JPH0517976A - Distribution water amount estimating device - Google Patents

Abstract

PURPOSE:To ensure economical and stable supply of water in waterworks by providing the supply such that a daily distribution amount of water can be accurately estimated regardless of yearly fluctuation of various conditions. CONSTITUTION:In an actual data classifying means 10, distribution amounts from a day of the week, weather relation information and a distribution water flow detector 4 are fetched, stored respectively in files 11, 14, thereafter converted into a classified data required in each classifying arithmetic means 12, 15 and stored in classifying files 13, 16 in each month to also store the day distribution water amount, serving as a teach signal, in the classifying file 16. An estimated model 22 is created by using these classified data, but here by using each month classified data and each month neural network, each month estimated model 22 is created by repeating a study and also estimating the day distribution water amount by using an estimation modeled neural network 31 of the concerned month.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an improvement of a distribution amount predicting apparatus for predicting daily distribution amount from the viewpoint of ensuring planned operation of a water supply facility.

[0002]

2. Description of the Related Art In general, in water supply facilities, raw water is taken from dam-type reservoirs and rivers in the mountains, and the raw water taken at the water intake site is taken to a water purification plant through a headrace. Therefore, in such a water supply facility, there is a large delay in introducing water to the raw water from the water intake to the water purification plant. In the water purification plant, the water purification process of coagulation and sedimentation takes place, resulting in a large time delay before distribution. Therefore, the time from intake to distribution will change significantly. For this reason, it is difficult to feedback-control the intake amount of the intake plant and the intake amount of the water treatment plant based on the fluctuation of the distribution amount of water at the final end. Therefore, predict the water distribution on the day in advance,
It is important to make a daily operation plan for water intake plants, water purification plants and water distribution plants. By the way, the amount of water distribution fluctuates due to 1) weather, temperature, snowfall, sunshine hours, etc., and 2) social life conditions such as days of the week, national holidays, consecutive holidays in May, summer holidays, and special days such as New Year holidays. To do.

Therefore, conventionally, in consideration of such demands and influences, a method of predicting a daily distribution amount from past actual data according to living conditions such as a day of the week, and a distribution model while considering a weather condition in a statistical model of the daily distribution amount. The method of predicting the daily water distribution by correcting the daily distribution (for example, the amount of daily water distribution is high on days with high temperature), and the daily distribution by repeating learning while varying the weighting factor using one neural network. A method of predicting the amount of water is being considered.

[0004]

However, in the former prediction method for each living condition such as day of the week and the prediction method considering the weather condition, there is no learning ability to use the actual data obtained every day as a teacher signal, and the population movement and the piping There are problems such as not being able to follow long-term structural changes due to construction, etc., and the prediction accuracy being low because both prediction methods are based on one condition.

On the other hand, in the latter prediction method using the neural network, since the learning result in consideration of the variation of various conditions throughout one year is recorded in one neural network, for example, the temperature which is one of the weather conditions is calculated. Very wide range of maximum and minimum temperature of the year,
As a result, the accuracy of forecasting daily water distribution becomes so low that it cannot be predicted accurately.

The present invention has been made in view of the above-mentioned circumstances, and it is possible to accurately predict the daily water supply amount regardless of the annual fluctuation of various conditions, thereby ensuring the economical and stable water supply in the water supply facility. The object is to provide a water distribution prediction device.

[0007]

SUMMARY OF THE INVENTION In order to solve the above problems, the present invention performs a required classification process based on actual weather data and actual distribution amount data, and includes a unique day for each month, or a unique day. Day-of-week information that does not include days, weather information (for example, weather,
The deviation from the average maximum temperature, the deviation from the average minimum temperature) and the distribution data for each month are provided for the actual data classification means, and the neural network for each month is provided in advance. Each day by inputting the day of the week information / weather information that does not include a singular day and the water distribution amount for each month, including the singular day of each month classified by the above, and learning by changing the weighting factor. A monthly forecast model creating means for creating a daily water distribution forecast model for each month is provided, and a neural network of the relevant month is selected from the forecast modeled according to the relevant month, and a specific day is set for this neural network. , Which does not include any specific day, inputs the day of the week information, weather information, and the daily distribution of the previous day to predict the daily distribution of the day. It is a measuring apparatus.

[0008]

Therefore, according to the present invention, by taking the above-mentioned means, the result data classifying means fetches weather information and day-of-week information including a peculiar day as necessary to obtain day-of-week information for each month. For example, the weather, the deviation from the average maximum temperature,
Meteorological information such as deviation from the average minimum temperature is classified and stored in a file. Similarly, the distribution amount from the distribution flow rate detector is taken in and classified into monthly distribution information for each month and stored in the file.

If the day-of-week information, weather information, and water distribution amount for each month are classified into files as described above, the monthly prediction model creating means predicts the daily water distribution amount using the classified information. , Here, a plurality of neural networks are prepared in advance, a neural network is selected for each month, the classified information is input to the neural network of the month, and the daily water distribution on the day is given as a teacher signal, and , Predicting models for each month are created by repeating the learning with variable weighting factors.

After creating the prediction model in this way,
In the daily water distribution forecasting method, select the neural network of the relevant month, input the day of the week information including the specific day, weather information and daily water distribution of the previous day to the neural network, and predict the daily water distribution of the day. By doing so, the daily distribution of water can be predicted with high accuracy.

[0011]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing an embodiment of the device of the present invention. In this device, a distribution flow rate detector 4 is installed in the middle of distribution of tap water from a distribution reservoir 1 through a distribution pipe 2 to a distribution pipe network 3 by natural flow, that is, in the middle of the distribution pipe 2. The distribution flow rate detected by 4 is taken in by the input device (process interface) 5, and
The day of the week, weather information, and the daily water distribution of the previous day are taken in by the input device 6 such as a keyboard / mouse to predict the daily water distribution of the day. In order to predict the daily water distribution of the day, actual data classification is performed. A means 10, a month prediction model creation means 20, a daily water distribution amount prediction means 30 and the like are provided.

The result data classification means 10 includes day / weather information (for example, temperature) fetched from the input device 6.
Based on the day of the week / meteorological information file 11 storing the day of the week / meteorological information of this result file 11, for example, if it is the 30th day of the month, a predetermined process is performed for each day for 1 to 30 days. A day of the week / weather classification data calculating means 12 for classifying into day / meteorological performance data, and a day / weather classification file 13 for storing the day of the week / meteorological performance data classified by this classification calculation means 12 It is provided.

Further, the actual result data classifying means 10 stores the actual amount of distributed water stored in the input device 5 for storing the amount of distributed water, and based on the actual amount of distributed water in the actual amount of distributed water file 14, for example, on the 30th day of the month. If there is, a predetermined amount of water is processed for each day for 1 to 30 days, and the distribution amount classification calculation means 15 for classifying into the distribution amount actual result for each month (each month). A distribution amount classification file 16 for storing the classified distribution amount actual data for each month is provided. It should be noted that the water distribution amount file 16 also stores the water distribution amount of the day, which is a teacher signal (output).

Next, each month prediction model creating means 20
Twelve sets of neural networks corresponding to two months are provided, and after selecting the neural network of the corresponding month based on the calendar (month) data, the result classification from the two classification files 13 and 16 is performed for this neural network. Learning function 2 that inputs data and teacher signals and repeatedly obtains the optimum output while changing the weight coefficient
The learning model 1 is used to create the prediction model 22 for each month.

Further, the daily distribution amount predicting means 30 selects the neural network of the month corresponding to the calendar (month) data from the prediction model 21 and fetches the selected neural network 31 from the input device 6. It inputs the day of the week, weather information, and the daily water distribution on the previous day, and predicts and outputs the normalized daily water distribution on that day. Reference numeral 32 is a signal conversion means for converting the normalized daily water distribution amount into the actual daily water distribution amount. The daily water distribution amount converted by the signal conversion means 32 is displayed on the screen of the display device 7 such as a CRT. To be done. Next, the operation of the above-described device will be described separately for the performance data classification means 10, each month prediction model creation means 20, and the daily water distribution prediction means 30. A. Operation of the performance data classification means 10

In the result data classifying means 10, the day of the week and the weather information sequentially fetched by the input device 6 are stored in the day of the week and the meteorological result file 11, and the distribution amount detected by the distribution flow rate detector 4 is inputted to the input device 5. And stores it in the actual distribution amount file 14.

Here, the day of the week / weather classification calculating means 12 performs a predetermined classification process described later based on the day of the week and the weather information record data stored in the file 11 to classify the day of the week / meteorological record data for each month. Stored in the classification file 13, and the distribution amount calculation means 15 also performs a predetermined classification (normalization) process based on the distribution amount actual data stored in the file 14 to classify the distribution amount actual data for each month. Then, the water distribution amount on the day when it becomes a teacher signal (output) is also subjected to a predetermined classification (normalization) process and stored in the classification file 16. Therefore, first, the classification processing of the day of the week, the weather, and the distribution amount of the previous day for inputting to the neural network of each month prediction model creating means 20 will be described. A-1 Information regarding day of week Information regarding this day of the week is converted into a 7-bit 0, 1 pattern shown below and stored in the classification file 13. Sunday → 1000000 Monday → 0100000 Tuesday → 00100000 Wednesday → 0001000 Thursday → 0000100 Friday → 0000010 Saturday → 00000001 A-2 Meteorological information A-2a. For weather information, see

[0018] Expressed separately on the day before and on the morning and afternoon of the day. The representation method is divided into, for example, three types of fine, cloudy, and rain, converted into the following 3-bit patterns of 0 and 1, and stored in the classification file 13. Fine → 100 Cloudy → 010 Rain → 001 A-2b. Information on temperature is divided into maximum temperature and minimum temperature, and the deviation from the average maximum temperature and average minimum temperature is calculated and classified.

(A). Information about maximum temperature ... First, the average maximum temperature and the variance of the maximum temperature are calculated from the actual temperature data. Assuming that the maximum temperature on the i-th day is θmax (i) [° C], the average maximum temperature is the average θmax [° C].
C] is

[0020]

[Equation 1] Can be expressed as Furthermore, this average θmax [° C]
The maximum temperature variance σθmax [° C] is

[0021]

[Equation 2] The calculation formula is Here, n is the number of data.

Next, the actual maximum temperature on the i-th day θmax (i)
[° C], average θmax [° C] and maximum temperature variance σ
The average θma from θmax [° C] based on the following equation (3)
Find the deviation (tilde θmax) from x (i).

[0023]

[Equation 3]

(B). Information about minimum temperature ... Classify information about minimum temperature in the same way as maximum temperature. i
If the daily minimum temperature is θmin (i) [° C], the average minimum temperature is θmin [° C]

[0025]

[Equation 4] Can be expressed as Furthermore, this average θ min [° C]
The minimum temperature variance σθ min [° C] is

[0026]

[Equation 5] Can be calculated from Here, n is the number of data.

Further, the minimum temperature actual result on the i-th day θmin (i)
[° C], average θ min [° C] and minimum temperature variance σ
The average θ from θ min [° C] based on the following equation (6)
Find the deviation (tilde θ min) from min (i).

[0028]

[Equation 6] A-3 Information on water distribution on the previous day

Information on the amount of water distribution on the previous day can be
Classify. Here, the actual amount of daily water distribution on day (i-1) [m
³ / Day] is Q (i-1)
The daily water distribution Qyester (i) can be expressed by Qyester (i) = {Q (i-1) -Qmin} / (Qmax-Qmin) ... (7). Where Qmax is the maximum daily water supply
[M³ / Day], Qmin is the minimum daily water supply [m³ / Day], n
Is the number of data. A-4 Classification of prediction target information

Here, classification is performed according to the following method from the information on the feature of the amount of water distribution on the day (i), which is the teacher signal of the neural network forming the month prediction model creating means 20. Also in this case, i
Actual daily water distribution [m ³ / Day] is Q (i), the normalized daily water distribution Qlearn (i) is Qlearn (i) = {Q (i) −Qmin} / (Qmax−Qmin). 8)

Therefore, the day of the week / weather classification calculating means 12 determines, for example, the day of the week, the weather (clear,
Cloudy, rain), deviation from average θmax (i) (tilde θmax)
And the classification information such as deviation (tilde θmin) from the average θmin (i) is obtained and stored in the day / weather classification file 13,
On the other hand, the daily water distribution classification calculation means 15 obtains the daily water distribution Qlearn, which is a teacher signal, and stores it in the water distribution classification file 16 in addition to the daily water distribution Qyester for the number of days of the relevant month. B. About the operation of each month prediction model creating means 20

The month prediction model creating means 20 selects the neural network corresponding to the month of the calendar and classifies the days of the week and the weather-related items stored in the files 13 and 16 in the input layer neurons of the neural network. In addition to the information, the daily water distribution classified the previous day is input, and the daily water distribution classified as the teacher signal for the output of the output layer neuron is also input. Therefore, the weight coefficient between each neuron including the intermediate layer neuron is input. While repeating learning with the learning function 21 while changing the, the prediction model 22 for each month having an optimum weighting coefficient is created.

The back propagation method is used for this learning. The back-propagation method is a learning method for a neural network having a hierarchical structure in which a network error propagates back from the output layer to the input layer. The learning procedure will be described below. The suffix for each month is omitted because the formula becomes complicated.

Step 1: Day-of-week and weather-related information are input to the input layer of the neural network, and the intermediate layer and the output layer are operated according to the following neuron model. The output Hj of the j-th neuron in the hidden layer is

[0035]

[Equation 7]

It is obtained from the equation: Where I ₁ is the output of the i-th neuron of the input layer, W _1j is the weighting coefficient of the i-th neuron of the input layer and the j-th neuron of the intermediate layer, 1 is the number of input layers, and m is
Is the number of intermediate layers, and f () is the threshold function of the intermediate layers. On the other hand, the output ok of the k-th neuron in the output layer is

[0037]

[Equation 8] Obtained from Here, W _jk is a weighting coefficient of the j-th neuron in the _hidden layer and the k-th neuron in the output layer, m is the number of hidden layers,
n is the number of output layers.

Step 2: Learning is performed by modifying the weighting coefficient of the network so as to minimize the sum of squared errors between the output ok of the k-th neuron in the output layer and the teacher signal yk of the k-th neuron in the output layer. The learning of the weighting factors of the intermediate layer and the output layer is to calculate _{ΔW jk in the} following equation and correct W _jk . W _jk (t-1) = W _jk (t) + ΔW _jk (t) …… (11) △ W _jk (t) = −ε ・ d _k (t) ・ H _j (t) …… (12) ) D _k (t) = ok (t) -yk (t) (13) where t is the number of learnings, ε is a parameter that determines the size of one modification, and d _k is the output layer It is an error.
Next, in learning the weighting factors of the input layer and the output layer, _{ΔW ik} is calculated based on the following equation, and W _ik is corrected.

[0039]

[Equation 9] In the above equation, d _j is the back propagation error of the intermediate layer, and f () is f
It is the derivative of (). Furthermore, the following equation is used to reduce vibration and accelerate learning convergence. ΔW _jk (t) = −ε · d _k (t) · H _j (t) + α · _{ΔW ik} (t-1) (17) ΔW _ij (t) = −ε · dj (t) I _i (t) + α · ΔW _ij (t-1) (18) In the above equation, the former term is the correction amount, and the latter term excluding α is the previous weight. α is a parameter for stability.

Therefore, the monthly prediction model creating means 20 creates the prediction model 22 by writing the weighting factors obtained by the above learning into different neural networks for each month. Therefore, it is possible to obtain the prediction model 22 including twelve neural networks for each month. C. Operation of the daily water distribution forecasting means 30

The daily water distribution forecasting means 30 selects the neural network 31 corresponding to the month of the calendar from the forecast model 22 for each month. Then, when the day of the week of the day, the weather information, and the daily water distribution of the previous day are input to the neural network 31, since the weighting coefficient of the neural network 31 is set in advance by learning, the neural network 31 is calculated. It is possible to predict daily water supply. Since the predicted daily water distribution for the day is normalized, the signal conversion means 32 converts the daily water distribution into the actual daily water distribution and displays it on the display device 7.

Therefore, according to the configuration of the above embodiment, the day of the week, the weather record and the distribution record of the distribution flow rate detector 4 are converted into classification data by performing the required classification process. It is possible to make a file for each month so that it can be used in the optimal state for the neural network.
Moreover, each month prediction model creating means 20 prepares an independent neural network for each month, selects the neural network of the corresponding month, inputs the filed classification data, and repeats learning to obtain optimum weights. Since the prediction model 22 with the coefficient is created, the fluctuation range of the maximum and minimum temperatures in the monthly unit is small, and the daily distribution of water with high accuracy can be predicted. Also, the prediction model 22 can be created without a complicated structure.

Further, since the prediction model 22 is created by using the daily water distribution amount obtained on a daily basis as a teacher signal and the daily water distribution amount is predicted using the prediction model 22, for example, long-term operation due to population migration or plumbing work. It can sufficiently follow structural changes.

Incidentally, FIG. 2 is a diagram showing the result of identifying the daily water supply amount. The solid line in the figure shows the actual water distribution and the dotted line in the figure shows the predicted water distribution, but the predicted water distribution is almost the same as the actual water distribution. It can be seen that the prediction accuracy is still high.

Although the day of the week information is stored in the file 11 in the above embodiment, the signal may be directly converted into the classification file 13 and stored instead of being stored in the file 11. Also, the device for predicting the daily water distribution by learning the daily water distribution results was explained. For example, regarding the number of neurons in the middle layer of the neural network for each month, while changing the number variously when creating the model. Build the model,
You should select a model with a high legitimacy rate. Further, the present apparatus is applied to the prediction of daily water distribution, but it can also be applied to other prediction targets such as power prediction as long as the neural network is constructed according to the features such as the day of the week and the weather. Further, by adding inputs of specific days (May holidays, Bon holidays, etc.) to the input layer of the neural network, it is possible to perform more accurate prediction. In addition, although the daily water distribution on the day is output, this device has a causal relationship with the input and output, such as large water distribution when the temperature is high / clear.
It can be easily applied to things like small water distribution when the temperature is low and it is raining. Besides, the present invention can be variously modified and implemented without departing from the scope of the invention.

[0046]

As described above, according to the present invention, a prediction model is created using the classification information for each month and the neural network for each month, and the neural network of the corresponding month is selected to be distributed daily. Since the amount of water is predicted, a daily amount of water distribution can be accurately predicted even if there are large annual fluctuations of various conditions, and it is possible to provide a water distribution amount prediction device that can sufficiently secure an economical and stable water supply in a water supply facility.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing an embodiment of a distribution amount prediction apparatus according to the present invention.

FIG. 2 is a diagram showing the relationship between the actual distribution amount of daily distribution amount and the predicted distribution amount.

[Explanation of symbols]

1 ... water reservoir, 2 ... water pipe, 4 ... water flow detector, 5, 6
...... Input device, 7 ... Display device, 10 ... Result data classification means, 11 ... Day of the week / meteorological result file, 12 ... Day of the week / meteorological classification calculation means, 13 ... Day of the week / meteorological classification file, 14 ...
Water distribution amount actual file, 15 ... Water distribution amount classification calculation means, 16
... Water distribution amount classification file, 20 ... Monthly prediction model creating means, 21 ... Learning function, 22 ... Monthly prediction model, 30
... Daily water distribution predicting means, 31 ... Neural network for the relevant month, 32 ... Data converting means.

Claims (1)

[Claims] [Claim 1] In a distribution amount predicting device for predicting a distribution amount of water distributed from a water supply facility, required classification processing is performed based on the meteorological record information and the distribution record record information, and day of the week information and weather Information and a result data classification means for classifying water distribution amount for each month, and a neural network for each month is provided in advance, and the day of week information and weather information for each month classified for the neural network of the corresponding month and Monthly prediction model creating means for creating a daily water distribution forecasting model for each month by inputting the water distribution for each month and varying and learning the weighting coefficient, and the forecasting modeling according to the corresponding month Select the neural network of the relevant month from among the specified, input the day of the week information, weather information and water distribution on the previous day to this neural network and enter the day Water distribution amount predicting device being characterized in that a daily water distribution amount predicting means for predicting the amount of water.