JP2020111892A - Flow rate prediction device and flow rate prediction method - Google Patents

Abstract

To provide a flow rate prediction device and a flow rate prediction method allowing prediction of a flow rate with high accuracy when the flow rate of a river is huge, and facilitating them to be introduced.SOLUTION: A flow rate prediction device 1 is provided to predict a flow rate of a specific river, the flow rate prediction device 1 being provided with: a prediction model learning section 20 generating a plurality of prediction models 31 through ensemble learning of a plurality of low models 21 in order to predict a flow rate after a previous point of time which is any of previous points of time using weather observation data 2, weather prediction data 3, and flow rate data 4 as learning data with relation to the specific river at the previous point of time; and a flow rate prediction section 30 predicting a future flow rate of the specific rive based on a plurality of temporary future prediction flow rates through output of the temporary future prediction flow rates by inputting the weather observation data 2, the weather prediction data 3, and the flow rate data 4 with relation to the specific river in each of the plurality of prediction models 31 as it is.SELECTED DRAWING: Figure 1

Description

The present invention relates to a flow rate prediction device and a flow rate prediction method.

From the past, it has been widely performed to predict the flow rate of a river based on weather information. Since it is highly difficult to predict the flow rate of a river, it has been proposed to automate it and predict the flow rate of the river easily.

In the automation of river flow rate prediction, a prediction model using a neural network may be used as disclosed in Patent Document 1, for example.
The flow rate predicting device disclosed in Patent Document 1 is a weighting ratio determining unit that determines a weighting ratio that represents the degree of influence of rainfall data at a certain location on the downstream flow rate for each of the rainfall data at a plurality of locations, and a weighting ratio. And the corresponding rainfall data, the model construction means for constructing the discharge forecast model using the catchment average rainfall data and the catchment average rainfall data by the determined weighting ratio are input to the constructed forecast model. And a predicting unit that generates predicted flow rate data related to the predicted flow rate.
The load ratio increases the weighting ratio of the rainfall data at the points where the increase or decrease of rainfall has a large effect on the increase or decrease of the downstream flow rate. It is determined by reducing the weighting ratio. More specifically, the load ratio and the flow rate prediction model are evaluated based on the correlation coefficient, which is the degree of correlation between the rainfall amount and the flow rate, and the prediction error. The flow rate prediction model is learned by repeating learning and prediction.

Learning data such as meteorological observation data and meteorological prediction data is required in order to construct a prediction model using a neural network as described above. However, the amount of rainfall, which has a large impact on it, such as typhoons and torrential rains, which requires particularly high prediction accuracy, is much smaller than the data with small amounts of rainfall and rainfall. .. Therefore, the prediction accuracy may not be improved when the flow rate and the rainfall are large.

On the other hand, Patent Document 2 discloses predicting the flow rate of a river or the like at a predetermined future time by a flow rate prediction device using a prediction model based on the following neural network. The prediction model is constructed on the basis of learning data for constructing a prediction model, which is obtained by processing forecast rainfall data, actual rainfall data, actual discharge data, or the like. In the prediction model construction data, data in which the average rainfall amount, the discharge amount or the change amount of the discharge amount is equal to or less than a certain value is deleted, and the data in which the average rainfall amount, the discharge amount or the change amount of the discharge amount is equal to or greater than a certain value are multiplied. As a result, it is possible to deal with the deviation of the data regarding the rainfall amount or the flow rate.
In the flow rate predicting device of Patent Document 2, the average rainfall is a point average value obtained by adding together the rainfalls at a plurality of points and dividing by the number of points.

JP, 2007-205001, A JP, 2007-226450, A

In the flow rate predicting device of Patent Document 1, as described above, the prediction accuracy when the flow rate is large may not be desirable.
Further, since learning and prediction are repeated, it takes a lot of time to learn the prediction model. Therefore, the introduction of the flow rate predicting device is not easy.

In the flow rate predicting device of Patent Document 2, the data in which the average rainfall, the flow rate, or the change in the flow rate is a certain value or more is simply duplicated, that is, duplicated. That is, since the learning data with new information is not added, the information amount of the learning data as a whole remains unchanged.
In addition, since the average rainfall is the average value of points, geographical information is not reflected with high accuracy.
Furthermore, by comparing the flow rate or flow rate change in the data with a fixed value, which data is to be deleted or to be multiplied is determined, but human arbitraryness intervenes in the determination of this fixed value. As a result, an appropriate value may not be set.
From the above, there is still a possibility that the prediction accuracy when the flow rate is large is not desirable.

The problem to be solved by the present invention is to provide a flow rate predicting apparatus and a flow rate predicting method capable of performing highly accurate flow rate prediction even when the flow rate of a river is large and being easy to introduce.

The present invention adopts the following means in order to solve the above problems. That is, the present invention is a flow rate predicting device that predicts the flow rate of a specific river, and at a past time that is an arbitrary time in the past, the meteorological observation data, the weather forecast data, and the flow rate data related to the river. As learning data, by ensemble learning a plurality of weak learners so as to predict the flow rate after the past time, a prediction model learning unit that generates a plurality of prediction models, and each of the plurality of prediction models, the present, , The weather observation data, the weather forecast data, and the flow rate data related to the river are input to output a provisional future forecast flow rate, and based on a plurality of the provisional future forecast flow rates. There is provided a flow rate predicting device, comprising: a flow rate predicting unit for predicting a future flow rate of the river.

In addition, the present invention is a flow rate prediction method for predicting the flow rate of a specific river, and includes meteorological observation data, meteorological prediction data, and flow rate data related to the river at a past time that is an arbitrary time in the past. As learning data, by ensemble learning a plurality of weak learners so as to predict the flow rate after the past time, to generate a plurality of prediction models, in each of the plurality of prediction models, the current, in the river The relevant meteorological observation data, the meteorological forecast data, and the discharge data are input to output a tentative future predicted discharge, and the future of the river based on the plurality of tentative future predicted discharges. A flow rate prediction method for predicting a flow rate is provided.

Advantageous Effects of Invention According to the present invention, it is possible to provide a flow rate prediction device and a flow rate prediction method that can accurately predict a flow rate even when the flow rate of a river is large and that can be easily introduced.

It is a block diagram of the flow prediction device in the embodiment of the present invention. It is explanatory drawing regarding the input/output of the prediction model learning part of the flow volume prediction apparatus in the said embodiment. It is explanatory drawing regarding the input/output of the flow volume estimation part of the flow volume estimation apparatus in the said embodiment. In the flow rate predicting method in the above embodiment, (a) is a flowchart for learning the flow rate, and (b) is a flowchart for predicting the flow rate. It is a block diagram of the flow prediction device in the 1st modification of the above-mentioned embodiment. It is explanatory drawing of the seasonal term data production|generation part of the flow volume estimation apparatus in the said 1st modification. It is a block diagram of the flow prediction device in the 2nd modification of the above-mentioned embodiment. It is explanatory drawing of the machine learning device and the feature extraction model in the said 2nd modification.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
The flow rate predicting apparatus in the present embodiment is a flow rate predicting apparatus that predicts the flow rate of a specific river, and at a past time that is an arbitrary time in the past, the weather observation data, the weather forecast data, and the flow rate related to the river. Using the data as learning data, a prediction model learning unit that generates a plurality of prediction models by ensemble learning multiple weak learners so as to predict the flow rate after the past time, and each of the plurality of prediction models. Of the river, the weather forecast data, the weather forecast data, and the discharge data related to the river are input to output the provisional future forecast discharge, and the future discharge of the river is based on the multiple provisional future forecast discharges. And a flow rate predicting unit that predicts.

FIG. 1 is a block diagram of a flow rate predicting apparatus according to this embodiment. FIG. 2 is an explanatory diagram regarding the input/output of the prediction model learning unit of the flow rate prediction device in the present embodiment. FIG. 3 is an explanatory diagram regarding the input/output of the flow rate predicting unit of the flow rate predicting apparatus according to the present embodiment.
The flow rate prediction device 1 in the present embodiment targets a specific river and predicts the future flow rate of the river.

The flow rate predicting device 1 is an information processing device such as a personal computer. The flow rate prediction device 1 includes a meteorological observation data storage unit 11, a weather forecast data storage unit 12, a flow rate storage unit 13, a data processing unit 14, a learning unit 15, a prediction model parameter storage unit 16, and a prediction unit 17. The learning unit 15 includes a prediction model learning unit 20. The prediction unit 17 includes a flow rate prediction unit 30.
Among the components of the flow rate prediction device 1, the data processing unit 14, the prediction model learning unit 20, and the flow rate prediction unit 30 may be software or programs executed by the CPU in the information processing device, for example. Further, the meteorological observation data storage unit 11, the weather forecast data storage unit 12, the flow rate storage unit 13, and the prediction model parameter storage unit 16 are realized by a storage device such as a semiconductor memory or a magnetic disk provided inside or outside the information processing apparatus. It may have been done.

The input data of the flow rate prediction device 1 includes meteorological observation data 2, weather forecast data 3, and flow rate data 4.
The meteorological observation data 2 is meteorological data actually observed at an observatory or the like. The meteorological observation data 2 includes at least rainfall. The meteorological observation data 2 may include the temperature, the amount of solar radiation, the amount of snowfall, the amount of snowfall, etc. in addition to the amount of rainfall. In the present embodiment, in particular, the rainfall amount is based on the latitude/longitude on the map of the observation point data 2o observed at the observation points provided in each area and the area included as the observation target observed by radar observation or the like. It has mesh data 2m divided into a mesh shape. The meteorological observation data 2 related to the river to be predicted by the flow rate prediction device 1 from the past to the present is stored in the meteorological observation data storage unit 11.
The weather forecast data 3 is, for example, future forecast weather data at the time of forecasting weather, which is provided by a weather forecast service. The weather forecast data 3, like the weather observation data 2, includes at least rainfall. The weather forecast data 3 may include the temperature, the amount of solar radiation, the amount of snowfall, the amount of snowfall, etc. in addition to the amount of rainfall. In the present embodiment, particularly, the rainfall amount is provided with the mesh data 3 m, like the meteorological observation data 2. The weather forecast data 3 related to the river that is the forecast target of the flow rate forecasting device 1 from the past to the present is stored in the weather forecast data storage unit 12.
The flow rate data 4 is flow rate data actually observed at an observatory or the like. The flow rate data 4 of the river that is the prediction target of the flow rate prediction device 1 from the past to the present is stored in the flow rate storage unit 13.

Based on each data stored in the weather observation data storage unit 11, the weather prediction data storage unit 12, and the flow rate storage unit 13, the flow rate prediction unit 30 predicts the future flow rate of the river. In order to perform this prediction effectively, the prediction model learning unit 20 is equipped with a machine learning device, as will be described later, and the learning data including the meteorological observation data 2, the meteorological prediction data 3, and the flow rate data 4 is machined. It is input to the learning device and machine learning is performed to generate predictive model parameters for predicting future river discharge.
That is, the flow rate predicting device 1 is roughly classified into two types of operations, namely, learning of the flow rate of the river and prediction of the flow rate. In order to simplify the explanation, first, each component of the flow rate predicting apparatus 1 at the time of learning the flow rate will be described, and then the behavior of each component at the time of predicting the flow rate will be described.

At the time of learning the flow rate, the machine learning device assumes the first past time (past time), which is an arbitrary time in the past, and sets each data based on the first past time, that is, at the time of the first past time. Of the actual weather (the weather observation data at the past time), the weather forecast situation at the first past time (the weather forecast data at the past time), and the flow rate at the first past time (the past The event that actually occurred or was predicted at the first past time, such as the flow rate data at the time), is input. In response to these inputs, the machine learning device uses the second past, which is a future time relative to the first past time, such as one hour, two hours, or several hours after the first past time. Predict the discharge in the time zone up to the time.
Since the second past time is a past time when the machine learning device is learning, the flow rate in the time zone from the first past time to the second past time is actually observed and is used as data. Exists. The actual flow rate data after the first past time is compared with the flow rate after the first past time predicted by the machine learning device, and the predicted flow rate is close to the actually observed flow rate data. The machine learner is learned so that

That is, when learning the flow rate, the meteorological observation data 2 used as an input to the machine learning device is the meteorological observation data before the first past time stored in the meteorological observation data storage unit 11 (the meteorological observation data at the past time). Data) 2A. In addition, when learning the flow rate, the weather forecast data 3 used as an input to the machine learning device is the weather forecast data stored in the weather forecast data storage unit 12 after the first past time (weather at the past time). Predicted data) 3A. Furthermore, when learning the flow rate, the flow rate data 4 used as an input to the machine learning device is the flow rate data before the first past time (flow rate data at the past time) 4A.
As the weather forecast data 3A after the first past time, it is desirable to use the latest weather forecast data 3 before the first past time and as close as possible to the first past time.
The result output by the machine learning device to which the above values are input, that is, the predicted flow rate 5A after the first past time (flow rate after the past time) is the correct value, and the river after the first past time is the correct value. Of the actual flow rate data (flow rate data at the past time) 4C.
These weather observation data 2A before the first past time, weather forecast data 3A after the first past time, flow rate data 4A before the first past time, and correct value, that is, flow rate data 4C after the first past time. Is the learning data of the machine learning device.

The data processing unit 14 includes, from the meteorological observation data storage unit 11, the meteorological prediction data storage unit 12, and the flow rate storage unit 13, the meteorological observation data 2A before the first past time, the weather prediction data 3A after the first past time, The flow rate data 4A before the first past time and the flow rate data 4C after the first past time are acquired, and each is converted into the input/output format of the machine learning device.
The data processing unit 14 sends each converted data to the learning unit 15.

The learning unit 15 receives each data from the data processing unit 14.
The learning unit 15 causes the prediction model learning unit 20 to learn the machine learning device to generate a learned prediction model.
In the present embodiment, the machine learning device is trained by ensemble learning. Ensemble learning is a learning method using a plurality of weak learners as machine learning devices, that is, a weak learner having a certain degree of correlation with truly correct regression. Ensemble learning generates a plurality of prediction models by training a plurality of weak learners under different conditions, rather than performing prediction using one prediction model that may not have sufficiently high prediction accuracy. It is based on the idea that the accuracy of prediction is improved by combining multiple prediction results with a prediction model.
That is, the prediction model learning unit 20 includes a plurality of weak learners 21, and the prediction model learning unit 20 performs ensemble learning on the plurality of weak learners 21 to generate a plurality of prediction models 31.

In this embodiment, boosting is applied as a learning method for ensemble learning. In boosting, a plurality of weak learners 21 are sequentially learned and combined one by one to improve the prediction accuracy of the weak learners 21 as a whole. More specifically, when a certain weak learner 21 is trained, the weak learner 21 that has already been learned before the predicted value (in the present embodiment, the predicted flow rate 5A after the first past time), Data having a large error from the correct value (in the present embodiment, the flow rate data 4C after the first past time) is preferentially learned so that the error becomes smaller. For example, when the error between the predicted value 5A and the correct value 4C in the first data is small and the error in the second data is large in the already learned weak learner 21, the weak learner 21 to be learned next is Prioritize and prioritize the data of 2. The data may be prioritized by, for example, weighting each data, and regarding the data considered to have eliminated the error in the already learned weak learner 21, the next weak learner 21 It may be ignored in learning.

Based on the above idea, the prediction model learning unit 20 first uses the learning data received from the data processing unit 14 to train the first weak learning device 21A. The predicted value 5A, which is the output of the first weak learning device 21A, and the correct value 4C are compared by, for example, the comparator 23 and the result is reflected in the parameters and the like provided inside the first weak learning device 21A. The first weak learner 21A is learned by a method such as. When it is determined that the learning of the first weak learning device 21A is completed, if there is data with a large error between the predicted value 5A and the correct answer value 4C when learning the first weak learning device 21A, this is prioritized. Then, the second weak learner 21B is learned.
Similarly to the first weak learner 21A, the second weak learner 21B is also learned by reflecting the result of comparison between the predicted value 5A and the correct answer value 4C in the second weak learner 21B. If it is determined that the learning of the second weak learning device 21B is completed, and if there is data with a large error between the predicted value 5A and the correct answer value 4C when learning the second weak learning device 21B, this is prioritized. Then, the third weak learner 21C is further learned. The above process is repeated until the learning of all the weak learners 21 is completed.

As a result of learning each of the plurality of weak learners 21 under different conditions by the above learning, when the same input is given to each of the plurality of weak learners 21, the plurality of weak learners 21 Make different predictions. In the present embodiment, the flow rate prediction device 1 predicts the flow rate 6 by inputting the meteorological observation data 2, the weather forecast data 3, and the flow rate data 4, so the difference in this condition is, for example, the difference in the flow rate. That is, it includes a case where the flow rate is large and a case where the flow rate is small. For example, a certain weak learning device 21 performs learning that emphasizes the case where the flow rate is small, and another weak learning device 21 performs learning that emphasizes the case where the flow amount is large.
Here, when a particularly high prediction accuracy is required, for example, when the flow rate is expected to be large during an abnormal situation such as a typhoon or heavy rainfall, the data corresponding to this is small It is much less than the data. Therefore, it is basically not easy to perform learning so that the prediction accuracy becomes high when the flow rate is large. On the other hand, particularly in the present embodiment, since boosting is applied as a learning method of ensemble learning, for example, when the flow rate is large in the first weak learning device 21A, the error between the predicted value 5A and the correct value 4C is large. Becomes larger, another weak learner 21 to be subsequently learned can perform learning with an emphasis on the case where the flow rate is large.

The prediction model learning unit 20 trains the plurality of weak learners 21 as described above. When the learning ends, the prediction model learning unit 20 stores the parameter adjusted by the learning of each weak learner 21 in the prediction model parameter storage unit 16 as the prediction model parameter. The prediction model parameters stored in the prediction model parameter storage unit 16 are acquired by the prediction unit 17 described later, and a plurality of prediction models 31 that predict the flow rate are realized corresponding to each weak learning device 21.
That is, the prediction model learning unit 20 is used as a program module that is a part of artificial intelligence software to generate a learned prediction model 31 in which appropriate learning parameters have been learned.

Next, the behavior of each component when predicting the flow rate will be described.

At the time of predicting the flow rate, in order to predict the future flow rate of the river, the meteorological observation data 2, the weather forecast data 3, and the flow rate data 4 are input to the prediction model 31 for which the learning has been completed. .. That is, the input of the prediction model 31 is before the present, that is, the current and past meteorological observation data (current meteorological observation data) 2B, and after the present, that is, the future meteorological forecast data (current meteorological forecast data) 3B. , And the current flow data before (current flow data) 4B.
The data processing unit 14 stores the meteorological observation data 2B before and after the present, the meteorological forecast data 3B at the future, and the before-current flow rate data 4B from the meteorological observation data storage unit 11, the weather forecast data storage unit 12, and the flow rate storage unit 13. It is acquired and each is converted into the input format of the flow rate prediction unit 30.
The data processing unit 14 transmits the converted data to the prediction unit 17.

The prediction unit 17 receives each data from the data processing unit 14.
The prediction unit 17 predicts the future flow rate 6 of the river by the flow rate prediction unit 30 in which a plurality of prediction models 31 are built.
Each of the plurality of prediction models 31 is constructed corresponding to each of the plurality of weak learners 21 learned by the prediction model learning unit 20. For example, each of the first prediction model 31A, the first prediction model 31B, and the first prediction model 31C corresponds to the first weak learning device 21A, the second weak learning device 21B, and the third weak learning device 21C, respectively. ing. The flow rate prediction unit 30 acquires the prediction model parameters corresponding to each weak learning device 21 stored in the prediction model parameter storage unit 16, and builds a prediction model 31 corresponding to each weak learning device 21 based on each of these. To do.
The flow rate prediction unit 30 inputs each data received from the data processing unit 14 into each of the plurality of prediction models 31.
As described above, the plurality of weak learners 21 are learned under different conditions. That is, the prediction models 31 corresponding to each of the plurality of weak learners 21 output different values even if the same data is input to each.

The flow rate predicting unit 30 includes a coupler 32. The flow rate prediction unit 30 inputs the provisional future predicted flow rate 5B, which is the output of the plurality of prediction models 31, to the coupler 32. The coupler 32 calculates and predicts the future flow rate 6 based on the plurality of provisional future predicted flow rates 5B. The combiner 32 predicts the future flow rate 6 by calculating, for example, the sum, average, or weighted sum of a plurality of provisional future predicted flow rates 5B. When the combiner 32 calculates the weighted sum, the weight is determined depending on the performance of each weak learner 21 such that the weight of the weak learner 21 having a low error rate, that is, good regression accuracy, becomes large. To be done.

Next, a flow rate prediction method by the flow rate prediction device 1 will be described with reference to FIGS. 1 to 3 and 4. FIG. 4A is a flowchart for learning the flow rate, and FIG. 4B is a flowchart for predicting the flow rate.
This flow rate prediction method is a flow rate prediction method that predicts the flow rate of a specific river, and learns meteorological observation data, meteorological prediction data, and flow rate data related to the river at a past time that is an arbitrary time in the past. As data, ensemble learning is performed on multiple weak learners to predict the flow rate after the past time, and multiple prediction models are generated. The observational data, the meteorological prediction data, and the discharge data are input to output the tentative future predicted discharge, and the future discharge of the river is predicted based on the multiple tentative predicted future discharges.

First, the operation of each component of the flow rate predicting device 1 when learning the flow rate will be described.
The data processing unit 14 includes, from the meteorological observation data storage unit 11, the meteorological prediction data storage unit 12, and the flow rate storage unit 13, the meteorological observation data 2A before the first past time, the weather prediction data 3A after the first past time, The flow rate data 4A before the first past time and the flow rate data 4C after the first past time are acquired, and each is converted into the input/output format of the weak learning device 21.
The data processing unit 14 sends each converted data to the learning unit 15.
The learning unit 15 receives each data from the data processing unit 14.
The learning unit 15 causes the prediction model learning unit 20 to learn the weak learner 21 to generate the learned prediction model 31 (step S1). In the present embodiment, the prediction model learning unit 20 uses the weather observation data 2A before the first past time, the weather prediction data 3A after the first past time, the flow rate data 4A before the first past time, and the correct value. The plurality of weak learners 21 are ensemble-learned so that the flow rate data 4C after the first past time is used as learning data and the predicted flow rate 5A after the first past time is predicted.
The prediction model learning unit 20 trains the plurality of weak learners 21 as described above. In this embodiment, since boosting is applied as a learning method for ensemble learning, a plurality of weak learnings are performed until it is determined by the determination processing shown as step S3 that all the weak learners 21 have finished learning. The devices 21 are sequentially learned one by one. When the learning ends, the prediction model learning unit 20 stores the parameter adjusted by the learning of each weak learner 21 as the prediction model parameter in the prediction model parameter storage unit 16 (step S5).

Next, the behavior of each component when predicting the flow rate will be described.
The data processing unit 14 stores the meteorological observation data 2B before and after the present, the meteorological forecast data 3B at the future, and the before-current flow rate data 4B from the meteorological observation data storage unit 11, the weather forecast data storage unit 12, and the flow rate storage unit 13. It is acquired and each is converted into the input format of the flow rate prediction unit 30.
The data processing unit 14 transmits the converted data to the prediction unit 17.
The prediction unit 17 receives each data from the data processing unit 14.
The flow rate prediction unit 30 acquires the prediction model parameters corresponding to each weak learning device 21 stored in the prediction model parameter storage unit 16, and builds a prediction model 31 corresponding to each weak learning device 21 based on each of these. Yes (step S11).
The flow rate prediction unit 30 inputs each data received from the data processing unit 14 into each of the plurality of prediction models 31. On the other hand, each of the plurality of prediction models 31 outputs the provisional future predicted flow rate 5B (step S13).
The flow rate predicting unit 30 acquires a plurality of provisional future predicted flow rates 5B output by the plurality of prediction models 31 and inputs them to the coupler 32. The coupler 32 calculates and predicts the future flow rate 6 based on the plurality of provisional future predicted flow rates 5B (step S15).

Next, effects of the flow rate predicting apparatus 1 and the flow rate predicting method described above will be described.

The flow rate prediction device according to the present embodiment is the flow rate prediction device 1 that predicts the flow rate 6 of a specific river, and the weather observation data 2A and the weather prediction data related to the river at a past time that is an arbitrary time in the past. 3A and flow rate data 4A and 4C as learning data, the prediction model learning unit 20 that ensemble-learns a plurality of weak learners 21 so as to predict a flow rate 5A after the past time and generates a plurality of prediction models 31. And, the plurality of prediction models 31 are each input with the meteorological observation data 2B, the meteorological prediction data 3B, and the flow rate data 4B related to the current river, and the provisional future predicted flow rate 5B is output. The flow rate predicting unit 30 for predicting the future flow rate 6 of the river based on the provisional future predicted flow rate 5B.
Further, the present flow rate prediction method is a flow rate prediction method for predicting the flow rate 6 of a specific river, and is related to a river at a past time that is an arbitrary time in the past, and the meteorological observation data 2A, the weather forecast data 3A, And the flow rate data 4A and 4C as learning data, the plurality of weak learners 21 are ensemble-learned so as to predict the flow rate 5A after the past time, and a plurality of prediction models 31 are generated. A plurality of provisional future forecasts are input by inputting the meteorological observation data 2B, the weather forecast data 3B, and the discharge data 4B related to the current river into the provisional future forecast discharge 5B, respectively. Predict future river flow 6 based on flow 5B.
According to the above configuration, a plurality of weak learners 21 are ensemble-learned to generate a plurality of prediction models 31, and thus the future flow rate 6 of the river is predicted. That is, the plurality of weak learners 21 are trained under different conditions including the difference in flow rate to generate a plurality of prediction models 31, and the future flow rate 6 is predicted based on the plurality of prediction results. Therefore, even in the case of a large amount of data, such as in the case of a typhoon or a torrential rain, a large number of data may not be obtained. There are 21. Therefore, it is possible to predict the flow rate with high accuracy even when the flow rate of the river is large.
Moreover, since the preparation before the introduction is completed by learning each weak learning device 21, the introduction is easy.

In particular, in the present embodiment, the data processing unit 14 only converts the input format and output format of data when learning each weak learning device 21, and does not change the data content itself such as deletion or duplication of data. Not not. Therefore, it is difficult for human arbitrariness to intervene in the learning of the weak learner 21, and the loss of information from the data is suppressed.
As a result, the effect that the accuracy of flow rate prediction is high can be more efficiently achieved.

The prediction model learning unit 20 also performs ensemble learning on each of the plurality of weak learners 21 by boosting.
For example, in the case of a typhoon or a torrential rain, it is expected that the flow rate will be large, but a large number of data may not be obtained. There is a possibility that the error between the correct answer value 4C and the correct answer value becomes large and sufficient learning cannot be performed.
Even in such a case, according to the above-described configuration, in any one of the weak learners 21 learned after the weak learner 21, sufficient learning is performed in the weak learner 21. Learning that emphasizes data that has a large flow rate is performed.
Therefore, even if the river flow is large, it is possible to predict the flow with high accuracy.

Further, the learning data includes meteorological observation data 2A before the past time related to the river, meteorological prediction data 3A after the past time related to the river, flow rate data 4A before the past time of the river, and the river as the correct value. Flow rate data 4C after the past time, and the flow rate predicting unit 30 includes, in each of the plurality of prediction models 31, the weather observation data 2B before and after the current related to the river, the future weather prediction data 3B related to the river, And the flow data 4B before and after the present of the river are input.
According to the above configuration, the flow rate predicting device 1 and the flow rate predicting method can be appropriately realized.

[First Modification of Embodiment]
Next, a first modified example of the flow rate predicting apparatus 1 and the flow rate predicting method shown as the above embodiment will be described with reference to FIGS. 5 and 6. FIG. 5 is a block diagram of a flow rate prediction device in the first modified example. FIG. 6 is an explanatory diagram of a seasonal term data generation unit of the flow rate prediction device in the first modified example. The flow rate predicting apparatus 40 and the flow rate predicting method in the first modified example are different in that the learning data further includes seasonal term data having different values depending on the season.

The flow rate prediction device 40 includes a seasonal term data generation unit 43. When the date and time is input, the seasonal term data generation unit 43 generates information regarding the season corresponding to the relevant date and time as seasonal term data. More specifically, the seasonal term data generation unit 43 includes a continuous function that outputs different values throughout the year corresponding to the date and time. When the date and time is input, the output of this continuous function is used as seasonal term data. Output.
Since December 31st, which is the end of the year, and January 1st, which is the beginning of the year, are days before and after each other, they are considered to be approximately the same seasonally. Therefore, the above continuous function is also a periodic function in which the values are repeated between the output corresponding to December 31st and the output corresponding to January 1st, that is, the value is repeated with one year as a cycle. ..
As such a function, the sin function 65 and the cos function 66 can be applied. These functions 65 and 66 are periodic functions whose values are repeated with a period of one year. Further, at least one of the values of the sin function 65 and the cos function 66 is different for different dates and times within a year. Therefore, by using these functions 65 and 66 together, when the dates and times are input. The corresponding seasonal term data is uniquely determined.
The seasonal term data generation unit 43 transmits the output obtained by applying the functions 65 and 66 to the input date and time to the data processing unit 44.

When learning the flow rate, when the date and time data 7 corresponding to the first past time is input, the seasonal term data generation unit 43 outputs the corresponding seasonal term data corresponding to the first past time to process the data. It is transmitted to the section 14.
The data processing unit 44, in addition to each data acquired from the weather observation data storage unit 11, the weather prediction data storage unit 12, and the flow rate storage unit 13, the season corresponding to the first past time received from the seasonal term data generation unit 43. The term data is converted as necessary and transmitted to the learning unit 45.
The learning unit 45 receives each data from the data processing unit 44.
The learning unit 45 causes the weak learning unit 51 to be learned by the prediction model learning unit 50, and generates the learned prediction model 61. In this modification, the prediction model learning unit 50 corresponds to meteorological observation data before the first past time, weather prediction data after the first past time, flow rate data before the first past time, and first past time. Ensemble learning is performed on the plurality of weak learners 51 so as to predict the predicted flow rate after the first past time by using the seasonal term data and the correct value, that is, the flow rate data after the first past time, as the learning data.
When the learning ends, the prediction model learning unit 50 stores the parameter adjusted by the learning of each weak learner 51 in the prediction model parameter storage unit 46 as the prediction model parameter.

At the time of predicting the flow rate, when the date and time data 7 corresponding to the present is input, the seasonal term data generation unit 43 outputs the corresponding seasonal term data corresponding to the present and transmits it to the data processing unit 14.
The data processing unit 44, in addition to each data acquired and converted from the weather observation data storage unit 11, the weather prediction data storage unit 12, and the flow rate storage unit 13, the seasonal term corresponding to the present time received from the seasonal term data generation unit 43. The data is transmitted to the prediction unit 47.
The prediction unit 47 receives each data from the data processing unit 44.
The flow rate prediction unit 60 acquires the prediction model parameters corresponding to each weak learning device 51 stored in the prediction model parameter storage unit 46, and builds a prediction model 61 corresponding to each weak learning device 51 based on each of these. To do.
The flow rate prediction unit 60 inputs each data received from the data processing unit 44 into each of the plurality of prediction models 61. On the other hand, each of the plurality of prediction models 61 outputs a provisional future predicted flow rate.
The flow rate predicting unit 60 acquires a plurality of provisional future predicted flow rates output by the plurality of prediction models 61 and inputs them to the coupler. The coupler calculates and predicts the future flow rate 6 based on the plurality of provisional future predicted flow rates.

In the flow rate predicting apparatus 40 of the present modification, the learning data further includes seasonal term data having different values depending on the season, and the flow rate predicting unit 60 causes each of the plurality of prediction models 61 to have a seasonal term corresponding to the present time. The data is further input as learning data to output the tentative future predicted flow rate, and the future flow rate 6 of the river is predicted based on the plurality of tentative future predicted flow rates.
Snowfall during the winter does not melt immediately and melts when the temperature rises in April or May. Therefore, the river flow is affected by snowmelt and tends to increase especially in April and May.
According to the above configuration, the weak learner 51 can be trained in consideration of such a seasonal factor such as snow melting, so that the future flow rate 6 is predicted in consideration of the seasonal factor. be able to.

The seasonal term data are continuous functions 65 and 66, and are output values of the periodic functions 65 and 66 whose values are repeated with a cycle of one year.
According to the above configuration, the flow rate predicting device 40 can be appropriately realized.

It goes without saying that the first modified example has other effects similar to those of the embodiment already described.

[Second Modification of Embodiment]
Next, a second modification of the flow rate predicting apparatus 1 and the flow rate predicting method shown as the above embodiment will be described with reference to FIGS. 7 and 8. FIG. 7 is a block diagram of a flow rate predicting apparatus according to the second modified example. FIG. 8 is an explanatory diagram of a machine learning device and a feature extraction model in the second modified example. The second modified example is a further modified example of the first modified example. The flow rate predicting apparatus 70 and the flow rate predicting method according to the second modified example include a mesh feature amount learning unit that inputs mesh data and causes a machine learning device to deeply learn the feature of the mesh data as a feature amount. The learning device is different in that the mesh data feature amount is input as learning data instead of the mesh data.

At the time of learning the flow rate, the data processing unit 44 acquires and receives from the meteorological observation data storage unit 11, the weather forecast data storage unit 12, the flow rate storage unit 13, and the seasonal term data generation unit 43, as in the second modification. Each data is converted and transmitted to the learning unit 75.
The learning unit 75 includes a mesh feature amount learning unit 80, a mesh feature amount parameter storage unit 82, and a mesh feature amount extraction unit 83. The mesh feature amount learning unit 80 and the mesh feature amount extraction unit 83 may be, for example, software or a program executed by the CPU in the information processing device. Further, the mesh feature amount parameter storage unit 82 may be realized by a storage device such as a semiconductor memory or a magnetic disk provided inside or outside the information processing device.

The mesh feature amount learning unit 80 receives the weather observation data before the first past time and the weather prediction data before the first past time from the data processing unit 44.
The mesh feature amount learning unit 80 includes a machine learning device 81. The machine learning device 81 performs convolutional deep learning, and is realized by a convolutional auto encoder. The machine learning device 81 is used as a program module that is a part of artificial intelligence software, and generates trained models 84 and 91 described below in which appropriate learning parameters have been learned.
In this modification, the output of the feature extraction model 84 is used as an input for the learning of the weak learner 86 in the prediction model learning unit 85. Therefore, the learning by the machine learning device 81 is performed before the learning by the weak learning device 86.
As described above, the meteorological observation data 2 and the meteorological prediction data 3 include mesh data. The mesh feature amount learning unit 80 inputs the mesh data included in each of the meteorological observation data before the first past time and the weather prediction data before the first past time to the machine learning device 81 as the input mesh data 81a.

The machine learning device 81 includes a convolution processing unit 81b, a total combination unit 81c, and a deconvolution processing unit 81d.
The convolution processing unit 81b executes convolution processing, pooling processing, and the like on the input mesh data 81a, and generates a feature map that is an input to the total combination unit 81c. The value of the feature map is stored in each input node of the total combining unit 81c.
In the total connection unit 81c, the weighted sum is calculated from the input node to the output node via the intermediate node based on the connection weight set between the nodes.
The value of each output node of the total combination unit 81c is input to the deconvolution processing unit 81d. In the deconvolution processing unit 81d, processing symmetrical to that of the convolution processing unit 81b is executed. That is, the input mesh data 81a was compressed by the convolution processing unit 81b into a low dimension before reaching the total combining unit 81c, but the deconvolution processing unit 81d operates so as to be restored from the low dimension compressed state. To do. As a result, output mesh data 81e having the same resolution as the input mesh data 81a is generated.
In the machine learning device 81, machine learning is performed so that the output mesh data 81e becomes mesh data that reproduces the input mesh data 81a.

As described above, the machine learning unit 81 is a convolutional auto encoder, and the input mesh data 81a is low-dimensionally compressed in the convolution processing unit 81b and is input to the total combination unit 81c to be maximally compressed, and then deconvoluted. The processing unit 81d restores the reproduced mesh data 81e. That is, the total connection part 81c is a part where the features of the input mesh data 81a are in the most compressed state, and the value of the output node of the total connection part 81c is an abstract condensation of the features of the input mesh data 81a. It can be treated as the feature amount, that is, the mesh data feature amount 81f.

The present modification conceptually utilizes the above-described characteristics of the convolutional auto encoder. That is, the input mesh data 81a is obtained by inputting each of the input mesh data 81a to the feature extraction models 84 and 91, which are the machine learning units 81 for which learning has been completed, and extracting the value of the output node of the total connection unit 81c. It is converted into the mesh data feature amount 81f. The prediction model learning unit 85 causes each weak learner 86 to perform ensemble learning using the mesh data feature amount 81f obtained by converting the mesh data instead of each mesh data.
For this reason, in the present modification, the portions from the convolution processing unit 81b to the total combination unit 81c, which are surrounded by the chain double-dashed line in FIG. 8, correspond to the feature extraction models 84 and 91 that are the learned models. Therefore, the value of each parameter corresponding to this portion, that is, the value of the weight of each filter used in the convolution processing unit 81b, the value of each connection weight of the total combination unit 81c, and the like are used as mesh feature amount parameters. It is transmitted to and stored in the parameter storage unit 82.

The mesh feature amount parameter storage unit 82 stores, as mesh feature amount parameters, each parameter of the machine learning device 81 that has been learned by the mesh feature amount learning unit 80 as described above.

The mesh feature amount extraction unit 83 acquires the mesh feature amount parameters of the machine learning unit 81 stored in the mesh feature amount parameter storage unit 82, and constructs a feature extraction model 84 corresponding to the machine learning unit 81.
The mesh feature amount extraction unit 83 inputs each mesh data included in the meteorological observation data before the first past time and the meteorological prediction data before the first past time to the feature extraction model 84 as the input mesh data 81a, and each input. A mesh data feature amount 81f corresponding to the mesh data 81a is generated.
The mesh feature amount extraction unit 83 transmits each mesh data feature amount 81f to the prediction model learning unit 85.

The prediction model learning unit 85 trains the weak learner 86 to generate the learned prediction model 93. In this modification, the prediction model learning unit 85 replaces the meteorological observation data before the first past time in which the mesh data is replaced with the corresponding mesh data feature amount 81f and the mesh data with the corresponding mesh data feature amount 81f. The weather forecast data after the first past time, the flow rate data before the first past time, the seasonal term data corresponding to the first past time, and the correct value, that is, the flow rate data after the first past time are used as learning data. Ensemble learning is performed on the plurality of weak learners 86 so as to predict the predicted flow rate after the first past time.
When the learning ends, the prediction model learning unit 85 stores the parameter adjusted by the learning of each weak learner 86 in the prediction model parameter storage unit 46 as the prediction model parameter.

At the time of predicting the flow rate, the data processing unit 44 acquires and receives from the weather observation data storage unit 11, the weather prediction data storage unit 12, the flow rate storage unit 13, and the seasonal term data generation unit 43, as in the first modification. Each data is transmitted to the prediction unit 77.
The prediction unit 77 receives each data from the data processing unit 44.
The prediction unit 77 includes a mesh feature amount extraction unit 90 having a feature extraction model 91. The mesh feature amount extraction unit 90 and the feature extraction model 91 have the same configurations as the mesh feature amount extraction unit 83 and the feature extraction model 84 provided in the learning unit 75.
The mesh feature amount extraction unit 90 inputs each mesh data included in the weather observation data before the present and the future weather forecast data received by the prediction unit 77 into the feature extraction model 91 as the input mesh data 81a, and inputs each mesh data. A mesh data feature amount 81f corresponding to the mesh data 81a is generated.
The mesh feature amount extraction unit 90 transmits each mesh data feature amount 81f to the flow rate prediction unit 92.

The flow rate predicting unit 92 acquires the prediction model parameter corresponding to each weak learning device 86 stored in the prediction model parameter storage unit 46, and constructs the prediction model 93 corresponding to each weak learning device 86 based on each of these. To do.
The flow rate prediction unit 60 inputs each data received from the data processing unit 44 into each of the plurality of prediction models 93. Here, in the meteorological observation data before the present and the future meteorological prediction data, the mesh data included therein is replaced with the corresponding mesh data feature amount 81f, and then input to each prediction model 93. On the other hand, each of the plurality of prediction models 93 outputs a provisional future predicted flow rate.
The flow rate predicting unit 92 acquires a plurality of provisional future predicted flow rates output by the plurality of prediction models 93 and inputs them to the coupler. The coupler calculates and predicts the future flow rate 6 based on the plurality of provisional future predicted flow rates.

In the flow rate prediction device 70 of the present modification, each of the meteorological observation data 2 and the meteorological prediction data 3 includes mesh data 81a in which the area included as an observation/prediction target is divided into a mesh by latitude and longitude on a map, The mesh feature amount learning unit 80 that inputs the mesh data 81a and causes the machine learning unit 81 to perform deep learning using the features of the mesh data 81a as the feature amount 81f, and the feature extraction models 84 and 91 that are the machine learning unit 81 that has finished learning. On the other hand, a mesh feature amount extraction unit 83, 90 that inputs the mesh data 81a and extracts the feature amount 81f of the mesh data 81a is provided, and the prediction model learning unit 85 replaces each of the mesh data 81a with Using the feature amount 81f obtained by inputting each of the mesh data 81a to the mesh feature amount extracting unit 83, the plurality of weak learners 86 are ensemble-learned, and the flow rate predicting unit 92 replaces each of the mesh data 81a. Then, the future flow rate 6 of the river is predicted by using the feature amount 81f obtained by inputting each of the mesh data 81a to the mesh feature amount extracting unit 92.
If the river basin is large, the meteorological mesh may be large. Moreover, when it is desired to input values in many time series as the meteorological observation data 2 and the meteorological prediction data 3, the number of input dimensions becomes enormous. In such a case, if the mesh data is directly input to the weak learning device 86 and the weak learning device 86 is ensemble-learned, the learning efficiency may not be good and accurate prediction may not be possible.
According to the above configuration, instead of the mesh data included in the meteorological observation data 2 and the meteorological prediction data 3, the features of these mesh data are reflected, and the mesh data features are compressed and the data amount is reduced. The weak learner 86 is learned using the quantity 81f. As a result, the weak learner 86 can be learned more efficiently, and high-precision prediction with higher generalization performance is possible.

Further, the machine learning device 81 and the feature extraction models 84 and 91 are realized by a convolutional auto encoder.
With the above configuration, the flow rate predicting device 70 can be appropriately realized.

It goes without saying that the second modified example has other effects similar to those of the embodiment and the first modified example described above.

The flow rate predicting apparatus and the flow rate predicting method of the present invention are not limited to the above-described embodiment and each modified example described with reference to the drawings, and other various modified examples are conceivable within the technical scope thereof. To be

For example, in the above-described embodiment and each modification, the prediction model learning unit performs ensemble learning for each of the plurality of weak learners by boosting, but the present invention is not limited to this. That is, the prediction model learning unit may perform ensemble learning on each of the plurality of weak learners by bagging.
In bagging, a small amount of data is extracted from all the learning data and input to each weak learning device for learning. That is, in bagging, unlike the boosting, the weak learners are learned in parallel. As described above, the data in which the flow rate is expected to be large, such as in the case of a typhoon or a torrential rain, is significantly less than the data in which the flow rate is small. However, for example, by extracting data with a small flow rate and increasing the flow rate and specializing the data with a large flow rate to learn one or more weak learners, the prediction accuracy is high when the flow rate increases. A predictive model can be generated.
Thus, even when bagging is adopted instead of boosting, by extracting a small amount of data, it is possible to improve the prediction accuracy when the flow rate is large.

Other than this, the configurations described in the above-described embodiment and each modification can be selected or changed to other configurations without departing from the gist of the present invention.

1, 40, 70 Flow rate predictor 2 Meteorological observation data 2A Meteorological observation data before the first past time (weather observation data at past time)
2B Meteorological observation data before the present (current meteorological observation data)
2m, 3m Mesh data 3 Weather forecast data 3A Weather forecast data after the first past time (weather forecast data at past time)
3B Future weather forecast data (current weather forecast data)
4 Flow rate data 4A Flow rate data before the first past time (flow rate data at past time)
4B Flow data before and after present (current flow data)
4C Flow rate data after the first past time, correct value (flow rate data at past time)
5A Predicted flow rate after the first past time (flow rate after the past time)
5B Provisional future predicted flow rate 6 Future flow rate 7 Date/time data 16,46 Prediction model parameter storage section 20, 50, 85 Prediction model learning section 21, 51, 86 Weak learner 30, 60, 92 Flow rate prediction section 31, 61, 93 Prediction model 43 Seasonal term data generation unit 65, 66 Function 80 Mesh feature amount learning unit 81 Machine learning device 81a Input mesh data (mesh data)
81e Output mesh data 81f Mesh data feature amount (feature amount)
82 mesh feature amount parameter storage unit 83, 90 mesh feature amount extraction unit 84, 91 feature extraction model

Claims (9)

A flow rate prediction device that predicts the flow rate of a specific river,
A plurality of weak learners for predicting the flow rate after the past time, using the weather observation data, the weather prediction data, and the flow rate data related to the river at the past time that is any time in the past as learning data. And a prediction model learning unit that generates a plurality of prediction models by ensemble learning
Each of the plurality of prediction models, at present, related to the river, the meteorological observation data, the meteorological prediction data, and the flow rate data is input to output a provisional future predicted flow rate, A flow rate predicting unit that predicts a future flow rate of the river based on a provisional future predicted flow rate,
A flow rate predictor equipped with. The flow rate prediction apparatus according to claim 1, wherein the prediction model learning unit performs ensemble learning on each of the plurality of weak learners by boosting. The flow rate prediction apparatus according to claim 1, wherein the prediction model learning unit performs ensemble learning on each of the plurality of weak learners by bagging. The learning data is related to the river, meteorological observation data before the past time, weather forecast data after the past time related to the river, discharge data before the past time of the river, and as a correct value. Is the discharge data of the river after the past time,
The flow rate prediction unit inputs, into each of the plurality of prediction models, weather observation data before and after the current related to the river, future weather forecast data related to the river, and flow rate data before and after the river. The flow rate predicting device according to any one of claims 1 to 3. The learning data further includes seasonal term data having different values depending on the season,
The flow rate prediction unit inputs the seasonal term data corresponding to the present time as learning data to each of the plurality of prediction models to output the tentative future predicted flow rate, and outputs the plurality of tentative futures. The flow rate predicting device according to any one of claims 1 to 4, which predicts the future flow rate of the river based on the predicted flow rate of. The flow rate predicting apparatus according to claim 5, wherein the seasonal term data is a continuous function and is an output value of a periodic function whose value is repeated with a cycle of one year. Each of the meteorological observation data and the meteorological prediction data includes mesh data in which the area included as an observation and prediction target is divided into a mesh shape by latitude and longitude on a map,
A mesh feature amount learning unit for inputting the mesh data and causing a machine learning device to perform deep learning using the feature of the mesh data as a feature amount;
A mesh feature amount extraction unit that inputs the mesh data and extracts the feature amount of the mesh data with respect to the feature extraction model that is the machine learning device that has finished learning,
Equipped with
The prediction model learning unit uses the feature amount obtained by inputting each of the mesh data to the mesh feature amount extraction unit, instead of each of the mesh data, to ensemble the plurality of weak learners. Learn and
The flow rate predicting unit predicts the future flow rate of the river by using the feature amount obtained by inputting each of the mesh data into the mesh feature amount extracting unit instead of each of the mesh data. The flow rate predicting device according to any one of claims 1 to 6. The flow rate prediction device according to claim 7, wherein the machine learning device and the feature extraction model are realized by a convolutional auto encoder. A flow rate prediction method for predicting the flow rate of a specific river,
A plurality of weak learners for predicting the flow rate after the past time, using the weather observation data, the weather prediction data, and the flow rate data related to the river at the past time that is any time in the past as learning data. Train an ensemble to generate multiple prediction models,
Each of the plurality of prediction models, at present, related to the river, the meteorological observation data, the meteorological prediction data, and the flow rate data is input to output a provisional future predicted flow rate, A flow prediction method for predicting a future flow of the river based on a provisional future predicted flow.

Images