JP2955413B2 - Rainwater inflow prediction device using neural network - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION The present invention relates to a rainwater drainage pumping facility or a pumping facility for treating stormwater drainage, which uses a radar rain gauge to measure the peak inflow of rainwater for supporting pump operation using a neural network. BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a rainwater inflow predicting device, and more particularly to a neural network applied rainwater inflow predicting device capable of automatically and accurately predicting a rainwater peak inflow (time change).

[0002]

2. Description of the Related Art In recent years, especially in urban areas, urbanization has progressed and sewerage systems have been rapidly developed, and most of rainfall has flowed into pump facilities. The need to respond to inflows has arisen.
In this case, in order to operate the pump facility optimally during rainfall, it is necessary to predict the amount of rainwater inflow from the start of rainfall to the start of actual inflow.

FIG. 7 shows an example of the relationship between the rainfall intensity and the amount of rainwater inflow. As shown in FIG. 7, the inflow phenomenon has a time delay with respect to rainfall. From the standpoint of pump operation, there is a demand to know the temporal change of rainfall and to predict the peak inflow of rainwater (corresponding to the maximum flow rate / mark in the figure). If the rainwater peak inflow can be predicted in advance, it becomes possible to grasp the start timing of the pump operation necessary for avoiding inundation, and to prepare to start up the required number of private generators.

Heretofore, several models have been proposed as models for predicting the inflow of rainfall runoff, such as the unit map method. However, accurate prediction models for various types of rainfall have not yet been established. However, there is a limit to the development of mathematical models. The causes are as follows. (A) The mathematical model contains many unknown parameters, and a lot of experimental data and analysis adjustment work are required to determine them, which requires considerable manpower and time. (B) If the structure of the mathematical formula model itself cannot flexibly cope with various rainfalls, a separate model configuration is required according to the rainfall.

(C) The amount of runoff into a sewer due to rainfall is determined by the effective rainfall remaining after subtracting the rainfall loss due to ground penetration or the like in the basin rainfall. These are heavily dependent on the ratio of the impervious area and the infiltrated area of the basin, that is, land use, and land use is changing over time with urbanization. The mathematical model once established cannot flexibly cope with such aging.

Conventional rainfall observation is based on a surface rainfall gauge, which is a point rainfall. Assuming that the distribution of rainfall in a basin is uniform, the rainfall is predicted by a mathematical model using a surface rainfall gauge at a pump facility. . For this reason, when the basin is wide to some extent and there is local rainfall, the calculation error becomes large, and the problem of using the point rainfall is pointed out.

[0007] On the other hand, recently, a radar rain gauge has been introduced and installed, and it has become possible to grasp the state of rainfall distribution in detail in real time. As for rainfall data used for peak inflow prediction, there is a strong demand for a method for effectively utilizing area rainfall data from a radar rain gauge to improve the accuracy of inflow prediction.

[0008]

As described above, the conventional method for predicting the amount of inflow of rainwater using a mathematical model has problems in handling rainfall data as input data and setting (tuning) model parameters. Was.

An object of the present invention is to provide a highly reliable neural network-based rainwater inflow prediction which can automatically and accurately predict the rainwater peak inflow (time change) for pump operation support. It is to provide a device.

[0010]

SUMMARY OF THE INVENTION In order to achieve the above object, the present invention provides a method for pumping a pumping facility such as a sewerage pumping station or a treatment plant for rainwater drainage by a neural network using a radar rain gauge. in rainwater inflow prediction apparatus for predicting the rainwater inflow for driving support, rainfall strong mesh shape observations from radar rain gauges
It is a value obtained by integrating the rainfall intensity into
Learning to store discrete accumulated rainfall data at equal time intervals as input data for learning, calculate the rainwater inflow into the pumping facility from the water level gauge of the pumping facility and the discharge rate from the pump operation, and store it as learning output data. and use record information input means, based on the rainwater inflow is accumulated rainfall data and Science <br/>習用output data as input data for learning stored by the learning record information input means, a back propagation method The weighting factor of the neural network is learned by using the prediction model construction means for storing the learned weighting factor, and the learned weighting factor stored in the prediction model construction means, which is obtained from the observation value of the radar rain gauge. cumulative rainfall data or these rainfall the day, and the inflow amount prediction means for predicting the rain water inflow by the neural network, the inflow amount prediction It is configured with an output means for outputting the rainwater inflow amount predicted by stage.

In order to achieve the above object, according to the present invention, a pump operation support for a pump facility such as a sewerage pumping station or a treatment plant for rainwater drainage is provided by a neural network using a radar rain gauge. The rainwater inflow forecasting device, which predicts the rainwater inflow , converts the observation values from the radar rain gauge into rainfall intensity in mesh form.
Is the value obtained by integrating the rainfall intensity of
Discrete and cumulative rainfall data and observations on pre-rainfall
Pre-rainfall results obtained based on the values and a certain discriminant
Discriminates whether rainfall has had an effect on rainwater inflow based on criteria
And the discrimination value, which is the calculated value, is stored as learning input data, and the amount of rainwater flowing into the pump facility is calculated from the water level meter of the pump facility and the discharge amount due to the pump operation, and the learning result information is stored as learning output data. The weight of the neural network is calculated by the back propagation method based on the input means, the accumulated rainfall data and the discriminant value as the learning input data stored by the learning result information input means, and the rainwater inflow as the learning output data. A prediction model construction means for learning the coefficient and storing the learned weight coefficient, and an integrated rainfall on the rainy day obtained from the observation value of the radar rain gauge using the learned weight coefficient stored in the prediction model construction means. From the data and the corresponding discrimination value of the previous rainfall, the amount of rainwater inflow was calculated using a neural network. An inflow amount prediction means for measuring, are configured to include an output means for outputting the rainwater inflow amount predicted by inflow prediction means.

Further, as the prediction model constructing means, on the basis of the rainwater inflow is accumulated rainfall data, and learning the output data is a learning input data stored by the learning record information input means, backpropagation The neural network learning means learns the weight coefficient of the neural network by the method, and the learned weight coefficient storage means stores the weight coefficient learned by the neural network learning means.

Further, if necessary, a learning starting means for correcting the weight coefficient of the model due to aging by performing learning in the prediction model construction means at regular intervals is added.

[0014]

Therefore, in the rainwater inflow predicting apparatus applied to the neural network according to the present invention, the observation value from the radar rain gauge is converted into a mesh-shaped integrated rainfall, and discrete integrated rainfall data is obtained at equal time intervals from the start of rainfall. In addition to being stored as learning input data, the amount of rainwater flowing into the pump facility is calculated from the water level gauge of the pump facility and the discharge amount due to the pump operation, and the amount of rainwater flowing is stored as learning output data.

Next, the weighting factor of the neural network is learned and stored by the back propagation method based on the accumulated rainfall data as the learning input data and the rainwater inflow as the learning output data. . Then, from the accumulated rainfall data, which is the actual distribution of rainfall on the day of the rainfall, and the discrimination value, which is the previous rainfall actual, the rainwater inflow is predicted by the neural network using the stored learned weighting factors. The rainwater inflow is output.

[0016]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS In the present invention, when predicting rainwater peak inflow for pump operation support, an observation value from a radar rain gauge is taken in to obtain a rainfall distribution result and a previous rainfall result.
By learning the causal relationship with the actual inflow, the rainwater inflow is predicted. Hereinafter, an embodiment of the present invention based on the above concept will be described in detail with reference to the drawings. FIG. 1 is a block diagram showing an overall configuration example in a case where the neural network application inflow rate prediction device according to the present invention is applied to a pump facility of a sewer pumping station.

In FIG. 1, a radar rain gauge 1 for observing rainfall and outputting an observed value (electric energy) is provided in a sewer basin.
Is installed. The pump station is provided with a water level meter 2 that measures the water level and outputs a measured value, and a flow meter 3 that measures the amount of discharge from the pump station by pump operation and outputs the measured value.

On the other hand, the neural network applied inflow rate prediction device 4 includes a rainfall distribution result calculation device 41, a previous rainfall result determination device 42, a learning input data storage device 43,
Rainwater inflow calculating device 44, learning output data storage device 45, neural network learning device 46, learned weight coefficient storage device 47, neural network prediction device 48, and rainwater inflow display device 49 as output means.
And a learning activation switch 40.

It should be noted that the rainfall distribution actual calculation device 41, the previous rainfall actual determination device 42, the learning input data storage device 43, the rainwater inflow amount calculation device 44, and the learning output data storage device 45.
From the learning network, the neural network learning device 46, the learned weight coefficient storage device 4
7 constitutes a prediction model construction means.

Here, the rainfall distribution result calculation device 41 converts the observation value from the radar rain gauge 1 into an integrated rainfall in a mesh shape. In addition, the previous rainfall performance determination device 42
Is to determine whether or not there is any influence on the inflow amount of the previous rainfall based on a certain determination criterion. Further, the learning input data storage device 43 compares the discrete accumulated rainfall data at equal time intervals from the start of rainfall calculated by the rainfall distribution performance calculation device 41 with the determination value determined by the previous rainfall performance determination device 42. , As input data for learning.

On the other hand, the rainwater inflow calculating device 44 calculates the rainwater inflow flowing into the pump facility based on the measured values from the water level meter 2 and the flow meter 3. Further, the learning output data storage device 45 is for storing the rainwater inflow amount calculated by the rainwater inflow amount calculation device 44 as learning output data.

On the other hand, the neural network learning device 4
6 is based on the accumulated rainfall data and the discrimination value, which are the learning input data, and the rainwater inflow amount, which is the learning output data, stored by the learning input data storage device 43 and the learning output data storage device 45. The weight coefficient of the neural network is learned by the back propagation method. The learned weight coefficient storage device 47
Is for storing the weighting coefficient learned by the neural network learning device 46. Further, the neural network prediction device 48 uses the learned weight coefficient stored in the learned weight coefficient storage device 47 to
The rainwater inflow is predicted by a neural network from the integrated rainfall data, which is the actual rainfall distribution on the day of rainfall, and the discrimination value, which is the previous rainfall actual, input by the rainfall distribution actual operation device 41. Furthermore, the rainwater inflow amount display device 49 is for displaying and outputting the rainwater inflow amount predicted by the neural network prediction device 47.

The learning activation switch 40 is a command switch for performing the learning in the neural network learning device 46 at regular intervals to correct (review) the weight coefficient of the model due to aging. Next, the operation of the neural network application inflow rate prediction device 4 of the present embodiment configured as described above will be described.

In FIG. 1, an observation value (electric power amount) from the radar rain gauge 1 is converted into a mesh-shaped integrated rain amount by the rainfall distribution result calculation device 41. Then, discrete integrated rainfall data at equal time intervals from the start of rainfall is stored in the learning input data storage device 43 as learning input data.

That is, in this case, there are two methods of expressing the amount of rainfall: rainfall intensity and integrated rainfall. The rainfall intensity is the amount of rainfall per hour (mm / h), and the time integration of this is the integrated rainfall (mm). FIG. 2 shows the relationship between the rainfall intensity and the accumulated rainfall.

Although the rainfall observed by the radar rain gauge 1 is the rainfall intensity, the rainfall intensity is converted into an integrated rainfall by integrating it over time. It can be understood that it is easier to handle with the accumulated rainfall.

For example, a radar rainfall observation mesh is considered for the inflow prediction target basin as shown in FIG. The rain that falls in the basin flows out to the sewer main line,
in flows in. In this case, of the observation mesh, the integrated rainfall Ri (i = 1 to n) affects the basin. Therefore, the integrated rainfall data of the rain falling on the mesh i at the time t is Ri
(T).

The previous rainfall performance determination device 42 determines whether there is any influence on the inflow amount of the previous rainfall based on a predetermined determination criterion. The determined value is also stored in the learning input data storage device 43 as learning input data.

That is, if the rainfall occurred the day before or several hours ago, and the same rainfall occurred on the day,
It is known that the inflow is larger than when there is no previous rainfall record. This is caused by the limited capacity of rainwater to penetrate and retain water on the land (called percolation capacity). For rainfall exceeding this infiltration capacity, since the land is saturated, it does not penetrate the ground completely and flows into the public sewer as surface flow.

Therefore, as a factor affecting the inflow, it is necessary to incorporate the previous rainfall results into the learning in addition to the integrated rainfall. This is determined based on whether or not there has been an accumulated rainfall of Amm or more within T hours. The input data is “1” when there is a previous rainfall record, and “0” when there is no previous rainfall record.

That is, in this case, the previous rainfall record has only one data as an average value of the entire basin. In other words, if an integrated rainfall of Amm or more is found within T hours for each mesh, “1” is obtained. If not, set it to “0”,
An average value is obtained by dividing the sum of the watershed meshes by the number of meshes. If the average value is larger than 0.5, it is set to "1", and if smaller than that, it is set to "0".

On the other hand, the amount of rainwater flowing into the pump facility is calculated by the rainwater inflow calculating device 44 from the measured values from the water level meter 2 and the flow meter 3. Then, the rainwater inflow amount data is stored in the learning output data storage device 45 as learning output data. That is, in this case, the amount of inflow to the pumping station is usually calculated by a method because there is no measuring equipment. That is, the inflow is

[0033]

(Equation 1) It is given by the following expression. This will be described with reference to FIG.

That is, the past water level change during rainfall can be read from the water level meter record. Then, the change in water level is applied to the VH characteristic to calculate a flow rate change. On the other hand, the pump discharge amount is calculated by applying the discharge amount-water level relationship called QH characteristic from the operation record of the pump. Next, the amount of rainwater flowing into the pumping station is obtained by adding the amount of change in the flow rate and the pump discharge amount.

Next, the neural network learning device 4
At 6, the integrated rainfall data (rainfall distribution actual value) and the discriminant value (previously rainfall actual), which are the learning input data, stored by the learning input data storage device 43 and the learning output data storage device 45, The weighting factor is calculated by the back propagation method using the inflow amount actual value that is the output data for use. FIG. 5 is a diagram illustrating a configuration example of a neural network. In this case, it has a three-layer configuration of an input layer, an intermediate layer, and an output layer.

That is, prediction is performed by learning a weight coefficient between neurons in a neural network in which the previous rainfall result of the rainfall distribution result value is used as a neuron in the input layer, and the actual inflow amount of the teaching data is used as a neuron in the output layer. A model is obtained. In this case, a learning method called backpropagation, in which network errors are carried back from the output layer to the input layer, is used for learning. Hereinafter, the learning procedure will be described.

(Step 1): The integrated rainfall data (rainfall distribution actual value) and the discrimination value (previously rainfall actual) are input to the input layer, and the intermediate layer and the output layer are calculated according to the following neuron model. The output H _j of the jth neuron in the hidden layer is

[0038]

(Equation 2) The output O _k of the k-th neuron in the output layer is

[0039]

(Equation 3)

(Step 2): The sum of the square errors between the output O _k of the k-th neuron in the output layer and the actual inflow amount y _k which is the teaching signal of the k-th neuron in the output layer is minimized. It is learned by modifying the weighting factor of the network. Learning the weight coefficient between the hidden layer and the output layer

[0041]

(Equation 4) Δw _jk of is calculated, w _jk is modified. In addition, learning of the weight coefficient between the input layer and the hidden layer is as follows.

[0042]

(Equation 5) Of Δw _ik is calculated, w _ij is modified. The weighting factor obtained as a result of the learning as described above is stored in the learned weighting factor storage device 47.

Next, the neural network prediction device 4
At 8, the rain distribution result (integrated rainfall data) from the rain distribution result calculation device 41 on the day of the rain and the previous rain performance determination device 42
From the previous rainfall result (discrimination value), the rainwater inflow amount is predicted using the learned weight coefficient stored in the learned weight coefficient storage device 47.

That is, when the weight coefficients W _ij and W _{jk of} the neural network are determined, the data is given to the input layer, and the inflow of rainwater, which is the data of the output layer, is obtained. Specifically, time series data of the integrated rainfall of the radar rainfall observation mesh is given as input data. In this case, as a method of giving the time-series data, m integrated rainfalls are prepared by dividing the integrated rainfall at regular time intervals ΔT. Then, there are n sets of such data corresponding to the number of meshes.
Also, previous rainfall actual data is input. As shown in FIG. 6, the rainwater inflow amount predicted from the integrated rainfall is equal to a predetermined time Δ
The value is output to the output layer as a value for each T.

Finally, the predicted value of the amount of rainwater inflow from the neural network prediction device 48 thus predicted is displayed on the rainwater inflow display device 49, and a series of prediction calculations ends.

In the above description, if the rainfall until the present time is low at the early stage of the rainfall, it is not possible to obtain sufficient rainwater inflow prediction information for rainwater drainage. For this reason, it is necessary to forecast rainfall for several hours in advance if necessary. Since this is related to the accuracy of rainfall prediction, it can be said that in the case of the present invention, it is more practical that the watershed is wide and the time delay from the start of rainfall to the inflow of pump station rainwater is as large as possible. That is, the present invention is applicable to a case where rainfall distribution is observed until a time when the rainfall reaches the rainfall peak, and after a sufficient time has passed, rainwater flows into the pumping station.

As described above, the neural network-based inflow rate prediction device 4 of this embodiment includes a rainfall distribution result calculation device 41 that converts the observation value from the radar rain gauge 1 into a mesh-shaped integrated rainfall, Pre-rainfall performance discriminating device 42 for discriminating whether or not there is an influence on the inflow amount of pre-rainfall based on the discrimination criterion, and discrete integrated rainfall data at equal time intervals from the start of rainfall calculated by rainfall distribution result calculating device 41 , A discriminating value determined by the previous rainfall actual discriminating device 42, a learning input data storage device 43 for storing as learning input data, and a pump facility based on measured values from the water level meter 2 and the flow meter 3. Rainwater inflow calculator 44 for calculating the amount of rainwater inflow into the rainwater, and rainwater inflow calculator 44
Learning data storage device 45 for storing the rainwater inflow amount calculated by the above as learning output data, and learning input data storage device 43 and learning input data stored by the learning output data storage device 45. A neural network learning device 46 that learns the weighting factor of the neural network by the back propagation method based on the integrated rainfall data and the discrimination value that is the learning data, and the rainwater inflow amount that is the learning output data, and learning by the neural network learning device 46 Using the learned weight coefficient storage device 47 for storing the learned weight coefficient, and the learned weight coefficient stored in the learned weight coefficient storage device 47,
A neural network prediction device 48 for predicting rainwater inflow by a neural network from the integrated rainfall data which is the rainfall distribution result on the rainy day and the discrimination value which is the previous rainfall result input by the rainfall distribution result calculation device 41;
A rainwater inflow display device 49 for displaying and outputting the rainwater inflow predicted by the neural network prediction device 47;
The neural network learning device 46 comprises a learning activation switch 40 for performing learning at regular intervals and correcting (reviewing) the weight coefficient of the model due to aging. Therefore, the following effects can be obtained.

(A) Since the weighting factor of the prediction model is learned based on the relationship between the past rainfall actual value and the rainwater inflow amount actual value, it is only necessary to accumulate the data necessary for the learning, and to set the model. Work can be automated.
That is, it is possible to eliminate the need for manpower and time for data analysis and the like that are required for adjusting the mathematical model, as in the related art.

(B) By performing learning at regular intervals, the influence of land use over time on the inflow rate change
It is possible to reflect on the weight coefficient based on the actual value. (C) The rainfall distribution by the radar rain gauge, which is the area rainfall, is taken in instead of the ground rain gauge, which is a point rainfall, which has been a problem in the past. Therefore, it is possible to significantly improve the prediction accuracy of rainwater inflow. Becomes

In the above embodiment, a case has been described in which a display device for displaying and outputting the rainwater inflow amount predicted by the neural network prediction device 48 is provided as the output means. However, the present invention is not limited to this. It goes without saying that a printing device that prints out the rainwater inflow amount predicted by the network prediction device 48 may be provided.

Further, in the above-described embodiment, there is provided the pre-rainfall result discriminating device 42 for discriminating whether or not the influence of the pre-rainfall on the inflow amount is based on a predetermined discrimination criterion.
The determination value determined in step 2 is used as the rainfall distribution result calculation device 4
Although the case has been described in which the learning input data storage device 43 is stored as learning input data together with discrete integrated rainfall data at equal time intervals from the start of rainfall calculated by 1 in the learning input data storage device 43, However, it is not necessarily essential to the present invention.

Further, in the above-described embodiment, the case where the learning start switch 40 for correcting (reviewing) the weight coefficient of the model due to aging by performing the learning in the neural network learning device 46 at regular intervals has been described. The provision of the learning start switch 40 is not always essential to the present invention.

[0053]

As described above, according to the present invention, pump operation support for a pumping facility such as a sewerage pumping station or a treatment plant for draining rainwater by a neural network using a radar rain gauge is provided. In the rainwater inflow prediction device that predicts rainwater inflow, the observation value from the radar rain gauge is converted into a mesh-shaped rainfall intensity, and this rainfall intensity is calculated.
It is a value obtained by integrating the degrees and discrete at equal time intervals from the start of rainfall
Means for storing actual cumulative rainfall data as learning input data, calculating the rainwater inflow amount flowing into the pump facility from the water level gauge of the pump facility and the discharge amount due to the pump operation, and storing it as learning output data. , based on the rainwater inflow is accumulated rainfall data and Science習用output data as input data for learning stored by the learning record information input means, to learn the weighting factor of the neural network by the back propagation method, and prediction model constructing means for storing the weighting coefficient of the learned, using the weight coefficient learned stored in the prediction model constructing means the day rainfall derived from observations of the radar rain gauge integrated rainfall data or al, Inflow prediction means for predicting rainwater inflow by neural network, and rainwater predicted by inflow prediction means The output means for outputting the input amount is provided, so that it is possible to automatically and accurately predict the rainwater peak inflow amount (time change) for supporting the pump operation, which is extremely reliable. A neural network application rainwater inflow prediction device with high cost can be provided.

[Brief description of the drawings]

FIG. 1 is a block diagram showing an embodiment in which the neural network applied inflow rate prediction device according to the present invention is applied to a pump facility of a sewer pumping station.

FIG. 2 is a diagram for explaining integrated rainfall data Ri (t) in the embodiment.

FIG. 3 shows the same embodiment.

FIG. 4 is a view for explaining a method of calculating the amount of rainwater flowing into the pumping station in the embodiment.

FIG. 5 is a diagram showing a configuration example of a neural network in the embodiment.

FIG. 6 is an exemplary view for explaining an example of the relationship between input data and output data in the embodiment.

FIG. 7 is a diagram showing an example of the relationship between rainfall intensity and rainwater inflow.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Radar rain gauge, 2 ... Water level gauge, 3 ... Flow meter, 4 ... Neural network applied inflow forecasting device, 41 ... Rainfall distribution result calculating device, 42 ... Pre-rainfall result discriminating device, 43 ... Input data storage device for learning , 44 ... rainwater inflow calculation device,
45: learning output data storage device, 46: neural network learning device, 47: learned weight coefficient storage device
48: Neural network prediction device, 49: Rainwater inflow amount display device, 40: Learning start switch.

──────────────────────────────────────────────────続 き Continued on the front page (58) Fields investigated (Int. Cl. ⁶ , DB name) G05B 13/00-13/04 G06F 15/18 550 G06F 15/20 E03F 1/00 F04B 49/00-49 / 08

Claims (4)

(57) [Claims] 1. A rainwater inflow prediction device for predicting rainwater inflow for a pump operation support by a neural network using a radar rain gauge for a pump facility such as a sewer pumping station or a treatment plant for rainwater drainage. In the above, the observation value from the radar rain gauge is used as a mesh-shaped rainfall intensity.
It is a value obtained by integrating the rainfall intensity into
While storing discrete accumulated rainfall data as learning input data at equal time intervals from the same, the rainwater inflow into the pump facility is calculated from the water level gauge of the pump facility and the discharge amount due to the pump operation as learning output data. a learning record information input means for storing, on the basis of the rainwater inflow is beauty science習用output data Oyo integrated rainfall data as input data for learning stored by the learning record information input means, the back propagation method Learn the weight coefficient of the neural network,
And prediction model constructing means for storing the weighting coefficient of the learned, using said weighting coefficients of the predictive model stored in the construction unit already learned, or cumulative rainfall data of day rainfall derived from observations of the radar rain gauge A rainwater inflow predicting means for predicting rainwater inflow by a neural network; and an output means for outputting a rainwater inflow predicted by the inflow prediction means. Quantity prediction device. 2. A rainwater inflow prediction device for predicting rainwater inflow for pump operation support by a neural network using a radar rain gauge for a pump facility such as a sewer pumping station or a treatment plant for rainwater drainage. In the above, the observation value from the radar rain gauge is used as a mesh-shaped rainfall intensity.
It is a value obtained by integrating the rainfall intensity into
From the integrated accumulated rainfall data at equal time intervals
Pre-rainfall results obtained based on the above observations
The rainfall inflow of pre-rainfall based on certain criteria
A discriminant value, which is a value for discriminating the presence or absence of the influence, is stored as learning input data, and a rainwater inflow into the pump facility is calculated from a water level gauge of the pump facility and a discharge amount by pump operation, and a learning output is calculated. Learning result information input means to be stored as data; integrated rainfall data and discrimination values as learning input data stored by the learning result information input means; and rainwater inflow as learning output data. A prediction model construction unit that learns the weighting factor of the neural network by the propagation method, and stores the learned weighting factor.Using the learned weighting factor stored in the prediction model building unit, the radar rain gauge Based on the accumulated rainfall data on the day of rainfall obtained from the observed values and the corresponding discrimination value of the previous rainfall, Neural network-applied rainwater inflow, characterized by comprising: an inflow amount prediction means for predicting rainwater inflow by a neural network; and an output means for outputting the rainwater inflow predicted by the inflow amount prediction means. Prediction device. The method according to claim 3, wherein the prediction model constructing means on the basis of the learning result information inputted as input data for learning stored by means accumulated rainfall data, and rainwater inflow is learning output data, back 2. A neural network learning means for learning a weight coefficient of a neural network by a propagation method, and a learned weight coefficient storage means for storing a weight coefficient learned by the neural network learning means. Rainwater Inflow Prediction Applied to Neural Network. 4. The neural network applied rainwater inflow predicting apparatus according to claim 1, wherein learning in said predictive model construction means is performed at regular intervals to correct learning weight coefficients of the model due to aging. A rainwater inflow predicting device using a neural network, characterized by adding means.