learning model MD capable of making a prediction corresponding to the input time, from among the learning models MD stored in the information storage area 130. After that, the inflow volume estimation unit 118 inputs, for example, the rainfall intensity DT11 for each of one or more target areas acquired in step S11 and the water level DT12 for each of one or more target locations acquired in step S12 to the identified specific learning model MD.

[0130] Thereafter, the result output unit 119 of the information processing device 10 outputs, for example, the inflow volume DT13 acquired in step S13 to the output device of the information processing device 10 as a predicted result of the inflow volume DT3 at a timing a predetermined time after the measurement timing of the rainfall intensity DT11 and the water level DT12 (step S14 in Figure 14).

[0131] Specifically, for example, when a learning model MD capable of making predictions when the specified time is two minutes is used in step S13, the result output unit 119 outputs the inflow volume DT13 obtained in step S13 as a prediction result of the inflow volume DT3 two minutes after the measurement timing of the rainfall intensity DT11 and the water level DT12 (i.e., the most recent target period).

[0132] In this manner, for example, in the learning process, the information processing device 10 in this embodiment generates a learning model MD that predicts the inflow amount DT3 of the water to be treated flowing into the pump P that treats the water to be treated, including rainwater.

[0133] Specifically, the information processing device 10 in this embodiment, for example, identifies one or more target regions M3 in which the measurement state of the rainfall intensity DT1 of rainwater satisfies a condition among multiple candidate regions M1 in which the rainfall intensity DT1 of rainwater can be measured. Then, for example, the information processing device 10 acquires the rainfall intensity DT1 at a first timing for each of the identified one or more target regions M3. Also, the information processing device 1 acquires an inflow amount DT3 at a second timing, which is a predetermined time after the first timing. After that, the information processing device 10 generates, for example, teacher data DT4 including the rainfall intensity DT1 in each of the acquired one or more target regions M3 and the acquired inflow amount DT3. Then, the information processing device 10 generates a learning model MD by, for example, machine learning the generated teacher data DT4.

[0134] More specifically, in the information processing device 10 of the present embodiment, the rainfall intensity DT1 in each of the multiple candidate regions M1 is repeatedly measured, for example, by using a radar R that emits radio waves and detects the reflected waves of the radio waves. Then, the information processing device 10 specifies, for example, each of the multiple candidate regions M1 in which the success probability of measuring the rainfall intensity DT1 by the radar R is equal to or higher than a threshold, as one or more target regions M3.

[0135] Furthermore, the information processing device 10 in this embodiment identifies, for example, one or more target locations among a plurality of candidate locations where the water level DT2 in the pipe L of the water to be treated flowing toward the pump P can be measured, where the time required for the water to be treated flowing through each location to flow into the pump P is within a predetermined range. Then, the information processing device 10 acquires, for example, the water level DT2 at each location at a first timing for each of the identified one or more target locations. After that, the information processing device 10 generates teacher data DT4 including, for example, the rainfall intensity DT1 in each of the one or more target regions M3, the water level DT2 in each of the one or more target locations, and the inflow amount DT3 of the water to be treated flowing into the pump P.

[0136] Furthermore, in the prediction process, the information processing device 10 in this embodiment predicts the inflow volume DT3 using a learning model MD that predicts the inflow volume DT3 of the water to be treated flowing into a pump P that treats the water to be treated, including rainwater.

[0137] Specifically, in the prediction process, the information processing device 10 in this embodiment acquires new rainfall intensity DT11 of rainwater at the third timing for each of one or more target regions M3. The information processing device 10 also inputs new rainfall intensity DT11 for each of one or more target regions M3 to the learning model MD, and acquires a new inflow amount DT13 output from the learning model MD in conjunction with the input of the new rainfall intensity DT11. The information processing device 10 then outputs the acquired new inflow amount DT13 as a prediction result of the inflow amount DT3 at the fourth timing a predetermined time after the third timing.

[0138] More specifically, the information processing device 10 in this embodiment acquires, for example, a new water level in the pipe L of the water to be treated at the third timing for each of one or more target locations. After that, the information processing device 1 inputs, for example, a new rainfall intensity DT11 in each of one or more target regions M3 and a new water level DT12 in each of one or more target locations to the learning model MD, and acquires a new inflow amount DT13 output from the learning model MD in conjunction with the input of the new rainfall intensity DT11 and the new water level DT12.

[0139] That is, the information processing device 10 in this embodiment, for example, identifies each area where it is possible to determine that the probability of successful measurement of the rainfall intensity DT1 is high, and generates a learning model MD by using the rainfall intensity DT1 in each identified area.

[0140] This allows the information processing device 10 in the present embodiment to suppress the generation of inappropriate teacher data DT4, for example. Therefore, the information processing device 10 can suppress the use of inappropriate teacher data DT4 in the generation of the learning model MD, for example, and can improve the prediction accuracy of the learning model MD generated in the learning process (the prediction accuracy of the inflow amount DT3 of the water to be treated into the pump P).

[0141] Furthermore, the information processing device 10 in this embodiment can, for example, improve the prediction accuracy of the inflow amount DT3 of the water to be treated into the pump P, thereby reducing the workload of the operator in operating the pump P. In other words, the information processing device 10 can, for example, improve the prediction accuracy of the inflow amount DT3 of the water to be treated into the pump P, thereby reducing the number of cases in which the operator's judgment based on the practical experience is required. Therefore, the information processing device 10 can, for example, enable the operation of the pump P by an operator with little practical experience. Furthermore, the information processing device 10 can, for example, further promote automation of the operation of the pump P by the control device 20. [Explanation of symbols]

[0142] 10: Information processing device 20: Control device 100: Water treatment system 101: CPU 102: Memory 103: Communication device 104: Storage medium 105: Bus 130: Information storage area DT1: Rainfall intensity DT2: Water level DT3: Inflow amount DT4: Training data DT11: Rainfall intensity DT12: Water level DT13: Inflow amount L: Piping M: Map M0: Rectangular area M1: Candidate area M2: Exclusion area M3: Target area MD: Learning model P: Pump R: Radar W: Water level gauge W1: Water level gauge W2: Water level gauge

Claims

1. A learning device that generates a learning model for predicting an inflow amount of water to be treated that flows into a water treatment device that treats water to be treated, including rainwater, an area specifying unit that specifies one or more areas in which the measurement state of the rainfall intensity satisfies a condition among a plurality of areas in which the rainfall intensity of the rainwater can be measured; an intensity acquisition unit that acquires the rainfall intensity at a first timing for each of the one or more identified regions; an inflow amount acquisition unit that acquires the inflow amount at a second timing that is a predetermined time after the first timing; A data generating unit that generates teacher data including the rainfall intensity in each of the one or more acquired regions and the acquired inflow amount; A model generation unit that generates the learning model by machine learning of the generated teacher data. A learning device characterized by:

2. the rainfall intensity in each of the plurality of regions is repeatedly measured by using a radar that emits radio waves and detects reflected waves of the radio waves; The learning device according to claim 1 , wherein the region identifying unit identifies, as the one or more regions, each of the plurality of regions in which a success probability of the measurement of the rainfall intensity by the radar is equal to or higher than a threshold value.

3. Furthermore, a location identification unit that identifies one or more locations among a plurality of locations where the water level in the pipe of the water to be treated flowing toward the water treatment device can be measured, the location identification unit identifying one or more locations where the time required for the water to be treated flowing through each location to flow into the water treatment device is within a predetermined range. a water level acquisition unit that acquires the water level at the first timing for each of the one or more identified locations, The learning device according to claim 1 , wherein the data generation unit generates the teacher data including the rainfall intensity in each of the one or more regions, the water level in each of the one or more locations, and the inflow amount.

4. A prediction device for predicting an inflow amount of water to be treated that flows into a water treatment device that treats water to be treated, including rainwater, using a learning model for predicting the inflow amount of the water to be treated, a memory unit that stores a learning model generated by machine learning of teacher data including the rainfall intensity at a first timing in one or more regions where the rainfall intensity of the rainwater can be measured and the inflow amount at a second timing a predetermined time after the first timing in each of the one or more regions where the measured state of the rainfall intensity satisfies a predetermined condition; an intensity acquisition unit that acquires a new rainfall intensity of the rainwater at a third timing for each of the one or more regions; an inflow amount estimation unit that inputs the new rainfall intensity in each of the one or more regions to the learning model and acquires a new inflow amount output from the learning model in response to the input of the new rainfall intensity; and a result output unit that outputs the acquired new inflow amount as the inflow amount at a fourth timing that is the predetermined time after the third timing. A prediction device comprising:

5. The memory unit stores the learning model generated by machine learning of the teacher data, which includes the rainfall intensity of each of the one or more regions, and the water level and the inflow amount at the first timing of each of one or more locations among a plurality of locations where the water level in the pipe of the treated water can be measured, the time required for the treated water flowing through each location to flow into the water treatment device being within a predetermined range, and further A water level acquisition unit is provided for each of the one or more locations to acquire a new water level in the pipe of the water to be treated at the third timing of each location, The prediction device of claim 4, wherein the inflow estimation unit inputs the new rainfall intensity in each of the one or more areas and the new water level in each of the one or more locations to the learning model, and obtains the new inflow output from the learning model in response to the input of the new rainfall intensity and the new water level.

6. A learning method for generating a learning model for predicting an inflow amount of treated water flowing into a water treatment device that treats treated water including rainwater, comprising: Identifying one or more areas where the measurement state of the rainfall intensity satisfies a condition among a plurality of areas where the rainfall intensity of the rainwater can be measured; Acquire the rainfall intensity at the first timing for each of the one or more identified regions; acquiring the inflow amount at a second timing that is a predetermined time after the first timing; generating training data including the rainfall intensity in each of the one or more acquired regions and the acquired inflow amount; A learning method comprising: a computer executing a process comprising:

7. A method for predicting an inflow amount of water to be treated that flows into a water treatment device that treats water to be treated, including rainwater, using a learning model for predicting the inflow amount, storing, in a storage unit, a learning model generated by machine learning of training data including the rainfall intensity at a first timing in one or more regions where the measurement state of the rainfall intensity of the rainwater can be measured and the inflow amount at a second timing a predetermined time after the first timing; Acquire a new rainfall intensity of the rainwater at a third timing for each of the one or more regions; inputting the new rainfall intensity in each of the one or more regions into the learning model, and acquiring a new inflow amount output from the learning model in response to the input of the new rainfall intensity; outputting the acquired new inflow amount as the inflow amount at a fourth timing that is the predetermined time after the third timing; A prediction method for causing a computer to execute a process comprising the steps of:

8. a water treatment device for treating water to be treated, including rainwater; A prediction device that predicts an inflow amount of the water to be treated flowing into the water treatment device using a learning model; A water treatment system comprising: a control device that controls the water treatment device based on the inflow amount predicted by the prediction device, The prediction device comprises: a memory unit that stores a learning model generated by machine learning of teacher data including the rainfall intensity at a first timing in one or more regions where the rainfall intensity of the rainwater can be measured and the inflow amount at a second timing a predetermined time after the first timing in each of the one or more regions where the measured state of the rainfall intensity satisfies a predetermined condition; an intensity acquisition unit that acquires a new rainfall intensity of the rainwater at a third timing for each of the one or more regions; an inflow amount estimation unit that inputs the new rainfall intensity in each of the one or more regions to the learning model and acquires a new inflow amount output from the learning model in response to the input of the new rainfall intensity; and a result output unit that outputs the acquired new inflow amount as the inflow amount at a fourth timing that is the predetermined time after the third timing. A water treatment system comprising:

Images