JP2007205001A - Discharge forecasting apparatus - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a discharge forecasting apparatus which calculates drainage area mean precipitation data to reduce parameters by optimizing a loading ratio of precipitation in a number of locations or precipitation in a number of zones (a number of mesh precipitation values), construct a discharge forecasting model with good accuracy by improving learning effect, and achieves improved forecasting accuracy. <P>SOLUTION: The discharge forecasting apparatus 1 functions to forecast the discharge in a downstream area by using precipitation data at a plurality of locations in an upstream area, and it is formed of: a loading ratio determining means 70 for determining the loading ratio indicative of an effect extent of precipitation data of a certain location on a discharge in a downstream area for each of the precipitation data of a plurality of locations; a model constructing means 80 for constructing a discharge forecasting model by using a drainage area mean precipitation data calculated from the loading ratio and the precipitation data corresponding to the same; and a forecasting means 90 for inputting the drainage area mean precipitation data by the determined loading ratio into the constructed precipitation forecasting model, and generating forecasted discharge data concerning a forecasted discharge. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to a flow rate prediction device for predicting the flow rate of water flowing into a water storage facility such as a sewage treatment plant, a pumping station, a system control station, a power supply command station, a dam management station, a hydroelectric power station, or a river flow rate. .

  Predicting the flow rate of water flowing into the dam and the upstream river is very important for the safety and economics of dam operation. Predicting the flow rate several hours ahead can appropriately perform dam discharge during rainfall, and can help to ensure the safety of the basin. Forecasts from several hours to several days ahead, especially the forecast for the next day, can properly formulate a power generation plan, which can contribute to effective use of hydropower and improvement of economic efficiency.

  Conventionally, the prediction of the inflow of a dam is often performed based on the experience and intuitive knowledge of a skilled operator. Further, automation of prediction work up to several hours ahead has been realized, and as an example, various methods using a tank model, a storage function method, a neural network, etc. have been proposed, and the prediction accuracy has been improved. In the prediction method as described above, the amount of rainfall representing the rainfall situation around the dam and the upstream basin of the dam is input, and the flow rate or the change in the flow rate is output.

In this case, since the rainfall conditions around the dam and in the upstream basin of the dam vary depending on the point, rainfall at a plurality of points is used. Furthermore, in recent years, there is an example using not only a rain gauge installed on the ground but also a wide range of mesh rainfall using an analysis result by a rainfall radar.
Note that the rain gauge only measures the rainfall at a single point in a wide area, so it cannot be said that it represents the overall rainfall situation if the rainfall situation differs locally. Also, the mesh rainfall may be too narrow. Therefore, it is also practiced to average the rainfall at a plurality of points and the mesh rainfall at a plurality of sections.

  Here, when averaging the rainfall or the mesh rainfall, it is necessary to consider the time (flowing time) from the time when rainfall started at each point to the time when it flows into the prediction target dam. This is because, in general, there is a time delay of about several hours from the start of rain in the upstream until the amount of inflow at the downstream dam point increases, which naturally depends on the point of rainfall.

  For example, if the flow time from the upstream point A to the C dam is 1 hour and the flow time from the upstream point B to the C dam is 2 hours, the average rainfall data is calculated by shifting by 1 hour as shown in the following formula. To do.

  Thereby, the average rainfall with a strong correlation with the flow rate can be obtained. Since the rainfall is not only temporary rainfall, but also the rainfall for the past several hours accumulates and appears as a flow rate, the average rainfall is used for the past several hours for prediction. It goes without saying that how many hours of average rainfall should be used depends on the dam.

Further, as a method of averaging, there are a method of calculating and using an average rainfall for each river when there are a plurality of rivers, and a method of using a basin average rainfall that summarizes all the basins.
As a prior art, for example, Patent Document 1 (Japanese Patent Laid-Open No. 2001-167078, the title of the invention “flow rate prediction method by a dam or a river”) is disclosed. A method using rainfall is disclosed.
Patent Document 2 (Japanese Patent Laid-Open No. 5-134715, entitled “Neural Network Applied Rainwater Inflow Prediction Device”) relates to a method for predicting the amount of rainwater in a sewage treatment plant or a pump station, and has an upstream basin. A method for predicting the inflow amount using the mesh rain amount is disclosed.

JP 2001-167078 A (paragraph numbers 0015 to 0051) Japanese Patent Laid-Open No. 5-134715 (paragraph numbers 0016 to 0052)

  In the prior art described in Patent Document 1 above, a method of using the basin average rainfall of the upstream basin for the prediction of the inflow of the dam is disclosed. Specifically, the rainfall of each river is weighted using the basin area of the upstream river. On average, basin average rainfall is calculated. However, this method is considered to be able to calculate an accurate average basin rainfall if the basin is clearly separable, but generally the boundary of the basin is not clear. Accordingly, the basin area cannot be set appropriately, and it is difficult to calculate an appropriate basin average rainfall. Moreover, the calculation method about the case where several rain gauges exist in the same basin is not disclosed.

  Furthermore, if the disclosed technique of calculating the basin average rainfall according to the basin area ratio is applied to the mesh rainfall, a method of averaging the mesh rainfall and calculating the basin average rainfall can be considered. However, not all mesh rainfall over a wide area flows into the inflow of the storage facility to be predicted or the flow of the river. In other words, when using general point rainfall, it is difficult to calculate an appropriate basin average rainfall using point rainfall, and when using mesh rainfall, it is difficult to set an appropriate weight ratio for each mesh. It is.

  Moreover, in the prior art described in Patent Document 2, a method is used in which the mesh rainfall is not averaged and is used as it is. That is, it is a method that uses all rainfall data over the past several hours for each mesh. In this method, it is not necessary to explicitly consider the weight ratio for each mesh, but the number of input factors of the neural network used as the flow prediction model is enormous and the structure of the neural network is large and complicated. turn into.

  For example, if the number of meshes is 5 meshes vertically and horizontally, and the time to be considered as an input factor is from the present to the past 4 hours ago, input a considerable amount of rainfall data, 125 meshes for 5 meshes x 5 meshes x 5 hours. End up. It is good if the number of meshes is as small as several and the time to be considered as input is short, but in general, when the number of input factors increases, it takes a lot of time to learn the neural network, and the input / output relationship Because it is difficult to learn properly, it is not an appropriate method.

  Accordingly, the present invention has been made to solve the above-mentioned problems, and its purpose is to calculate basin average rainfall data by optimizing the weight ratio of rainfall at a plurality of points or rainfall at a plurality of sections (a plurality of mesh rainfalls). It is an object of the present invention to provide a flow rate predicting apparatus that reduces parameters and constructs a highly accurate flow rate prediction model by improving learning efficiency and also improves prediction accuracy.

A flow rate predicting apparatus according to claim 1 of the present invention includes:
A flow rate prediction device that predicts downstream flow rate using rainfall data at a plurality of upstream locations,
Using rainfall data at multiple points or sections upstream, and weight ratios that represent the degree of influence of rainfall data at a certain location on the downstream flow rate for each of the multiple rainfall data, A basin average rainfall calculating means for calculating a basin average weight for calculating a plurality of basin average rainfall data for each time point or a sum of a product of the weight ratio related to the rainfall data for a plurality of points or a plurality of sections;
Correlation coefficient calculating means for calculating a correlation coefficient by using a plurality of basin average rainfall data and flow rate data related to the basin average rainfall data by time,
A weight ratio evaluation means for determining whether the evaluation is a high weight ratio based on a correlation coefficient;
A weight ratio optimization means for determining a weight ratio that generates a new weight ratio by changing the weight ratio based on a predetermined rule;
The weight ratio determining basin average rainfall calculating means, the correlation coefficient calculating means, the weight ratio evaluating means, and the weight ratio determining weight ratio optimizing means are repeatedly functioned to correlate the basin average rainfall data with the discharge data. A weight ratio determining means for determining the weight ratio so as to increase,
Estimate the flow rate prediction model using the basin average rainfall data, which is the sum of the data of the rain data at a certain location and the weight ratio determined by the weight ratio determination means and the weight ratio determined for the multiple points or sections. A prediction model construction means to be constructed;
A prediction unit that inputs the basin average rainfall data based on the weight ratio determined by the weight ratio determination unit to generate the prediction flow data related to the predicted flow, with respect to the flow prediction model constructed by the prediction model construction unit;
It is characterized by providing.

A flow rate predicting apparatus according to claim 2 of the present invention is
A flow rate prediction device that predicts downstream flow rate using rainfall data at a plurality of upstream locations,
Using rainfall data at multiple points or sections upstream, and weight ratios that represent the degree of influence of rainfall data at a certain location on the downstream flow rate for each of the multiple rainfall data, A basin average rainfall calculation means for predictive model construction for calculating a plurality of basin average rainfall data for each time point or a sum of the products of the weight ratio relating to this rainfall data for a plurality of points or a plurality of sections;
A prediction model learning means for learning a flow prediction model using a plurality of basin average rainfall data and flow rate data related to the basin average rainfall data according to time;
Predicted by the flow prediction model using the basin average rainfall data related to the past results and the corresponding flow rate data to generate predicted flow data, and evaluated using the prediction error between the actual flow data and the predicted flow data And a prediction model / weight ratio evaluation means for determining whether the evaluation is a high flow rate prediction model and a weight ratio,
A weight ratio optimization means for constructing a prediction model for generating a new weight ratio by changing the weight ratio based on a predetermined rule;
The prediction model construction basin average rainfall calculation means, the prediction model learning means, the prediction model / weight ratio evaluation means and the prediction model construction weight ratio optimization means are repeatedly functioned so that the prediction error is reduced. A prediction model construction means for determining a prediction model and a weight ratio;
Predictive means for generating predicted flow rate data related to the predicted flow rate by inputting the basin average rainfall data based on the weight ratio determined by the predictive model constructing unit to the flow rate predictive model constructed by the predictive model constructing unit;
It is characterized by providing.

A flow rate predicting apparatus according to claim 3 of the present invention is
A flow rate prediction device that predicts downstream flow rate using rainfall data at a plurality of upstream locations,
Using rainfall data at multiple points or sections upstream, and weight ratios that represent the degree of influence of rainfall data at a certain location on the downstream flow rate for each of the multiple rainfall data, A basin average rainfall calculating means for calculating a basin average weight for calculating a plurality of basin average rainfall data for each time point or a sum of a product of the weight ratio related to the rainfall data for a plurality of points or a plurality of sections;
Correlation coefficient calculating means for calculating a correlation coefficient by using a plurality of basin average rainfall data and flow rate data related to the basin average rainfall data by time,
A weight ratio evaluation means for determining whether the evaluation is a high weight ratio based on a correlation coefficient;
A weight ratio optimization means for determining a weight ratio that generates a new weight ratio by changing the weight ratio based on a predetermined rule;
The weight ratio determining basin average rainfall calculating means, the correlation coefficient calculating means, the weight ratio evaluating means, and the weight ratio determining weight ratio optimizing means are repeatedly functioned to correlate the basin average rainfall data with the discharge data. A weight ratio determining means for determining the weight ratio so as to increase,
Multiple basin average rainfall data for each point in time is calculated by summing the product of the rainfall data at a certain location and the weight ratio determined by the weight ratio determination means for multiple points or multiple sections. Basin average rainfall calculation means for predictive model construction,
A prediction model learning means for learning a flow prediction model using a plurality of basin average rainfall data and flow rate data related to the basin average rainfall data according to time;
Predicted by the flow prediction model using the basin average rainfall data related to the past results and the corresponding flow rate data to generate predicted flow data, and evaluated using the prediction error between the actual flow data and the predicted flow data And a prediction model / weight ratio evaluation means for determining whether the evaluation is a high flow rate prediction model and a weight ratio,
A weight ratio optimization means for constructing a prediction model for generating a new weight ratio by changing the weight ratio based on a predetermined rule;
The prediction model construction basin average rainfall calculation means, the prediction model learning means, the prediction model / weight ratio evaluation means and the prediction model construction weight ratio optimization means are repeatedly functioned so that the prediction error is reduced. A prediction model construction means for determining a prediction model and a weight ratio;
Predictive means for generating predicted flow rate data related to the predicted flow rate by inputting the basin average rainfall data based on the weight ratio determined by the predictive model constructing unit to the flow rate predictive model constructed by the predictive model constructing unit;
It is characterized by providing.

A flow rate predicting apparatus according to claim 4 of the present invention includes:
In the flow prediction apparatus according to any one of claims 1 to 3,
The flow rate prediction model is a neural network.

  According to the present invention as described above, the basin average rainfall data is calculated by optimizing the weight ratio of the rainfall at a plurality of points or the rainfalls at a plurality of sections (a plurality of mesh rainfalls), and the learning efficiency is improved. Therefore, it is possible to provide a flow rate prediction device that constructs a highly accurate flow rate prediction model and also realizes improvement of prediction accuracy.

Hereinafter, a flow rate predicting apparatus according to the best mode of the present invention will be described with reference to the drawings. FIG. 1 is a configuration diagram of a flow rate prediction apparatus according to this embodiment.
The flow rate prediction apparatus 1 includes an input unit 10, an output unit 20, a recording medium reading unit 30, a data transmission unit 40, a data collection unit 50, a data storage unit 60, a weight ratio determination unit 70, a prediction model construction unit 80, and a prediction unit 90. It has.

The input means 10 is a keyboard for performing manual input.
The output unit 20 is a display, a printer, or the like, and also performs various data display / printing.
The storage medium reading means 30 is a device that reads and writes information from and to a storage medium such as an FD (Flexible Disc), an MO (Magnet Optical Disc), or a USB memory.
The data transmission means 40 is a means for connecting to the network 2 and can output information from the data storage means 60 via the data transmission means 40 and the network 2 to the outside.

The data collecting means 50 is means for acquiring various data via a network 2 from a monitoring control system, a rain gauge or a weather information company in a plurality of areas, and further includes a rain data collecting means 51 and a flow data collecting means 52. .
The rainfall data collection unit 51 is a unit that collects past, present, and future rainfall data. First, the past and present rainfall data will be described. The past and present rainfall data here refers to, for example, rainfall data collected and input from measuring devices and sensors such as rain gauges installed independently in each region, or analysis results by rainfall radar This is rain data representing mesh rainfall in a predetermined section using. Location data indicating where the rainfall is is also added to the rainfall data, and time data, rainfall data, and location data are associated and registered. Moreover, the rainfall data about the past and present results may be rain data distributed from the weather information service company via the network 2. This rainfall data is also registered in association with time data, rainfall data, and location data.

  As the details of the rainfall data, it is assumed that continuous rainfall data for a certain period before and after the weather with different rainfall, such as particularly fine, light rain, heavy rain, and typhoon, is registered. It is also assumed that continuous rainfall data for a certain period of the period in which the rainfall and flow rate change from light rain to heavy rain is also registered. Furthermore, it is assumed that rainfall data that changes the rainy conditions such as light rain and heavy rain by changing the date is registered. These past and present rainfall data are stored in the rainfall database 62 of the data storage means 60 as shown in FIGS.

The rainfall data will be described assuming a model as shown in FIG. FIG. 2 is an explanatory diagram for explaining rainfall data and flow data in a river model.
In this model, tributaries 1, 2,..., I,... N enter the main stream upstream, and water with a flow rate Q flows into the prediction target 100 (for example, a water storage facility) downstream. Rain gauges 1, 2,..., I,... N are arranged near each basin of these tributaries, and rainfall data R (1), R (2). (I), ... R (n) is collected. As the necessary rainfall data, the rainfall data R (1),... R (n) of all the locations at a certain time is set as one set, and a plurality of sets of data are obtained for each time.
Of course, weather data including, for example, weather (sunny, rain, cloudy, snow), temperature, humidity, amount of sunlight, etc. may be collected in addition to the rainfall.

  Next, future rainfall data will be described. It also collects future weather forecasts and collects rainfall data included in weather forecasts. The rain data related to this prediction is included in the weather data obtained by receiving the weather forecast from the weather information service company, and is the rain data representing the predicted rainfall at a certain location / time. Future rainfall data is also stored in the rainfall database 62 of the data storage means 60 as shown in FIGS. Rainfall data is accumulated as a series of past, present, and future data.

  The flow rate data collecting unit 52 is a unit that collects flow rate data of a river to be predicted. As shown in FIG. 2, the flow rate data is data representing the flow rate Q of the river mainstream that flows into the prediction target 100. For example, the flow rate data is input every predetermined period or constantly from a flow meter or a monitoring system installed in a river near the prediction target 100. The flow meter and the monitoring system are flow meters that are connected via the network 2. This flow rate data is registered in association with time data. These past and present flow rate data are stored in the flow rate database 63 of the data storage means 60 as shown in FIGS.

  These rainfall data and flow rate data may be stored in the data storage means 60 and updated daily to store these data. The data storage means 60 may be a database in which rain data and flow rate data are registered in association with specific data that can specify the measurement time, season, and degree of rain as a main key. A plurality of other data are registered in the data storage unit 60, which will be described later.

The weight ratio determining means 70 reads the past rainfall data at a plurality of upstream points or sections from the rainfall database 62, reads the past flow data from the flow database 63, and based on such rainfall data and flow data. Thus, it is a means for determining a weight 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. To determine the weighting ratio here, increase the weighting ratio of the rainfall data where the increase / decrease in rainfall greatly affects the increase / decrease in downstream flow, and increase / decrease in rainfall will have a small effect on increase / decrease in downstream flow. It is to reduce the weight ratio of the rainfall data at the location.
Next, the reason for using the weight ratio will be described. Rainfall data was collected first, but because it covers multiple points and multiple sections of the basin, multiple rainfall data used for prediction are handled, and there is a problem that the number of parameters is too large. Furthermore, when the number of parameters is large, it becomes a complicated flow prediction model, and there is a problem that time and labor are required for model construction and flow prediction. In view of this, it is expressed as basin average rainfall data in which the weight of the rainfall data at a location that does not have a high influence on the flow rate is low and the weight of the rainfall data at a location that has a high influence on the flow rate is high. This basin average rainfall data is a parameter having a high influence on the flow rate because the increase and decrease of the basin average rainfall data and the increase and decrease of the flow rate are related. Furthermore, since the basin average rainfall data itself is a parameter that combines multiple rainfall data at multiple points, the number of parameters is reduced. And less time and effort.

  Specifically, as shown in the structure diagram of the weight ratio determining means in FIG. 3, the weight ratio determining means 70 is a weight ratio determining basin average rainfall calculating means 71, a correlation coefficient calculating means 72, a weight ratio evaluating means 73, a weighting. A ratio determining weight ratio optimization means 74 is provided. And each data of the weight ratio database 61 with which the data storage means 60 is provided, the rainfall database 62, and the flow volume database 63 is read and used. In the weight ratio database 61, a weight ratio in the middle of optimization and an optimized weight ratio are registered. The rainfall database 62 uses, in particular, past rainfall data among the rainfall data. In the flow rate database 63, past flow rate data among the flow rate data is used.

  First, the basin average rainfall calculation means 71 for determining the weight ratio reads the weight ratio from the weight ratio database 61, and also reads the rain data at a certain time in the past from the rain database 62 to calculate the basin average rainfall data. To do. This basin average rainfall data, which will be described later, is the sum of the product of the rainfall data at a certain point or section upstream and the weighting ratio relating to this rainfall data for a plurality of points or sections. The initial value of this weight ratio is an appropriate value (however, the sum is 1.0).

  In addition, as shown in FIG. 2, the downflow time until the rainfall R (1), R (2)..., R (i),. And the downstream side are different from each other, and each has a time delay. The flow time until the rainfall R (1) of the basin 1 appears in the downstream flow rate Q is a1,..., The flow time until the rainfall R (i) of the basin i appears in the downstream flow rate Q is ai,. The flow time an until the rainfall R (n) in the basin n appears in the downstream flow rate Q In this way, it is influenced by past rainfall.

  These rainfall amounts R (1), R (2)..., R (i),... R (n) are multiplied by a weight ratio. The weight ratios are w (1), w (2),..., W (i),. Even if it rains, the influence of the rainfall R (i) is small in the area where it flows into the river only a little, so the weight ratio w (i) is small. Since the influence is great, the weight ratio w (i) becomes large. In this case, consideration is given so that the sum of the weight ratios w (1),..., W (i),. The value of the weight ratio may be, for example, a binary value of 0 or a constant a. In this case, when the weight ratio is 0, the rainfall is not used, and when the weight ratio is a, the rain is used. For example, n = 6, w (1) = 0, w (2) = 0.25, w (3) = 0.25, w (4) = 0.25, w (5) = 0, w (6 ) = 0.25. Even in this case, the total sum of w (1),..., W (i),.

Also, the value of the weight ratio may be a decimal value from 0 to 1. In this case, for example, n = 6, w (1) = 0.1, w (2) = 0.5, w (3) = 0.05, w (4) = 0, w (5) = 0. .2, w (6) = 0.15. For example, when n = 6, w (1) = 0.01, w (2) = 0.01, w (3) = 0.005, w (4) = 0.005, w (5) = 0 .01, w (6) = 0.96. Thus, w (i) can take a value from 0 to 1, but in any case, the sum of the weight ratios w (1),..., W (i),. Take care to meet zero. A plurality of such basin average rainfall data are calculated for each time. The basin average rainfall data of a certain time t and R _t. In addition, since there is a downflow time ai that is delayed until the influence of the rainfall R (i) of a certain tributary on the upstream side appears in the flow rate Q of the prediction target 100 downstream, the amount of rain that flows into the prediction target 100 at a certain time t Let R (i) be R (i) _t-ai . Expressing basin average rainfall data R _t in the formula in river model shown in FIG. 2 in this case is as follows.

Thereby obtaining a basin average rainfall data R _t taken rainfall at a plurality of locations and multiple compartments. A plurality of the basin average rainfall data _Rt are acquired at different times t.

Subsequently, the correlation coefficient calculating unit 72 reads the basin average rainfall data R _t from the weight ratio determining basin average rainfall calculating unit 71 and the past flow rate data from the flow rate database 63, respectively, and these basin average rainfall data, It is a means for calculating a correlation coefficient for evaluating the weight ratio using the flow rate data. First, the correlation coefficient r is expressed by the following equation.

  The correlation coefficient is a statistical index that measures the degree of correlation between two data. The correlation coefficient takes a value of −1 ≦ r ≦ + 1, meaning that positive correlation is obtained when r> 0, negative correlation is given when r <0, and no correlation is given when r = 0. For example, when the correlation coefficient r is close to 1 (for example, 0.8 or more), it is determined that the increase / decrease in the basin average rainfall and the increase / decrease in the flow rate show the same tendency.

  The weight ratio evaluation means 73 determines whether or not the evaluation is a high weight ratio based on the correlation coefficient r calculated by the correlation coefficient calculation means 72. If the evaluation is high, this weight ratio (more specifically, w (1),..., W (i),..., W (n)) are registered in the weight ratio database 61. First, the evaluation function E is used in the evaluation based on the correlation coefficient r. Although various calculation methods can be considered as the evaluation function E, the following equation is shown as an example.

Here, a value output by inputting a correlation coefficient of a specific value into the evaluation function E is set as an evaluation value, and evaluation is performed based on the evaluation value. When the evaluation value is close to 1, it indicates that the correlation coefficient is close to 1 (positive correlation) or -1 (negative correlation), and the correlation between the basin average rainfall data and the discharge data is high and the evaluation is high. Further, when the evaluation value is close to 0, it indicates that the correlation coefficient is close to 0 (no correlation), and the correlation between the basin average rainfall data and the discharge data is low and the evaluation is low.
That is, it is determined whether or not the evaluation value obtained by inputting a specific correlation coefficient r into the evaluation function E is an evaluation value that can obtain a higher evaluation than before. Since the data is a parameter that has a large influence on the increase or decrease of the downstream flow rate, the weight ratio of the basin average rainfall data is registered in the weight ratio database 61. At the first time, the weight ratio used as the initial value is registered. Thereafter, the weight ratio is overwritten and registered in the weight ratio database 61 every time a higher evaluation is obtained than before.

  In the weight ratio optimization means 74 for determining the weight ratio, the weight ratio obtained by the weight ratio evaluation means 73 is further changed by increasing / decreasing the weight ratio based on a predetermined rule, thereby changing the weight ratio database as a new weight ratio. 61 is a means for separately registering. It should be noted that the original weight ratio obtained with high evaluation is registered in the weight ratio database 61. Here, the weighting ratio can be changed by applying various optimization methods such as a mathematical programming method and a metaheuristic method as a predetermined rule for changing the weighting ratio. The details of the optimization technique to be applied are not the gist of the present invention, and will be omitted.

Thereafter, by using the weight ratio newly added and registered in the weight ratio database 61, the basin average rainfall calculating means 71, the correlation coefficient calculating means 72, and the weight ratio evaluating means 73 are made to function to calculate the basin average rainfall data. The weight ratio is evaluated, and the previous evaluation and the newly obtained evaluation are compared. If the evaluation is better than the previous one, the new weight ratio is overwritten and registered in the weight ratio database 61 as the best weight ratio. The weight ratio optimization means 74 rewrites the weight ratio based on a predetermined rule. Hereinafter, the weight ratio determining basin average rainfall calculating means 71, the correlation coefficient calculating means 72, the weight ratio evaluating means 73, and the weight ratio determining weight ratio optimizing means 74 are repeatedly functioned, and finally the highest evaluation (correlation) , W (i),..., W (n) are determined as the weight ratio. Note that if the evaluation does not become the highest for a predetermined period after the evaluation of the evaluation function E becomes the highest, the repetition is terminated.
Such basin average rainfall data is determined such that the flow rate on the downstream side increases or decreases with the same tendency according to the increase or decrease of the rainfall in each basin i.

Subsequently, model construction is performed by the prediction model construction means 80.
As shown in detail in FIG. 4, the prediction model construction unit 80 includes a basin average rainfall calculation unit 81 for prediction model construction, a prediction model learning unit 82, and a prediction model evaluation unit 83.
And each data of the weight ratio database 61 with which the data storage means 60 is provided, the rainfall database 62, the flow volume database 63, and the prediction model database 64 is used.

  In order to construct a flow rate prediction model, rainfall data that is a past record that has already occurred and flow rate data that is a past record are used. In particular, continuous rainfall data and flow rate data for a certain period of time before and after the weather with different rainfall such as clear, light rain, heavy rain, and typhoon are used. Further, continuous rainfall data and flow rate data for a certain period of time during which the rainfall and flow rate change, such as from light rain to heavy rain, are used. Further, rainfall data and flow rate data are used in which the rainy conditions such as light rain and heavy rain differ by changing the month and day. Rainfall data at multiple locations is acquired while satisfying the above conditions. Further, flow rate data corresponding to the time is acquired.

  First, the basin average rainfall calculating means 81 for constructing the prediction model reads out the optimum weight ratio having a high evaluation previously determined by the weight ratio determining means 70 from the weight ratio database 61, and the past rainfall from the rain database 62. The data is read out, and basin average rainfall data is calculated by equation (2). Furthermore, basin average rainfall data for a plurality of times is calculated for each time.

The prediction model learning means 82 reads the basin average rainfall data from the prediction model construction basin average rainfall calculation means 81, reads a plurality of flow rate data corresponding to the basin average rainfall data in time from the flow rate database 63, and The flow rate prediction model is constructed by using the basin average rainfall data calculated by the basin average rainfall calculation means 81 for the flow rate data and the prediction model for a plurality of times.
Here, since the details of the prediction method are not the gist of the present invention, a detailed description thereof will be omitted, but typical flow prediction models include multiple regression models, neural network models, fuzzy inference models, autoregressive models, Kalman filters. Models, case-based reasoning models, etc. In this embodiment, a prediction method when a neural network model is adopted as a flow rate prediction model will be briefly described. Since the neural network model is described in various documents, it will be briefly described. Generally, the neural network model has a three-layer neural network structure including an input layer, an intermediate layer, and an output layer. Each layer of the layer and the output layer is provided with a neuron element expressed by a sigmoid function, and has a connection between elements of the input layer and the intermediate layer and between elements of the intermediate layer and the output layer.

  In this neural network, elements in the input layer correspond to input factors, and elements in the output layer correspond to output factors. The degree of coupling between neurons is represented by a coupling coefficient, and this coupling coefficient is a coefficient for representing the weight of coupling between elements of the neural network. If the coupling coefficient is large, the coupling has a weight, that is, a necessary coupling, and if the coupling coefficient is small, the coupling has no weight, that is, an unnecessary coupling. . By updating the magnitude of the coupling coefficient w (i), it is possible to learn a nonlinear relationship between input and output.

The number of input layer elements is n, the number of intermediate layer elements is m, the number of output layer elements is 1, the coupling coefficient between the input layer and the intermediate layer is w _ij , the coupling coefficient between the intermediate layer and the output layer is w _j , and the input is x = ( x ₁ , x ₂ , x ₃ ,..., x _n ), the input / output relationship is represented by the following equations. The input to the intermediate layer element j is as follows.

  Further, the output of the intermediate layer element j is as follows.

  The input to the output layer element is as follows.

  Further, the output of the output layer element is as follows.

In such a neural network model, nonlinear input / output relations can be processed by changing the model structure and the coupling coefficient, so that it is frequently used as a flow rate prediction model.
Various prediction model construction of such neural network is to obtain a desired output value from an output layer element (output factor) for input values (time series data) input to a plurality of input layer elements (input factors). In other words, the coupling coefficient between the input layer and the intermediate layer and between the intermediate layer and the output layer is changed. As a result, the coupling coefficient is determined.

  Learning data is necessary for the learning of the neural network. The basin average rainfall data at a certain time in the past calculated by the basin average rainfall calculating means 81 for constructing the prediction model and the flow data at the same time read from the flow database 63 are slightly later. The flow rate data at the time is used. They collect basin average rainfall data and flow rate data at a certain time for input, and flow data at a little later time for output over a plurality of times, and prepare a plurality of input / output data sets for the neural network. Thus, for example, a neural network is constructed using various known learning algorithms such as a known back-propagation method. The data of the coupling coefficient of the constructed flow rate prediction model is written and stored in the prediction model database 64.

  In this embodiment, basin average rainfall data (for example, three points before a time, a certain time T, and after a time) and flow data at a certain time T are used as input factors. Is a means for constructing a flow rate prediction model using flow rate data at a time delayed for a predetermined period from (for example, a time after a certain time T) as an output factor. In the prediction based on the flow rate prediction model by the neural network of this embodiment, when the basin average rainfall data and current flow rate data a hour before, present, and a hour in the future are input, flow data in the future a time is obtained.

  The prediction model evaluation unit 83 is a unit that evaluates the flow rate prediction model. If the evaluation is low, learn again. Thereafter, the prediction model learning unit 82 and the prediction model evaluation unit 83 are repeatedly functioned to learn the flow rate prediction model until it approaches the best flow rate prediction model. As a result, a flow prediction model corresponding to the basin average rainfall data is constructed.

Subsequently, the prediction unit 90 performs prediction using these flow rate prediction models.
As shown in detail in FIG. 5, the prediction unit 90 includes a prediction basin average rainfall calculation unit 91 and a prediction flow rate calculation unit 92.
First, the basin average rainfall calculation means 91 for prediction reads out the optimum weight ratio previously determined by the weight ratio determination means 70 from the weight ratio database 61, and the past, present, and future data read from the rainfall database 62. Using the rainfall data, basin average rainfall data is generated a hour ago, currently, and a hours in the future.

The predicted flow rate calculation unit 92 uses the flow rate prediction model determined by the coupling coefficient from the prediction model database 64, and the basin average rainfall data calculated by the basin average rainfall calculation unit 91 or the flow rate data read from the flow rate database 63. Is input to the flow rate prediction model to generate future predicted flow rate data. For example, in the previous example, the basin average rainfall data and current flow rate data a hour before, currently, and a hours in the future are input to the flow rate prediction model to obtain predicted flow rate data.
The predicted flow rate data obtained in this way is registered in the predicted value database 65 of the data storage means 60. Further, for example, it is displayed on another terminal via the data transmission means 40 and the network 2, is output by the output means 20, or is recorded by the recording medium reading means 30.

  The flow prediction apparatus 1 of this embodiment has been described above. In this embodiment, the computer includes one of the above units, the input unit 10 is a keyboard, the output unit 20 is a display or printer, the recording medium reading unit 30 is an FD or USB memory, and the data storage unit 60 is a hard disk or main storage device. The weight ratio determination means 70, the prediction model construction means 80, and the prediction means 90 may be configured to function by a CPU program.

  According to the present invention described above, it is possible to reduce the cost by reducing the construction time and prediction time of the flow rate prediction model by reducing the parameters. It is also possible to obtain a prediction value with high prediction accuracy. Previously, there was no standard for calculating average basin rainfall, but the present invention simplifies the rainfall of multiple basins as average basin rainfall, thereby simultaneously reducing costs and improving prediction accuracy by shortening the development period. Is possible.

Next, an embodiment in which the flow rate prediction apparatus described above is further improved will be described with reference to the drawings. FIG. 6 is a block diagram of a flow rate prediction apparatus according to another embodiment, and FIG. 7 is a structural diagram of another model construction means.
As shown in FIG. 6, the flow rate predicting apparatus 1 ′ includes an input unit 10, an output unit 20, a recording medium reading unit 30, a data transmission unit 40, a data collection unit 50, a data storage unit 60, a prediction model construction unit 80 ′, Prediction means 90 is provided. In this embodiment, since the weight ratio is determined within the improved prediction model construction means 80 ′ as shown in FIG. 7, the weight ratio determination means 70 is eliminated. In this embodiment, the configuration is the same as that of the previous embodiment except for the prediction model construction means 80 ', and the same reference numerals are given and redundant description is omitted. As shown in FIG. 7, the prediction model construction means 80 ′ includes a prediction model construction basin average rainfall calculation means 81, a prediction model learning means 82, a prediction model / weight ratio evaluation means 84, and a weight ratio optimization means for prediction model construction. 85. Compared with the prediction model construction means 80 shown in FIG. 4, the prediction model / weight ratio evaluation means 84 and the prediction model construction weight ratio optimization means 85 are particularly different in FIG. Explain with emphasis on the differences.

  First, the basin average rainfall calculation means 81 for predictive model construction reads the weight ratio from the weight ratio database 61 and also reads the rain data from the rain database 62 to calculate the basin average rain data. At first, a weight ratio which is an appropriate initial value is read out.

Then, a flow prediction model is constructed using this basin average rainfall data.
The prediction model learning means 82 reads the basin average rainfall data from the prediction model construction basin average rainfall calculation means 81, reads out the basin average rainfall data and the flow rate data corresponding to the time from the flow rate database 63, and these flow rates A flow prediction model is constructed using data and basin average rainfall data. In this embodiment, a neural network is used as the flow rate prediction model.
Also in this embodiment, as in the previous embodiment, the basin average rainfall data _Rt and the actual flow data Q are input as input factors of the neural network, and the flow data Q at the corresponding time are input as output factors, respectively. To learn.

The prediction model / weighted ratio evaluation means 84 is means for evaluating the learned flow rate prediction model. For details, it generates a predicted flow rate data Qf predicted by a flow rate prediction model by using the flow rate data Q performance and the corresponding basin average rainfall data R _t according to past experience, and the flow rate data Q Actual predicted flow rate data Qf And a means for determining whether the evaluation is a flow rate prediction model and a weighted ratio with high evaluation.
First, to calculate the prediction error e between the flow rate data Q of the predicted flow rate data Qf and actual case of using a watershed average rainfall data R _t. Here, the suffix i indicates the time of the predicted flow rate data Qf and the actual flow rate data Q. Estimated flow rate data Qf _i obtained by inputting the flow rate data Q _{i at the} time i with past results for evaluation accumulated in the flow rate database 63 and the basin average rainfall data R _t at the corresponding time i into the flow rate prediction model, Are calculated for a large number of times, and a prediction error e _i at a certain time i is calculated by the following equation using the flow rate data Q _i and the predicted flow rate data Qf _{i at the} multiple times.

The prediction error e _i at a plurality of times is calculated by varying the time i. These prediction errors e _i are substituted into the evaluation function E. Although various calculation methods can be considered as the evaluation function E, the following equation is shown as an example.

Here, the value output to the evaluation function E by entering a specific prediction error e _i as the evaluation value is evaluated on the basis of the evaluation value. When the evaluation value is large, it indicates that the difference between the predicted flow rate data Qf and the actual flow rate data Q is small, and the prediction accuracy of the prediction model is high and the evaluation is high. Further, when the evaluation value is small, it indicates that there is a large difference between the predicted flow rate data Qf and the actual flow rate data Q, and the prediction accuracy of the prediction model is low and the evaluation is low.
Then, it is determined whether or not the evaluation is higher than before. If the evaluation is higher (if the evaluation value is higher than before), the coupling coefficient that is the learning result is rewritten and registered in the prediction model database 64. The weight ratio when the evaluation value becomes high is also rewritten and registered in the weight ratio database 61.
In the weight ratio optimization means 85 for constructing the prediction model, the weight ratio evaluated as high by the prediction model / weight ratio evaluation means 84 is further increased or decreased based on a predetermined rule such as a mathematical programming method or a metaheuristic method. It is a means of registering separately in the weight ratio database 61 as a new weight ratio. The details of the optimization technique to be applied are not the gist of the present invention, and will be omitted. For the weight ratio w (i) registered in the weight ratio database 61, the weight ratio used as the initial value is registered at the first time.

  Subsequently, new basin average rainfall data using an optimum weight ratio is generated by the prediction model construction basin average rainfall calculation means 81, the flow prediction model is learned by the prediction model learning means 82, and the prediction model / weight ratio evaluation means 84 is generated. If the evaluation is lower than the previous evaluation, the coupling coefficient and the weight ratio w (i) are discarded. If the evaluation is higher, the coupling coefficient and the weight ratio w (i) are rewritten and registered. Then, a new weight ratio w (i) is determined by the prediction model construction weight ratio optimization means 85, and these prediction model construction basin average rainfall calculation means 81, prediction model learning means 82, prediction model / weight ratio evaluation. The means 84 and the weight ratio optimization means 85 for constructing the prediction model are repeatedly functioned to determine the coupling coefficient of the flow rate prediction model while determining the weight ratio w (i) having a high evaluation. As a result, a flow prediction model corresponding to the basin average rainfall data is constructed.

  Then, the prediction is performed by the prediction means 90. The prediction by the prediction means 90 is the same as the prediction described above, and a duplicate description is omitted. According to this embodiment, since the weight ratio w (i) that reduces the output error of the flow prediction model is obtained, the optimum weight ratio w (i) can be obtained while improving the accuracy of the flow prediction model.

Next, an embodiment in which the above-described flow rate prediction apparatus is further improved will be described. In the previous embodiment, the weight ratio determination means 80 ′ did not use the weight ratio determination means 70, but in this embodiment, the weight ratio determination means 70 shown in FIG. 3 and the prediction model construction means shown in FIG. 80 'is used in combination, and the weight ratio is determined by both. For this reason, the configuration is such that a prediction model construction unit 80 ′ is arranged in place of the prediction model construction unit 80 in the flow rate prediction apparatus 1 of FIG.
In addition, although mentioned later, the point evaluated using the evaluation function E shared by the weight ratio evaluation means 73 of the weight ratio determination means 70 and the prediction model / weight ratio evaluation means 84 of the prediction model construction means 80 ′ is also the previous form. It is different.

More specifically, as shown in FIG. 3, the weight ratio determination means 70 is a weight ratio determination basin average rainfall calculation means 71, a correlation coefficient calculation means 72, a weight ratio evaluation means 73, a weight ratio determination weight ratio optimization means. 74.
More specifically, as shown in FIG. 7, the prediction model construction means 80 ′ includes a prediction model construction basin average rainfall calculation means 81, a prediction model learning means 82, a prediction model / weight ratio evaluation means 84, and a weight ratio for prediction model construction. Optimization means 85 is provided.

First, the weight ratio determining means 70 reads past rainfall data at a plurality of upstream points or sections from the rainfall database 62, reads past flow data from the flow database 63, and rain data and flow data like these. Is a means for determining a weight 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.
First, the basin average rainfall calculation means 71 for determining the weight ratio reads the weight ratio from the weight ratio database 61, and also reads the rainfall data at a certain time in the past from the rainfall database 62 to calculate the basin average rainfall data. To do. The initial value of the weight ratio is an appropriate value (however, the sum is 1.0). The basin average rainfall data is as shown in Equation 2 above. As a result, the basin average rainfall data that captures rainfall at a plurality of locations is acquired. A plurality of the basin average rainfall data are acquired at different times.

Subsequently, the correlation coefficient calculating unit 72 reads the basin average rainfall data R _t from the weight ratio determining basin average rainfall calculating unit 71 and the past flow rate data from the flow rate database 63, respectively. This is a means for calculating a correlation coefficient for evaluating the weight ratio using data. Specifically, the correlation coefficient r is as shown in the above equation 3.

  The weight ratio evaluation means 73 determines whether or not the evaluation is a high weight ratio based on the correlation coefficient r calculated by the correlation coefficient calculation means 72. If the evaluation is high, the weight ratio is used as the weight ratio database. 61 is a means for registering in 61. First, the correlation coefficient r is substituted into the evaluation function E. The evaluation function E is an expression as shown in the following expression. Here, a and b represent the weight of each term of the evaluation function.

Here, a specific correlation coefficient r in the first term of the evaluation function E, also the value that is to enter outputs a specific prediction error e _i in the second term as an evaluation value, the evaluation value Based on evaluation. Consider separately into the first term and the second term.
First, consider the first term. The first term is close to a, i.e. r ² is 1 correlation coefficient when close to 1 (positive correlation) and -1 (negative correlation) to be one that indicates that close the basin average rainfall data and flow data High correlation and high evaluation. In addition, when the first term is close to 0, that is, when r ² is close to 0, this indicates that the correlation coefficient is close to 0 (no correlation), and the correlation between the basin average rainfall data and the discharge data is low and the evaluation is low. Acts to lower.
Next, the second term will be examined. Prediction error e _i is the prediction model be one indicating that little difference between the flow rate data Q of the predicted flow rate data Qf and the actual when the second term close to zero indicates a large value (e.g. a value of 1000 or 10000) It works so that the prediction accuracy is high and the evaluation is high. Moreover, so that the prediction accuracy of the prediction model A represents the often a difference between the flow rate data Q of the predicted flow rate data Qf and the actual lower rating is low when second term large prediction error e _i is close to 0 Act on.
That is, the larger the evaluation value of the evaluation function, the higher the evaluation, and the smaller the evaluation value, the lower the evaluation.
Note that only the first term has an effect on the evaluation by the weight ratio evaluation means 73 of the weight ratio determination means 70, and there is no influence on the second term. Then, it is determined whether or not the evaluation is the highest value, and when the evaluation is high, the weight ratio so far in the weight ratio database 61 is rewritten. Note that the weight ratio used as the initial value is registered at the first time, but thereafter, the weight ratio with the highest evaluation value is registered.

  The weight ratio optimization means 74 for determining the weight ratio is a means for separately registering in the weight ratio database 61 as a new weight ratio by changing the weight ratio with a high evaluation value based on a predetermined rule. Here, the weighting ratio can be changed by applying various optimization methods such as a mathematical programming method and a metaheuristic method as a predetermined rule for changing the weighting ratio.

Thereafter, using the weight ratio newly registered in the weight ratio database 61, the weight ratio determining basin average rainfall calculating means 71, the correlation coefficient calculating means 72, and the weight ratio evaluating means 73 are functioned and evaluated, and the previous evaluation is performed. If the evaluation is lower than before, the new weight ratio is discarded, and if the evaluation is higher than before, the new weight ratio is registered in the weight ratio database 61, and the weight for determining the weight ratio A ratio optimization unit 74 generates a weight ratio. Thereafter, the weight ratio determining basin average rainfall calculating means 71, the correlation coefficient calculating means 72, the weight ratio evaluating means 73, and the weight ratio determining weight ratio optimizing means 74 are repeatedly functioned, and finally the weight ratio w having a high evaluation is obtained. (1), ..., w (i), ..., w (n) are determined as weighting ratios. In addition, repetition is complete | finished when higher evaluation is not obtained over a predetermined period after obtaining high evaluation.
The average rainfall by such a weight ratio is determined so as to increase or decrease with the same tendency of the downstream flow rate according to the increase or decrease of the rainfall in each basin. As a result, a weight ratio with a high evaluation is determined in advance.

Subsequently, learning is performed by the prediction model construction means 80 ′.
First, the basin average rainfall calculating means 81 for constructing the prediction model reads the optimum weight ratio previously determined by the weight ratio determining means 70 from the weight ratio database 61, and also reads the rain data from the rain database 62, Basin average rainfall data is calculated.

Then, a flow prediction model is constructed using this basin average rainfall data.
The prediction model learning means 82 reads the basin average rainfall data from the prediction model construction basin average rainfall calculation means 81, reads out the basin average rainfall data and the flow rate data corresponding to the time from the flow rate database 63, and these flow rates A flow prediction model is constructed using data and basin average rainfall data. In this embodiment, a neural network is used as the flow rate prediction model.

The prediction model / weighted ratio evaluation means 84 is means for evaluating the learned flow rate prediction model. First, to calculate the prediction error e between the flow rate data Q of the predicted flow rate data Qf and actual case of using a watershed average rainfall data R _t. The prediction error e _i at each time i is calculated by Equation 9 using the past flow data for evaluation and the flow prediction data stored in the flow database 63. Then, an evaluation value is calculated by substituting these prediction errors e _i into the evaluation function E. The evaluation function E is calculated by the equation 11 shown above. It should be noted that in the evaluation by the prediction model construction means 80 ′, only the second term of Equation 11 is affected, and the first term is not affected. Then, it is determined whether or not the evaluation is high. If the evaluation is high, the coupling coefficient that is the learning result is rewritten and registered in the prediction model database 64. Also, the weight ratio is rewritten and registered in the weight ratio database 61.

  The weight ratio optimization means 85 for constructing the prediction model changes the weight ratio w (i) to a new weight ratio by applying various optimization methods such as mathematical programming and metaheuristic methods. The new weight ratio w (i) is additionally registered in the weight ratio database 61. The immediately preceding weight ratio w (i) is left in the weight ratio database 61.

  Thereafter, new basin average rainfall data is generated by the prediction model construction basin average rainfall calculation means 81, the flow prediction model is learned by the prediction model learning means 82, and evaluated by the prediction model evaluation means 83. If it is lower, the coupling coefficient and the weight ratio w (i) are discarded, and if it is higher, the coupling coefficient and the weight ratio w (i) are rewritten and registered. Then, a new weight ratio w (i) is determined by the prediction model construction weight ratio optimization means 85, and these prediction model construction basin average rainfall calculation means 81, prediction model learning means 82, prediction model / weight ratio evaluation. The means 84 and the weight ratio optimization means 85 for constructing the prediction model are repeatedly functioned to determine the coupling coefficient of the flow rate prediction model while determining the weight ratio w (i) having a high evaluation. As a result, a flow prediction model corresponding to the basin average rainfall data is constructed.

  Then, the prediction is performed by the prediction means 90. Also in this embodiment, since the weight ratio w (i) that reduces the output error of the flow rate prediction model is obtained, the optimum weight ratio w (i) can be obtained while improving the accuracy of the flow rate prediction model.

  Here, it has been described that the iterative process of repeatedly making the prediction model construction means 80 ′ function after the weight ratio determining means 70 is made to function repeatedly is performed once. However, such an iterative process is performed more than once. Then, the weight ratio may be determined and the flow rate prediction model may be learned so that the evaluation by the evaluation function E is maximized. Thereby, it is also possible to construct a flow rate prediction model with higher prediction accuracy.

  Next, prediction results obtained by performing actual prediction using the flow rate prediction apparatus of the present invention will be described in detail with reference to the drawings. FIG. 8 is an explanatory diagram for explaining rainfall data and flow data in another river model. Further, as the flow rate predicting device, the flow rate predicting device 1 (the device that adjusts the weighting ratio only by the weighting ratio determining means 70) illustrated in FIG.

Hereinafter, specific numerical values of the rainfall data and the flow data will be described. In this embodiment, as shown in FIG. 8, a simplified river model is considered in which tributaries 1 and 2 flow into the main stream at substantially the same location at the same time. In this example, a water storage facility is illustrated as the prediction target 100. Let Q be the flow rate flowing into the prediction target 100 at a certain time t. As upstream rainfall, there are two rain gauges, and the rainfall output from the rain gauge is R (1) and R (2), respectively. As will be described later, for the sake of simplicity, the flow time until the upstream rainfall R (1) appears in the downstream flow rate Q and the downstream rainfall R (2) until the downstream flow rate Q appears. The flow-down time has a relationship in which the time delay is the same because it flows in almost the same part of the river at approximately the same time. That is, a1 = a2 = a. In this example, in order to further simplify the description, it is assumed that there is no time delay and that a1 = a2 = 0. In this case, calculation formula of the basin average rainfall data R _t is expressed by the following equation.

Then, explanation will be made by applying specific numerical values. The table below shows rainfall R (1) and R (2) at points 1 and 2 and weighting ratio w (1) of 0.5 and w (2) at a certain time zone (1-10 o'clock). The basin average rainfall data _Rt is shown when the value is 0.5 (the rainfall unit in the table is mm / h).

In addition, examples of the basin average rainfall data _Rt and the flow data Q are shown below (the rainfall unit in the table is mm / h, and the flow unit is m ³ / s).

When the correlation coefficient r at this time is calculated according to Equation 3, the correlation coefficient r is 0.904. When the evaluation value by the evaluation function E of Equation 4 is calculated, it is 0.818.
Next, the weighting ratios w (1) and w (2) are optimized. As an example of optimization, for example, the evaluation value by the evaluation function E when w (1) and w (2) are changed in 0.1 steps is shown in the characteristic diagram representing the weight ratio-evaluation function in FIG. In the characteristic diagram of FIG. 9, the horizontal axis represents values of w (1) from 0 to 1.0. Since w (1) + w (2) = 1.0, if the value of w (1) is determined, the value of w (2) is also automatically determined, so w (1) If all the values are covered, the optimum weight ratio w (1) and w (2) can be selected. In this example, when w (1) is 0.2 and w (2) is 0.8, the evaluation value of the evaluation function E is maximum at E = 0.896, and the weighting ratio at this time is the optimum weighting ratio. As required.
From the above, in this case, the weight ratio determining means 70 determines w (1) as 0.2 and w (2) as 0.8.

Subsequently, the basin average rainfall calculation means 81 for predictive model construction of the prediction model construction means 80 calculates the basin average rainfall data R _t using Equation 12 using the weight ratio optimally determined by the weight ratio determination means 70, The prediction model learning unit 82 and the prediction model evaluation unit 83 are repeatedly functioned to learn the flow rate prediction model until the best flow rate prediction model is approached. An example in which a neural network is used as the learning flow rate prediction model will be described. When using a neural network for prediction, it is necessary to first learn the neural network. Although detailed explanation is omitted, the neural network is learned using the basin average rainfall data _Rt and the flow rate data Q for the past several hours calculated by the prediction model construction basin average rainfall calculation means 81. When the flow rate data Q is predicted, the basin average rainfall data R _t for the past several hours (different depending on the dam) at the time of prediction or the predicted rainfall is input to the neural network to obtain the predicted flow rate data Qf of the flow rate data Q. . The prediction is performed using the flow rate prediction model thus obtained. In this way, by using the basin average rainfall data _Rt, it is possible to reduce the input factor to the neural network, increase the learning speed, and obtain a good flow prediction model with improved prediction accuracy.

Another embodiment 2 will be described. In the first embodiment, the time delay was not taken into consideration, but in this second embodiment, the time delay occurs because the upstream and downstream flow into the river with a time interval. FIG. 10 is an explanatory diagram for explaining other rainfall data and flow data. Also in this example, a water storage facility is illustrated as the prediction target 100. At a certain point in time, the flow rate flowing into the prediction target 100 is Q, and there are two rain gauges as upstream rainfall, and the rainfall data output by the rain gauge is R (1) and R (2), respectively. . Specifically, the rainfall data R (1) is delayed by 1 hour, and the rainfall data R (2) is delayed by 2 hours. Here, the weight ratio determining basin average rainfall calculating means 71 of the weight ratio determining means 70 calculates the basin average rainfall data R _t at time _t by the following equation.

Equation 13 is obtained by setting n = 2, a1 = 1, and a2 = 2 in Equation 2. Further, w is a value from 0 to 1, and the weight ratios of the rainfall data R (1) and R (2) are w (1) and w (2), respectively.
Here the weighted ratio w (1), specific values of the basin average rainfall data R _t w (2) and hourly rainfall data R (1) and by R (2), for example, the table as shown in the following table Is done.

Hereinafter, the weight ratio determining means 70 functions in the same manner as in the first embodiment to perform an optimal flow rate prediction. In this way, the basin average rainfall data _Rt may be calculated.

The flow rate prediction apparatus of the present invention has been described above. The prediction of the flow rate of water flowing into dams and water storage facilities and the flow rate of rivers are directly linked to dam gate operation, and accurate prediction is required due to safety issues in the downstream area. In these predictions, the flow prediction accuracy varies greatly depending on how to consider rainfall over a wide area.
Moreover, when planning the installation of a rain gauge on the ground in the future, it is possible to select the most suitable rainfall point for the flow rate prediction using the mesh rainfall data provided from the weather company in advance. If a rain gauge installation point is determined based on this selection result, it is possible to devise an appropriate facility plan without incurring unnecessary costs.
The present invention can automatically determine which point of rainfall should be used for prediction. Moreover, before the construction of the prediction system, it is possible to select a rain point to be measured by calculating an optimal weight ratio that maximizes the correlation between the flow rate and the rainfall. If a prediction system has already been constructed or if simulation of prediction calculation is possible on a computer, it is possible to automatically calculate an optimal weight ratio that minimizes the prediction error. In either case, a highly accurate flow rate predicting means can be provided.

It is a block diagram of the flow volume prediction apparatus of the best form for implementing this invention. It is explanatory drawing explaining the rainfall data and flow volume data in a river model. It is a structural diagram of a weight ratio determination means. It is a structural diagram of a model construction means. It is a structural diagram of a prediction means. It is a block diagram of the flow volume prediction apparatus of another form. It is a structural diagram of another model construction means. It is explanatory drawing explaining the rainfall data and flow volume data in another river model. It is a characteristic view showing a weight ratio-evaluation function. It is explanatory drawing explaining the rainfall data and flow volume data in another river model.

Explanation of symbols

1, 1 ': Flow rate prediction device 10: Input unit 20: Output unit 30: Recording medium reading unit 40: Data transmission unit 50: Data collection unit 51: Rainfall data collection unit 52: Flow rate data collection unit 60: Data storage unit 61 : Weight ratio database 62: Rainfall database 63: Flow rate database 64: Prediction model database 65: Predictive value database 70: Weight ratio determination means 71: Basin average rainfall calculation means 72 for weight ratio determination 72: Correlation coefficient calculation means 73: Weight ratio Evaluation means 74: Weight ratio determining weight ratio optimization means 80, 80 ': Model construction means 81: Prediction model construction basin average rainfall calculation means 82: Prediction model learning means 83: Prediction model evaluation means 84: Prediction model / weighting Ratio evaluation means 85: weight ratio optimization means for predictive model construction 90: prediction means 91: basin average rainfall calculation means for prediction 2: predicting flow rate calculation unit 2: Network 100: candidate prediction

Claims (4)

A flow rate prediction device that predicts downstream flow rate using rainfall data at a plurality of upstream locations,
Using rainfall data at multiple points or sections upstream, and weight ratios that represent the degree of influence of rainfall data at a certain location on the downstream flow rate for each of the multiple rainfall data, A basin average rainfall calculating means for calculating a basin average weight for calculating a plurality of basin average rainfall data for each time point or a sum of a product of the weight ratio related to the rainfall data for a plurality of points or a plurality of sections;
Correlation coefficient calculating means for calculating a correlation coefficient by using a plurality of basin average rainfall data and flow rate data related to the basin average rainfall data by time,
A weight ratio evaluation means for determining whether the evaluation is a high weight ratio based on a correlation coefficient;
A weight ratio optimization means for determining a weight ratio that generates a new weight ratio by changing the weight ratio based on a predetermined rule;
The weight ratio determining basin average rainfall calculating means, the correlation coefficient calculating means, the weight ratio evaluating means, and the weight ratio determining weight ratio optimizing means are repeatedly functioned to correlate the basin average rainfall data with the discharge data. A weight ratio determining means for determining the weight ratio so as to increase,
Estimate the flow rate prediction model using the basin average rainfall data, which is the sum of the data of the rain data at a certain location and the weight ratio determined by the weight ratio determination means and the weight ratio determined for the multiple points or sections. A prediction model construction means to be constructed;
A prediction unit that inputs the basin average rainfall data based on the weight ratio determined by the weight ratio determination unit to generate the prediction flow data related to the predicted flow, with respect to the flow prediction model constructed by the prediction model construction unit;
A flow rate predicting device comprising: A flow rate prediction device that predicts downstream flow rate using rainfall data at a plurality of upstream locations,
Using rainfall data at multiple points or sections upstream, and weight ratios that represent the degree of influence of rainfall data at a certain location on the downstream flow rate for each of the multiple rainfall data, A basin average rainfall calculation means for predictive model construction for calculating a plurality of basin average rainfall data for each time point or a sum of the products of the weight ratio relating to this rainfall data for a plurality of points or a plurality of sections;
A prediction model learning means for learning a flow prediction model using a plurality of basin average rainfall data and flow rate data related to the basin average rainfall data according to time;
Predicted by the flow prediction model using the basin average rainfall data related to the past results and the corresponding flow rate data to generate predicted flow data, and evaluated using the prediction error between the actual flow data and the predicted flow data And a prediction model / weight ratio evaluation means for determining whether the evaluation is a high flow rate prediction model and a weight ratio,
A weight ratio optimization means for constructing a prediction model for generating a new weight ratio by changing the weight ratio based on a predetermined rule;
The prediction model construction basin average rainfall calculation means, the prediction model learning means, the prediction model / weight ratio evaluation means and the prediction model construction weight ratio optimization means are repeatedly functioned so that the prediction error is reduced. A prediction model construction means for determining a prediction model and a weight ratio;
Predictive means for generating predicted flow rate data related to the predicted flow rate by inputting the basin average rainfall data based on the weight ratio determined by the predictive model constructing unit to the flow rate predictive model constructed by the predictive model constructing unit;
A flow rate predicting device comprising: A flow rate prediction device that predicts downstream flow rate using rainfall data at a plurality of upstream locations,
Using rainfall data at multiple points or sections upstream, and weight ratios that represent the degree of influence of rainfall data at a certain location on the downstream flow rate for each of the multiple rainfall data, A basin average rainfall calculating means for calculating a basin average weight for calculating a plurality of basin average rainfall data for each time point or a sum of a product of the weight ratio related to the rainfall data for a plurality of points or a plurality of sections;
Correlation coefficient calculating means for calculating a correlation coefficient by using a plurality of basin average rainfall data and flow rate data related to the basin average rainfall data by time,
A weight ratio evaluation means for determining whether the evaluation is a high weight ratio based on a correlation coefficient;
A weight ratio optimization means for determining a weight ratio that generates a new weight ratio by changing the weight ratio based on a predetermined rule;
The weight ratio determining basin average rainfall calculating means, the correlation coefficient calculating means, the weight ratio evaluating means, and the weight ratio determining weight ratio optimizing means are repeatedly functioned to correlate the basin average rainfall data with the discharge data. A weight ratio determining means for determining the weight ratio so as to increase,
Multiple basin average rainfall data for each point in time is calculated by summing the product of the rainfall data at a certain location and the weight ratio determined by the weight ratio determination means for multiple points or multiple sections. Basin average rainfall calculation means for predictive model construction,
A prediction model learning means for learning a flow prediction model using a plurality of basin average rainfall data and flow rate data related to the basin average rainfall data according to time;
Predicted by the flow prediction model using the basin average rainfall data related to the past results and the corresponding flow rate data to generate predicted flow data, and evaluated using the prediction error between the actual flow data and the predicted flow data And a prediction model / weight ratio evaluation means for determining whether the evaluation is a high flow rate prediction model and a weight ratio,
A weight ratio optimization means for constructing a prediction model for generating a new weight ratio by changing the weight ratio based on a predetermined rule;
The prediction model construction basin average rainfall calculation means, the prediction model learning means, the prediction model / weight ratio evaluation means and the prediction model construction weight ratio optimization means are repeatedly functioned so that the prediction error is reduced. A prediction model construction means for determining a prediction model and a weight ratio;
Predictive means for generating predicted flow rate data related to the predicted flow rate by inputting the basin average rainfall data based on the weight ratio determined by the predictive model constructing unit to the flow rate predictive model constructed by the predictive model constructing unit;
A flow rate predicting device comprising: In the flow prediction apparatus according to any one of claims 1 to 3,
The flow rate prediction apparatus, wherein the flow rate prediction model is a neural network.

Images