JPH06110555A - Flow-in water quantity predicting device - Google Patents

Abstract

PURPOSE:To prevent processing from being made into a black box by clearly expressing the intuition of an operator and to facilitate the tuning of a membership function. CONSTITUTION:This device is provided with a weather information storage means 31 and result flow-in water quantity storage means 32 for storing inputted weather information and the result value of flow-in water quantity, fuzzy rule storage means 34 for preserving relation between the weather information and the flow-in water quantity as a fuzzy rule, predicting means 33 to calculate predictive flow-in water quantity based on the input information by using the fuzzy rule, and output device 4 to show the predicted result. At such a flow-in water quantity predicting device, a tuning means 36 depending on a neural network is added for tuning the membership function by the learning function of the neural network.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an inflow water amount predicting device for predicting an inflow amount of water into a reservoir or a river from weather information using an electronic computer.

[0002]

2. Description of the Related Art Conventional methods for predicting inflow to rivers and reservoirs represent the meteorological information used to predict the inflow of water by a membership function, and calculate the inflow of water to the reservoir by fuzzy operation (eg, 1991 power technology). Research paper PE-9
1-5) Alternatively, there is also a method in which a neural network is applied to the prediction, and the weather information and the inflow water amount, which are past actual values, are given as teacher data of the neural network to make the neural network learn and predict the inflow water amount ( As an example of using a neural network for prediction, there is a paper of the 91st Power Technology Study Group PE-91-12, PE.
-91-13, PE-91-14, PE-91-1
5).

[0003]

When a fuzzy operation is used to calculate a model that changes over time, such as when predicting the inflow of a reservoir, it is necessary to retune the membership function according to the change over time of the model. Moreover, this kind of tuning work has no deciding factor and requires trial and error, which is a heavy burden on the operator. Further, the prediction of the inflow of water using the neural network leads to a black box treatment, and there is a drawback that the validity of the prediction result of the neural network cannot be guaranteed under the meteorological condition that has not been experienced in the past. The present invention has been made in view of the above circumstances, and predicts the inflow of water by facilitating the tuning of the membership function while avoiding the black boxing of the processing by expressing the intuition of the operator intelligibly by using the fuzzy operation. The purpose is to provide a device.

[0004]

In order to achieve the above object, the present invention is a means for storing the input meteorological information and the actual value of the inflow water amount, and a means for storing the relationship between the meteorological information and the inflow water amount as a fuzzy rule. And a means for calculating the predicted inflow water amount based on the input information using the fuzzy rule, and an inflow water amount prediction device having an output device for presenting the prediction result, by giving the past actual inflow water amount and meteorological information to the neural network. A means for tuning the membership function of fuzzy rules is added by the learning function of the network.

First, the meteorological information and the actual value of the inflow water amount are input to the electronic computer via the input device and stored. The forecast of the inflow of water is performed automatically at regular intervals or according to the operator's request.
The inflow water amount is calculated by using the meteorological information stored in the electronic computer and the fuzzy rule stored in the electronic computer which is defined in advance, and the result is presented to the output device. As a result of the comparison display, if the predicted result of the inflow water amount deviates, or if a factor that changes the inflow water amount of the reservoir or river occurs, the operator activates the membership function tuning means. The activated tuning means gives the inflow water amount and the meteorological information, which are past performance values, as learning data of the neural network, and tunes the membership function.

[0005]

Embodiments will be described below with reference to the drawings. Figure 1
FIG. 1 is a configuration diagram of an embodiment of an inflow water amount prediction device according to the present invention. In the figure, 1 is a transmission line for transmitting weather information and information about the amount of inflow water, 2 is an input device, 3 is an electronic computer, and 4 is an output device. The electronic computer includes a weather information storage unit 31, a record inflow water amount storage unit 32 that stores a record inflow amount, weather information stored in the weather information storage unit 31, fuzzy rules and membership defined in advance in the fuzzy rule storage unit 34. It is stored in the prediction means 33 for performing the fuzzy calculation using the function and to calculate the predicted inflow water quantity, the predicted inflow water quantity storage means 35 for storing the predicted inflow water quantity, the weather information storage means 31, and the actual inflow water quantity storage means 32. The neural network tuning means 36 performs tuning of the membership function stored in the fuzzy rule storage means 34 using the meteorological information and the actual value of the inflow water amount.

Next, the operation will be described. The weather information and the inflow water information are transmitted through the transmission line 1, and the input device 2
It is stored in the weather information storage means 31 and the actual inflow water storage means 32 of the electronic computer 3 via. The operation of the predicting means 33 will be described using an example of the fuzzy rule of FIG. 2 and an example of the membership function of FIGS. 3 and 4. In FIG.
The rule of "Winter 1" means that if the temperature is low and the temperature difference from the normal year is negative, the predicted inflow is slightly less than the normal year. An example of a membership function that indicates that the temperature is low, the temperature difference from the normal year is negative (minus), and the amount of inflow water is slightly smaller than the normal year is as shown in FIG. The fact that the amount of inflow is slightly smaller than that of the normal year corresponds to the decrease 3 of the membership function of the predicted inflow difference shown in FIG.
The processing flow of the prediction means will be described with reference to the flowchart of FIG. In step S1, first, the difference between the predicted precipitation amount, the early snow depth (calculated in units of 10 days), and the temperature in the meteorological information storage means 31 from the past average value is calculated. In step S2, the predicted precipitation amount, the amount of snowfall in early season, the difference between the normal temperature value and the temperature calculated in step S1 are used as the input information of the antecedent if part of each fuzzy rule, and the predicted inflow of the consequent part then is calculated. The difference from the normal value of is calculated using, for example, the max-min composition method. This is calculated for all fuzzy rules, and the consequent part is defuzzified using the center of gravity method, etc., and the difference from the normal value for the predicted inflow is calculated. At step S3, the normal value is added to the difference from the normal value obtained at step S2 to calculate the predicted inflow water amount. Output device 4 with calculated predicted inflow
To present. In the presentation to the output device 4, the predicted value and the actual value are compared and output.

Next, the tuning means 36 by the neural network when the predicted value and the actual value do not match
The flow of the process will be described according to the flow of FIG. First,
In step Sn1, the fuzzy rules and membership functions stored in the fuzzy rule storage means are expanded into a neural network. When developing into a network, the antecedent part is approximated by a sigmoid type for the trapezoid type and the bell type is approximated for a triangular type, and the consequent part can use the trapezoidal membership function. Figure 7 shows the approximate antecedent membership function.
Shown in. FIG. 8 shows an expansion of the rule of FIG. 2 in the network. In the network shown in FIG. 8, the output is defuzzified from the middle third layer to the output layer. In FIG. 8, black circles are threshold units, and shaded circles are trapezoidal membership function corresponding units of the consequent part. In step Sn2, the difference between the average influent water amount, the temperature, the rainfall amount, and the snow depth data stored in the meteorological information storage means and the actual inflow water amount storage means is calculated from the normal value, and the neural network in FIG. The difference with the value, temperature, the difference with the normal value of the temperature, the difference with the normal value of the precipitation, and the difference with the normal value of the snow depth are given as the teaching data of learning, and it learns by the back-propagation method. In step Sn3, the membership function whose position has been changed in step Sn2 is converted into a trapezoid for the sigmoid type and a triangle type for the bell type. According to the above-mentioned embodiment, the inflow water amount is predicted by the fuzzy calculation using the fuzzy rule which is easy for the operator to perform, and the complicated membership function tuning is tried by the operator by using the learning function of the neural network. It can be done automatically without repeating mistakes.

The following embodiments are also possible. Instead of transmitting the weather information and the inflow water amount through the transmission path and inputting the input device, the operator may input the weather information and the inflow water amount using an interactive input / output device. Further, the predicting means and the tuning means by the neural network can be performed by different electronic computers.

[0009]

As described above, according to the present invention, the inflow water amount of a reservoir can be predicted by fuzzy arithmetic processing using a fuzzy rule which is easy for the operator to understand, so that the inflow water amount is safe for the operator. It is possible to provide the prediction device of. Moreover, since the tuning of the membership function which is troublesome for the operator is performed by the automatic tuning function of the inflow water amount predicting device, the operator is freed from the tuning work,
It is possible to obtain highly accurate prediction results that match the reservoir and river models that have changed over time.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an embodiment of an inflow water amount prediction device according to the present invention.

FIG. 2 shows an example of fuzzy rules.

[Fig. 3] An example of a membership function of predicted precipitation difference, early snow depth difference, temperature, and temperature difference.

FIG. 4 is an example of a membership function of a predicted inflow water amount difference.

FIG. 5 is a processing flow of a prediction unit.

FIG. 6 is a processing flow of a tuning means using a neural network.

FIG. 7 shows the membership function converted into a sigmoid function and a bell-shaped function.

FIG. 8 shows fuzzy rules deployed on a network.

[Explanation of symbols]

 1 Transmission Line 2 Input Device 3 Electronic Computer 4 Output Device 31 Meteorological Information Storage Means 32 Actual Inflow Water Storage Means 33 Prediction Means 34 Fuzzy Rule Storage Means 35 Predicted Inflow Water Storage Means 36 Neural Network Tuning Means

Claims (1)

[Claims] 1. A means for storing the input meteorological information and the actual value of the inflow water amount, a means for storing the relationship between the meteorological information and the inflow water amount as a fuzzy rule, and a prediction based on the input information using the fuzzy rule. An inflow water amount predicting device having a means for calculating the inflow water amount and an output device for presenting a prediction result, wherein a means for tuning a membership function of a fuzzy rule by a learning function of a neural network is added. Prediction device.