JPWO2020090104A1 - Water level prediction device and water level prediction method - Google Patents

Abstract

In the water level predictor (1), the virtual pump well model (20) is used to learn the pump operation parameters related to the pump operation, and the water level change showing the relationship between the river water level and the water level increase amount according to the pump operation. The parameters are learned and the water level of the river is predicted based on these parameters.

Description

The present invention relates to a water level prediction device and a water level prediction method.

For example, as a conventional technique for predicting the water level of a river, the water level prediction method described in Patent Document 1 has been proposed. In this method, the water level of a river is predicted by predicting the amount of water flowing out from the basin to the river using the storage function method.

JP 2015-94122

In the process of changing the water level of a river, in addition to the case where the amount of rainwater that has fallen in the river basin is saturated and continuously flows out to the river and the water level of the river changes, the water is dispersed to the river by a pump. The water level of the river may change due to the outflow to the river. For example, in a basin where the water level is lower than the river level and it is difficult for rainwater to flow into the river, rainwater is collected in the sewer and discharged to the river by a pump in order to prevent inundation damage caused by rainwater. However, since the storage function method cannot sufficiently model the discrete water outflow process to the river by the pump, the water level prediction method described in Patent Document 1 determines the water level of the river where the water is discharged by the pump. There was a problem that it could not be predicted accurately.

The present invention solves the above problems, and an object of the present invention is to obtain a water level prediction device and a water level prediction method capable of accurately predicting the water level of a river in which water is discharged by a pump.

The water level predictor according to the present invention is a virtual pump well model, which assumes a virtual pump well in which rainwater that has fallen in a river basin is stored and a pump operates according to the stored height to cause water to flow out to the river. The first learning unit that learns the pump operation parameters related to the pump operation using the past rainfall data and the pump operation history data in the virtual pump well model, and the pump operation parameters are used to respond to the pump operation. It includes a second learning unit that learns a water level change parameter that indicates the relationship between the water level of the river and the amount of increase in the water level, and a water level prediction unit that predicts the water level of the river based on the pump operation parameter and the water level change parameter.

According to the present invention, assuming a virtual pump well in which rainwater that has fallen in a river basin is stored and the pump operates according to the stored height to discharge water to the river, the pump operation parameters related to the pump operation are set. It learns and learns the water level change parameters that indicate the relationship between the water level of the river and the amount of increase in the water level according to the operation of the pump, and predicts the water level of the river based on these parameters. As a result, the change in the water level of the river according to the operation of the pump can be accurately modeled, so that the water level of the river in which the water is discharged by the pump can be predicted accurately.

It is a block diagram which shows the structure of the water level prediction apparatus which concerns on Embodiment 1. FIG. FIG. 2A is a diagram showing an outline of a virtual pump well model in which the storage height is below the threshold value. FIG. 2B is a diagram showing an outline of a virtual pump well model in which the storage height exceeds the threshold value. It is a flowchart which shows the water level prediction method which concerns on Embodiment 1. It is a figure which shows the outline of the virtual pump well model in Embodiment 1. FIG. FIG. 5A is a diagram showing the relationship between the water level of the river and the amount of increase in the water level when the number of pumps operated by the virtual pump well model of FIG. 4 is one. FIG. 5B is a diagram showing the relationship between the water level of the river and the amount of increase in the water level when the number of pumps operated by the virtual pump well model of FIG. 4 is two. FIG. 5C is a diagram showing the relationship between the water level of the river and the amount of increase in the water level when the number of pumps operated by the virtual pump well model of FIG. 4 is three. It is a block diagram which shows the structure of the water level prediction part. It is a flowchart which shows the operation of the water level prediction part of FIG. FIG. 8A is a block diagram showing a hardware configuration that realizes the function of the water level prediction device according to the first embodiment. FIG. 8B is a block diagram showing a hardware configuration for executing software that realizes the function of the water level prediction device according to the first embodiment.

Embodiment 1.
FIG. 1 is a block diagram showing a configuration of a water level prediction device 1 according to a first embodiment. The water level prediction device 1 predicts the water level of the river by using various data related to the river whose water level is to be predicted registered in the river database 2 and a virtual pump well model. The virtual pump well model is a model for reflecting the phenomenon that the water level of the river suddenly rises due to the operation of the pump in the water level prediction of the river, and is a storage tank that stores the rainwater that has fallen in the river basin. , It is assumed that the pump operates according to the storage height and drains water to the river.

Various data related to the river registered in the river database 2 include, for example, past rainfall data in the river basin, pump operation history data, current rainfall data in the river basin, and observed water level of the river. The operation history data of the pump includes, for example, the presence / absence of pump operation in the virtual pump well model and the outflow amount of stored water by the pump operated in the past.

FIG. 2A is a diagram showing an outline of the virtual pump well model 20 in which the storage height S is equal to or less than the threshold value Th. FIG. 2B is a diagram showing an outline of the virtual pump well model 20 in which the storage height S exceeds the threshold value Th. In the pump well assumed by the virtual pump well model 20 shown in FIGS. 2A and 2B, rainwater having an effective rainfall R that has fallen in the river basin is stored. The effective rainfall R is the amount of rainfall that directly flows into the river among the rainfall that has fallen in the river basin.

As shown in FIG. 2A, if the storage height S of the pump well of the virtual pump well model 20 is equal to or less than the threshold Th, the pump does not operate and the water stored in the pump well is not discharged to the river by the pump. On the other hand, as shown in FIG. 2B, when the storage height S of the pump well exceeds the threshold value Th, the pump operates and water having an outflow amount q flows out from the pump well to the river.

As shown in FIG. 1, the water level prediction device 1 includes a first learning unit 10, a second learning unit 11, and a water level prediction unit 12. The first learning unit 10 learns pump operation parameters related to pump operation by using the virtual pump well model 20, the past rainfall data of the basin read from the river database 2, and the operation history data of the pump.

For example, the first learning unit 10 identifies the storage height S of the pump well of the virtual pump well model 20 from the past rainfall data of the basin, and uses the outflow amount q included in the history data of the pump operation to pump the pump. The threshold value Th of the storage height S in which is operated is obtained. The threshold value Th is a value indicating the timing at which the pump operates in a situation where it rains in the basin and the storage height S of the pump well increases.

The second learning unit 11 learns the relationship between the water level of the river and the amount of water level increase using the threshold value Th learned by the first learning unit 10. The amount of increase in water level is the amount of increase in the water level of the river due to the water flowing from the pump well to the river by the pump. The second learning unit 11 identifies the timing at which the pump operates by the threshold Th, and learns the relationship between the water level of the river and the amount of increase in the water level, which indicates a situation in which the water level of the river increases from the timing at which the pump operates. To.

The water level prediction unit 12 predicts the water level of the river based on the threshold Th learned by the first learning unit 10 and the relationship between the water level of the river learned by the second learning unit 11 and the amount of increase in the water level. For example, the water level prediction unit 12 sets a threshold value Th and a value of a parameter indicating the relationship between the river water level and the amount of water level increase in the water level prediction model, and predicts the river water level according to this water level prediction model.

Next, the operation will be described.
FIG. 3 is a flowchart showing the water level prediction method according to the first embodiment, and shows a series of processes by the water level prediction device 1 until the water level of the river is predicted.
The first learning unit 10 learns the pump operation parameters including the threshold Th of the storage height at which the pump operates in the virtual pump well model 20 (step ST1).

For example, in the first learning unit 10, the rain amount data and the operation history data of the basin when it rains in the river basin and the pump operates, and the pump does not operate even if it rains in the river basin. A virtual pump well model 20 is constructed by using the rainfall data of the basin and the operation history data of the pump at that time. The first learning unit 10 predicts the outflow amount of the stored water by the pump operated by using the virtual pump well model 20, obtains the stored amount of the pump well in which the stored water flows out with the predicted outflow amount, and obtains the stored amount. Among the highs, the storage amount with the highest evaluation is learned as the threshold Th.

The pump malfunction probability may be used as an evaluation value for the evaluation of the storage amount. The pump malfunction probability may be a value obtained by dividing the number of times the pump operation prediction is erroneous by the total number of predictions using the virtual pump well model 20. The first learning unit 10 uses the virtual pump well model 20 to learn the threshold value Th so that the probability of malfunction of the pump is minimized. As the evaluation value of the stored amount, the difference between the outflow amount of the pump predicted by the virtual pump well model 20 and the outflow amount of the stored water by the pump operated in the past may be used. The first learning unit 10 uses the virtual pump well model 20 to learn the threshold value Th so that the difference is minimized.

FIG. 4 is a diagram showing an outline of the virtual pump well model 20 according to the first embodiment. As shown in FIG. 4, in the virtual pump well model 20, thresholds Th1, Th2, Th3 are set for the storage height S of the pump wells, and the number of pumps corresponding to each of the thresholds Th1, Th2, Th3 is set. Operate. For example, when the storage height S exceeds the threshold value Th1, one pump operates to discharge the stored water to the river with the outflow amount T1. When the storage height S exceeds the threshold Th2, the two pumps operate to discharge the stored water to the river with the outflow amount T2. When the storage height S exceeds the threshold value Th3, three pumps operate to discharge the stored water to the river with the outflow amount T3.

The following equation (1) is a relational expression showing the relationship between the storage height S of the virtual pump well model 20 and the outflow amount Q, and the following formula (2) shows the time change of the storage height S of the virtual pump well model 20. It is a relational expression shown. In the following formula (1) and the following formula (2), S is the storage height of the pump well. Q is the amount of stored water that continuously flows out from the pump well to the river regardless of the pump. R is the effective rainfall, and K and P are parameters indicating the relationship between the storage amount S and the runoff amount Q. T1, T2 and T3 are the outflow amounts of the stored water to the river for each number of operating pumps.
S = KQ ^P ... (1)
dS / dt = RQ-T1-T2-T3 ... (2)

In the above formula (2), when the storage height S is equal to or less than the threshold value Th1, since there are no operating pumps, T1, T2, and T3 are all "0". If the storage height S exceeds the threshold Th1 and is equal to or less than the threshold Th2, one pump operates, so T2 and T3 become “0”. If the storage height S exceeds the threshold Th2 and is equal to or less than the threshold Th3, T3 becomes “0” because there are two operating pumps.

The first learning unit 10 uses the rainfall data of the past basin and the outflow amount of the stored water by the pump operated in the past, and according to the above equations (1) and (2), for example, the effective rainfall R and the outflow amount Q. , The outflow amounts T1, T2 and T3 and the parameters K and P are learned. The pump operation parameters including the threshold values Th and the parameters K and P corresponding to the outflow amounts T1, T2, and T3 learned by the first learning unit 10 are transmitted to the second learning unit 11 and the water level prediction unit 12. It is output.

The second learning unit 11 learns the water level change parameter including the relationship between the water level of the river and the amount of increase in the water level according to the operation of the pump (step ST2). Utilizing the fact that the amount of stored water discharged by the pumps to the river takes a discrete value for each number of operating pumps, the second learning unit 11 sets the water level of the river for each number of operating pumps. Regression learning of the relationship with the amount of water level increase.

FIG. 5A is a diagram showing the relationship between the water level of the river and the amount of increase in the water level when the number of pumps operated by the virtual pump well model 20 is one. FIG. 5B is a diagram showing the relationship between the water level of the river and the amount of increase in the water level when the number of pumps operated by the virtual pump well model 20 is two. FIG. 5C is a diagram showing the relationship between the water level of the river and the amount of increase in the water level when the number of pumps operated by the virtual pump well model 20 is three.

The second learning unit 11 retraces and learns the relationship between the observed water level of the river read from the river database 2 and the amount of increase in the water level of the river according to the outflow of stored water by the pumps for each number of operating pumps. .. In regression learning, as shown in FIGS. 5A, 5B and 5C, linear regression (eg, least squares method) is used to identify the relationship between the water level and the amount of water level increase, and the slope and intercept in the specified relationship are determined. Learn as a water level change parameter. In this way, the water level change parameter when one pump operates, the water level change parameter when two pumps operate, and the water level change parameter when three pumps operate are learned.

In the explanation so far, the case where three threshold values are set for the storage amount of the pump well of the virtual pump well model 20 and one to three pumps operate is shown, but the present invention is not limited to this. .. In the virtual pump well model 20, since it is sufficient that the number of pumps corresponding to each of the plurality of thresholds for the storage height of the pump well operates, the number of thresholds is two or more, and the pumps that operate corresponding to the thresholds. The number of units may be two or more.

The water level prediction unit 12 predicts the water level of the river based on the pump operation parameters learned by the first learning unit 10 and the water level change parameters learned by the second learning unit 11 (step ST3). For example, in a situation where the storage height S of the pump well is rising due to rainfall in the river basin, if the pump is not operating, the water level prediction unit 12 will use the current rainfall data to determine the river when one pump is operating. Predict the water level. When one pump operates, the water level prediction unit 12 predicts the water level of the river when two pumps operate from the current rainfall data. Further, when two pumps are operated, the water level prediction unit 12 predicts the water level of the river when three pumps are operated from the current rainfall data. The water level predicted by the water level prediction unit 12 is displayed on a display (not shown).

For example, the water level prediction unit 12 predicts the outflow amount and the storage amount from the pump well and the operation timing of the pump from the rainfall data by calculating the numerical difference using the above equations (1) and (2). To do. Here, the variable U indicating the presence or absence of pump operation is expressed as U = g (R | w). The variable U is a variable representing the presence or absence of pump operation by 0 or 1. g is a function that predicts the variable U from the learning parameter w and the rainfall R. The learning parameter w is, for example, w = {K, P, Thn, Tn} (n is the number of operating pumps).

The first learning unit 10 calculates the learning parameter w so that the value of the evaluation function shown in the following equation (3) is minimized. In formula (3), R _i is the actual value of the actual rainfall, U _i is the observed value of the presence or absence of the pump operation (operation history data of the pump). i is a data number and N is the number of training data. The following equation (3) represents the difference between the predicted value of the pump operation by the function g and the observed value of the pump operation, and the learning parameter w is learned so that this difference becomes small. Here, the case where the learning parameter w is learned by focusing on the prediction error of the pump operation is shown, but the steepest descent method is focused on the error between the predicted outflow amount and the observed outflow amount by focusing on the outflow amount of the pump well. The learning parameter w may be learned by such means.

When the water level hr of the pump well can be observed, the first learning unit 10 learns the threshold value at which the pump operates using the observation data of the water level of the pump well, and the water level prediction unit 12 of the pump well Pump operation may be predicted using water level observation data. Let h be the function that predicts the operation of the pump, and let w2 = {ai, b, th} as the learning parameter w2. ai and b are parameters indicating the relationship between the rainfall ri and the water level hr, and th is the threshold value of the water level of the pump well in which the pump operates. Let the rainfall R be R = {r1, r2, r3, ..., Rk}. k is a time number, and ri is the amount of rainfall at each time. The following equation (4) is a model of the relationship between the rainfall, the water level of the pump well, and the pump operation, and the water level prediction unit 12 predicts the operation of the pump using the following equation (4).

In the first learning unit 10, the difference between the predicted value of the water level at which the pump operates and the observed value of the water level at which the pump operates is minimized by the function h, similarly to the learning of the function g of the above equation (3). As described above, the learning parameter w2 is learned. The water level prediction unit 12 predicts the operation of the pump from the water level hr and the rainfall R of the pump well according to the above equation (4) using the learning parameter w2 learned by the first learning unit 10.

FIG. 6 is a block diagram showing the configuration of the water level prediction unit 12. FIG. 7 is a flowchart showing the operation of the water level prediction unit 12, and shows the details of the process of step ST3 of FIG. As shown in FIG. 6, the water level prediction unit 12 includes a setting unit 120, a prediction unit 121, a likelihood evaluation unit 122, and a correction unit 123.

The setting unit 120 sets the pump operation parameter learned by the first learning unit 10 and the water level change parameter learned by the second learning unit 11 in the water level prediction model as initial values (step ST1a). The water level prediction model is a prediction model for predicting the water level of a river by using a pump operation parameter and a water level change parameter. For example, the water level prediction model predicts the pump operation using the function g or h learned by the first learning unit 10, and predicts the pump operation and the water level learned by the second learning unit 11. The water level is predicted from the result of regression prediction of the amount of increase in water level for each.

The prediction unit 121 predicts the water level of the river by using the water level prediction model in which the initial value is set (step ST2a). For example, in the water level prediction model, the setting unit 120 sets an initial value for each number of pumps, so that the prediction unit 121 predicts the water level of the river for each initial value.

The likelihood evaluation unit 122 calculates the likelihood of predicting the water level using the water level prediction model (step ST3a). For example, the likelihood evaluation unit 122 calculates the likelihood of water level prediction for each initial value using the predicted value of the water level for each initial value and the observed water level of the river read from the river database 2, and the calculated likelihood. Evaluate the necessity of correcting the initial value according to. As the likelihood, the difference between the predicted water level and the observed water level may be used. For example, if this difference is greater than a certain threshold, the initial value needs to be modified.

The correction unit 123 corrects the initial value evaluated as needing correction by the likelihood evaluation unit 122 (step ST4a). For example, the correction unit 123 may correct the initial values (pump operation parameter and water level change parameter) that need to be corrected by using a state estimation model in which the state vector is set. Further, the correction unit 123 may correct the initial value by using the resampling process in the particle filter which is one of the state estimation models. In this case, the particle filter is set with an initial value that needs to be corrected as a particle. For example, among a plurality of initial values, by reducing the particles with the initial value having a small likelihood and duplicating the particles with the initial value having a high likelihood, the number of particles with a small prediction error increases and the initial value Is fixed. In addition to the particle filter, the state estimation model includes a Kalman filter or an extended Kalman filter. The prediction unit 121 makes a plurality of predictions having different initial values using the state estimation model, and the correction unit 123 estimates the initial value having the maximum likelihood. The water level is predicted using the initial values estimated in this way.

Subsequently, the setting unit 120 sets the initial value corrected by the correction unit 123 in the water level prediction model (step ST5a). After that, the prediction unit 121 predicts the water level of the river using the water level prediction model in which the corrected initial value is set (step ST6a). In this way, the likelihood is calculated using the predicted water level and the observed water level, and the necessity of correcting the initial value is evaluated according to the calculated likelihood. Therefore, the river that matches the observed water level obtained in real time is evaluated. It is possible to predict the water level.

Although FIG. 6 shows a case where the initial value is corrected once, the initial value may be repeatedly corrected until the likelihood satisfies the allowable condition. For example, when the difference between the predicted water level and the observed water level is the likelihood, the correction of the initial value is repeated until the likelihood corresponding to each of the initial values becomes equal to or less than a certain threshold value. This makes it possible to predict the water level of a river that matches the observed water level obtained in real time.

Next, the hardware configuration of the water level prediction device 1 will be described.
The functions of the first learning unit 10, the second learning unit 11, and the water level prediction unit 12 in the water level prediction device 1 are realized by the processing circuit. That is, the water level prediction device 1 includes a processing circuit for executing the processes from step ST1 to step ST3 in the flowchart shown in FIG. The processing circuit may be dedicated hardware, or may be a CPU (Central Processing Unit) that executes a program stored in the memory.

FIG. 8A is a block diagram showing a hardware configuration that realizes the function of the water level prediction device 1 according to the first embodiment. Further, FIG. 8B is a block diagram showing a hardware configuration for executing software that realizes the function of the water level prediction device 1 according to the first embodiment. In FIGS. 8A and 8B, the database interface 100 is an interface that relays the exchange of data between the water level prediction device 1 and the river database 2. The past rainfall data of the river basin and the operation history data of the pump are output to the water level prediction device 1 via the database interface 100.

The display interface 101 is an interface that relays the transmission of the predicted water level information from the water level predicting device 1 to a display (not shown). The predicted water level of the river predicted by the water level prediction unit 12 is displayed on the display via the display interface 101.

When the processing circuit is the processing circuit 102 of the dedicated hardware shown in FIG. 8A, the processing circuit 102 may be, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, or an ASIC (Application Special Integrated Circuit). ), FPGA (Field-Programmable Gate Array), or a combination thereof.
The functions of the first learning unit 10, the second learning unit 11, and the water level prediction unit 12 in the water level prediction device 1 may be realized by separate processing circuits, and these functions are collectively realized by one processing circuit. You may.

When the processing circuit is the processor 103 shown in FIG. 8B, the functions of the first learning unit 10, the second learning unit 11, and the water level prediction unit 12 in the water level prediction device 1 are software, firmware, or a combination of software and firmware. Realized by. The software or firmware is described as a program and stored in the memory 104.

The processor 103 realizes the functions of the first learning unit 10, the second learning unit 11, and the water level prediction unit 12 in the water level prediction device 1 by reading and executing the program stored in the memory 104. That is, the water level prediction device 1 includes a memory 104 for storing a program in which the processes from steps ST1 to ST3 in the flowchart shown in FIG. 3 are executed as a result when executed by the processor 103. These programs cause the computer to execute the procedure or method of the first learning unit 10, the second learning unit 11, and the water level prediction unit 12 in the water level prediction device 1. The memory 104 may be a computer-readable storage medium in which a program for causing the computer to function as the first learning unit 10, the second learning unit 11, and the water level prediction unit 12 in the water level prediction device 1 is stored.

The memory 104 is, for example, a non-volatile semiconductor such as a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable Read Only Memory), an EEPROM (Electrolytically Magnetic Disc), or an EPROM (Electrically-EPROM). , Flexible discs, optical discs, compact discs, mini discs, DVDs, etc.

The functions of the first learning unit 10, the second learning unit 11, and the water level prediction unit 12 in the water level prediction device 1 may be partially realized by dedicated hardware and partly realized by software or firmware. For example, the first learning unit 10 and the second learning unit 11 realize the functions by the processing circuit 102, which is dedicated hardware, and the water level prediction unit 12 reads the program stored in the memory 104 by the processor 103. The function is realized by executing. In this way, the processing circuit can realize the above-mentioned functions by hardware, software, firmware or a combination thereof.

As described above, in the water level prediction device 1 according to the first embodiment, the pump operation parameters related to the pump operation are learned by using the virtual pump well model 20, and the river water level and the water level increase amount according to the pump operation. The water level change parameters that indicate the relationship with are learned, and the water level of the river is predicted based on these parameters. As a result, the change in the river water level according to the operation of the pump can be accurately modeled, so that the water level of the river in which the stored water is inflowed by the pump can be accurately predicted.

In the water level prediction device 1 according to the first embodiment, in the virtual pump well model 20, a number of pumps corresponding to each of a plurality of threshold values for the storage height of the pump wells operate. The first learning unit 10 learns pump operation parameters including the number of operating pumps. The second learning unit 11 learns the water level change parameter for each operating number of pumps included in the pump operating parameter. The water level prediction unit 12 predicts the water level of the river by using the water level prediction model in which the pump operation parameter and the water level change parameter for each operating number of pumps are set as initial values, and observes the predicted value of the river water level and the water level of the river. The likelihood of prediction using the water level prediction model is calculated based on the value, and the initial value is corrected according to the calculated likelihood. In this way, the likelihood is calculated using the predicted water level and the observed water level, and the necessity of correcting the initial value is evaluated according to the calculated likelihood. Therefore, the river that matches the observed water level obtained in real time is evaluated. Water level can be predicted.

In the water level prediction device 1 according to the first embodiment, the first learning unit 10 learns the threshold value Th that minimizes the malfunction probability of the pump. This makes it possible to learn the threshold value Th at which the pump operates in the virtual pump well model 20.

In the water level prediction device 1 according to the first embodiment, the water level prediction unit 12 corrects the initial value by using the state estimation model in which the pump operation parameter and the water level change parameter are set as the state vector. For example, the water level prediction unit 12 corrects the initial values by using a particle filter in which the pump operation parameter and the water level change parameter are set as particles. This makes it possible to modify the initial value set in the water level prediction model.

The present invention is not limited to the above-described embodiment, and within the scope of the present invention, it is possible to modify any component of the embodiment or omit any component of the embodiment.

Since the water level predicting device according to the present invention can accurately predict the water level of a river in which water is discharged by a pump, it can be used, for example, as a monitoring device for monitoring the water level of a river.

1 Water level predictor, 2 River database, 10 1st learning unit, 11 2nd learning unit, 12 Water level prediction unit, 20 Virtual pump well model, 100 Database interface, 101 Display interface, 102 processing circuit, 103 processor, 104 Memory, 120 setting unit, 121 prediction unit, 122 likelihood evaluation unit, 123 correction unit.

Claims (8)

A virtual pump well model that assumes a virtual pump well in which rainwater that has fallen in a river basin is stored and the pump operates according to the storage height to discharge water to the river, past rainfall data in the basin, and the virtual A first learning unit that learns pump operation parameters related to the operation of the pump using the operation history data of the pump in the pump well model, and
A second learning unit that learns a water level change parameter indicating the relationship between the water level of the river and the amount of increase in the water level according to the operation of the pump by using the pump operation parameter.
A water level prediction unit that predicts the water level of the river based on the pump operation parameter and the water level change parameter.
A water level predictor characterized by being equipped with. In the virtual pump well model, the number of the pumps corresponding to each of the plurality of thresholds for the storage height of the pump well operates.
The first learning unit learns the pump operation parameters including the number of operating pumps, and receives the pump operation parameters.
The second learning unit learns the water level change parameter for each operating number of the pump included in the pump operating parameter.
The water level prediction unit predicts the water level of the river by using the water level prediction model in which the pump operation parameter and the water level change parameter for each operating number of the pump are set as initial values, and the predicted value of the water level of the river. The first aspect of claim 1, wherein the likelihood of prediction using the water level prediction model is calculated based on the observed value of the water level of the river, and the initial value is corrected according to the calculated likelihood. Water level predictor. The water level prediction device according to claim 1 or 2, wherein the first learning unit learns a threshold value at which the probability of malfunction of the pump is minimized. The water level prediction device according to claim 2, wherein the water level prediction unit corrects the initial values by using a state estimation model in which the pump operation parameters and the water level change parameters are set as state vectors. The water level prediction device according to claim 2, wherein the water level prediction unit corrects the initial values by using a particle filter in which the pump operation parameter and the water level change parameter are set as particles. The water level prediction device according to claim 1, wherein the first learning unit learns a threshold value at which the pump operates using observation data of the water level of the pump well. The water level prediction device according to claim 1, wherein the water level prediction unit predicts the operation of the pump using the observation data of the water level of the pump well. The first learning unit is a virtual pump well model assuming a virtual pump well in which rainwater that has fallen in a river basin is stored and a pump operates according to the storage height to cause water to flow out to the river. A step of learning pump operation parameters related to the operation of the pump by using the past rainfall data and the operation history data of the pump in the virtual pump well model, and
The second learning unit uses the pump operation parameter to learn a water level change parameter indicating the relationship between the water level of the river and the water level increase amount according to the operation of the pump.
A step in which the water level prediction unit predicts the water level of the river based on the pump operation parameter and the water level change parameter.
A water level prediction method characterized by being equipped with.

Images