JP7399631B2 - Water demand forecasting system and its method - Google Patents

Description

The present invention relates to a water demand prediction system and method for predicting water demand for each reference period for a predetermined water distribution area where water is distributed from water distribution equipment via a water distribution pipe network.

Conventionally, water supply facilities have been subject to controls for distributing water to consumers including general households. Water demand not only fluctuates every day, every predetermined time, depending on the day of the week and the time of day, but also varies depending on external factors such as weather and temperature. Therefore, there are conventional examples that attempt to accurately predict water demand.

For example, in Japanese Patent Application Laid-open No. 2012-99049 , it is stated that the actual value of water demand for the same day of the week as the forecast day is obtained from a predetermined period in the past, and the average value for each time is calculated for 24 hours a day. A water distribution planning and forecasting system is disclosed that forecasts water demand using water demand pattern values.

This water distribution planning and forecasting system acquires the latest actual water demand values, and if the deviation between the water demand pattern value (predicted value) and the actual value deviates from a predetermined tolerance range, the system calculates the difference between the pattern value and the actual value. In order to correct the pattern value so that the water demand is reduced, a certain amount is added or subtracted from the pattern value in an attempt to improve water demand prediction accuracy.

Japanese Patent Application Publication No. 2012-99049

Although the above-mentioned water distribution planning and forecasting system can improve the prediction accuracy of water demand in cases where water demand uniformly increases or decreases, for example, when washing time changes depending on the weather, water demand In cases where water demand is shifted, brought forward, or postponed in time, correcting the pattern value will actually worsen the accuracy of water demand forecasting. Therefore, an object of the present invention is to provide a system that can accurately predict water demand even in cases where changes in water demand are not uniform.

In order to achieve the above object, the present invention is to predict the water demand for each reference period for a predetermined water distribution area where water is distributed from water distribution equipment via a water distribution pipe network, Water demand, weather information of the water distribution district, and day of the week information are acquired, and as related conditions for the water demand in the reference period, the weather information, day of the week information, and a predetermined range before the reference period are used for the predetermined period. The water demand is set, and the similarity between the related conditions in each reference period of the past predetermined period and the related conditions in the reference period for which the water demand is to be predicted is calculated. By preferentially using the actual value of water demand for a reference period with conditions, the water demand for the reference period for which the water demand is to be predicted is predicted.

According to the present invention, it is possible to provide a system that can accurately predict water demand even in cases where changes in water demand are not uniform.

It is an example of a block diagram of a water supply system including a computer system as a water demand prediction system. This is an example of a water demand data management table. This is an example of a day of the week/weather data management table. This is an example of a setting screen for water demand prediction conditions. This is an example of a prediction condition management table. FIG. 2 is an explanatory diagram of a kernel ridge regression model. It is a graph explaining the relationship between the latest water demand data on the prediction date and the learning date. It is a graph explaining the relationship between the latest water demand data on the prediction date and the learning date. It is a graph explaining the relationship between the latest water demand data on the prediction date and the learning date. This is an example of a water demand forecast data management table. This is an example of a prediction result display data management table. This is an example of a prediction result display screen. This is an example of a model parameter management table.

Embodiments of the present invention will be described in detail below with reference to the drawings. Note that the present invention is not limited to the embodiments described below.

FIG. 1 shows a block diagram of the water supply system 102. The water supply system includes a water distribution facility 100, a water distribution facility monitoring and control system (SCADA) 103, a computer system 101 for predicting water demand, and a network 111 that connects the monitoring and control system 103 and the computer system 101. .

The water distribution equipment includes a water treatment plant 104, a pump 105, a water distribution reservoir 106, a water distribution pipe network 109, and measuring instruments 107 and 108, and a monitoring control system 103 monitors and controls the pump 105 and the measuring instruments 107 and 108. .

Water purified at the water purification plant 104 is pumped to a distribution reservoir 106 by a pump 105, and after being temporarily stored in the distribution reservoir 106, it is distributed to a customer 110 via a water distribution pipe network 109. The pump 105, the water level meter 107, and the flow meter 108 each measure the operating state (ON/OFF state) of the pump 105, the water level of the water distribution reservoir 106, and the amount of water distributed to the customer 110 (water demand) at a one-minute cycle. , and is transmitted to the supervisory control system 103.

The monitoring control system 103 acquires the measured water demand (in units of 1 minute) of the target water distribution district, which is distributing water from the water distribution equipment via the water distribution pipe network 109, and calculates the sum of the hourly values for each hour. The hourly demand is aggregated and transmitted to the computer system 101 via the network 111.

A weather information providing system (weather information source) 112 that exists outside the water supply system 102 updates the actual values of weather information for the area including the target water distribution district and the predicted values of weather information for the near future such as the next day and the day after. Every time a value is obtained, it is distributed to the computer system 101 via the network 111.

The computer system 101 acquires water demand data from the monitoring control system 103 and weather information data (including forecasts) from the weather information providing system 112 via the network 111, and performs the above acquisition every day (every reference period). Using the data, a predicted calculation of the hourly demand from a predetermined time (for example, 0 o'clock) to 24 hours ahead is performed, and the predicted result is transmitted to the monitoring control system 103 via the network 111.

The monitoring and control system 103 acquires the water demand forecast results, and creates an hourly water operation plan (water purification plan for the water purification plant 104) from the predetermined time until 24 hours ahead according to the water demand forecast results. , an operation plan for the pump 105, a water storage plan for the water distribution reservoir 106, etc.), and controls the operation of the water purification equipment of the water purification plant 104 and the pump 105 based on the water operation plan. As described above, a daily water demand forecast and a water operation plan based on the forecast are performed, and the operation of the water supply system is controlled based on the water operation plan so as to supply an appropriate amount of water to consumers. ing.

The computer system 101 includes general computer hardware such as a controller (CPU), a storage device (RAM, hard disk, flash memory, etc.), an input section 121 (keyboard, mouse, etc.), a display section 122 (display, printer, etc.). Be prepared.

The storage device includes a data acquisition unit 123, a prediction condition setting unit 124, a water demand prediction unit 125, a prediction result display unit 126, and a model parameter determination unit 127 as functional modules realized by the controller executing a program. It is set. Each functional module may be referred to as a means, a unit, or the like.

The storage device stores a water demand data management table 131, a day of the week/weather data management table 132, a prediction condition management table 133, a model parameter management table 134, a water demand prediction data management table 135, and a prediction result display data management table 136. has been done. The controller uses these tables when executing the above module.

The data acquisition unit 123 acquires the water demand performance data transmitted from the monitoring control system 103 and the weather information data transmitted from the weather information providing system 112, and registers them in a predetermined table.

The prediction condition setting unit 124 inputs the prediction start time input by the administrator of the computer system 101 through the input unit 121, conditions related to the water demand (weather, day of the week, water demand in the immediate period, etc.), and the prediction calculation. Conditions used for predictive calculations, such as the number of most recent past days (study days) to be used, are set in a predetermined table.

The water demand prediction unit 125 predicts the water demand on the prediction date based on the kernel ridge regression model using the prediction date and the related condition data for each day of the past predetermined period, and monitors and controls the prediction result. It is distributed to the system 103.

The prediction result display section 126 creates a display screen of the water demand prediction results and displays this on the display section 122.

The model parameter determination unit 127 determines model parameters of the kernel ridge regression model used for predictive calculation of water demand.

The water demand data management table 131 is a data group for managing hourly water demand measurements.

The day of the week/weather data management table 132 is a data group for managing daily day of the week information and weather information (including forecasts).

The prediction condition management table 133 includes information such as the prediction start time, information related to water demand (weather, day of the week, water demand in the nearest period, etc.), and the number of days in the most recent past (learning date) used for prediction calculations. This is a data group for managing conditions for predictive calculations.

The model parameter management table 134 is a data group for managing model parameters of the kernel ridge regression model used for predictive calculation of water demand.

The water demand forecast data management table 135 is a data group for managing hourly water demand forecast values.

The prediction result display data management table 136 contains predicted water demand values, measured values (actual values), water demand accumulation errors, and reservoir water level conversion values of the accumulation errors for each hour from the prediction start time. , is a data group for managing display information of prediction results.

By executing the following processes (1) to (5), the computer system 101 calculates the water demand based on the water demand in the past day with conditions similar to the conditions related to the current water demand. Achieve highly accurate water demand forecasts that match changes in water demand by realizing forecasting.

(1) Obtaining water demand and weather information (2) Setting forecast conditions (3) Forecasting water demand (4) Displaying water demand forecast results (5) Determining model parameters Below, (1) to ( 5) will be explained based on FIGS. 2 to 13. Regarding the water demand/weather information acquisition process (1), the monitoring and control system 103 collects the latest water demand measurement value and its measurement time every time the hourly water demand of the target water distribution district is aggregated. It is transmitted to the computer system 101 via the network 111.

The data acquisition unit 123 of the computer system 101 sequentially acquires the transmitted water demand measurement values (actual values) and their measurement times, and registers them in the water demand data management table 131. FIG. 2 shows an example of the water demand data management table 131 registered by the data acquisition unit 123.

The weather information providing system 112 delivers the actual values of weather information for the area including the target water distribution district and the predicted values of the next day's weather information to the computer system via the network 111 once a day.

The data acquisition unit 123 acquires the distributed weather information, classifies the weather information into maximum temperature, whether it is sunny or not, whether there is rain or not, and whether there is snow or not, date, day of the week (Monday to Sunday, holidays), It is registered (updated) in the day of the week/weather data management table 132 along with the actual/forecast distinction. FIG. 3 shows an example of the day of the week/weather data management table 132 registered by the data acquisition unit 123.

The prediction condition setting process (2) will be explained. In general, the daily water demand (time series for 24 hours) changes depending on the weather, maximum temperature, day of the week, recent water demand (time series), etc. Days with similar related conditions tend to have similar water demands.

Therefore, the computer system 101 predicts the water demand on the prediction day using past water demand performance data (data for multiple days is possible) having the above-mentioned related conditions similar to those on the prediction day. The prediction condition setting process (2) is the water demand forecast calculation, which is necessary for the water demand forecast calculation described above, such as the forecast start time, related conditions for water demand, and the number of days in the most recent past day to be referenced in the forecast calculation. The necessary preconditions (prediction conditions) are set in advance.

The administrator of the computer system 101 uses the input unit 121 to input the above-mentioned prediction start time, related conditions for water demand (selected from the above-mentioned weather, maximum temperature, day of the week, latest water demand, etc.), and input information in the prediction calculation. Enter the number of days in the most recent past day to reference. Then, the prediction condition setting unit 124 of the computer system registers the set prediction conditions in the prediction condition management table 133.

FIG. 4 shows an example of a display screen for setting prediction conditions displayed on the display unit 122 by the prediction condition setting unit 124. This display screen displays conditions related to water demand, such as the current day's weather, maximum temperature, day of the week, and recent water demand (time series), which are appropriate for the target water distribution district. It is configured so that the administrator can select the

If the most recent water demand amount (time series) is selected, the administrator also inputs the number of hours of the most recent water demand amount to be used for forecast calculations and the period. For example, if the forecast start time is 0:00 and the period of the latest water demand is 8 hours, the water demand time series data from 16:00 to 00:00 (24:00) is used to predict the water demand from 0:00 to 24:00 the next day. Demand forecasting is done.

When the administrator selects the most recent water demand (time series), the administrator can use the water demand for forecast calculations as "water demand time series data alone" or as "water demand time series data + difference data". or use it as "water demand time series data + total amount data".

As will be described later, as a condition related to the above water demand, not only the most recent water demand time series data, but also their difference data or total amount data are used in the forecast calculation, so that forecast calculations with different characteristics (actual value trend and It is possible to make predictions with similar waveforms, predictions with small integrated errors from actual values, etc.).

When the administrator selects the day of the week information, the administrator can select whether to use the day of the week information for prediction calculations by "weekdays and holidays (2 categories)", "by each day of the week and holidays (8 categories)", or by "each day of the week/holiday (8 categories)". Choose whether to use the system by holiday (8 categories), but change the designated days of the week for consecutive holidays of 3 or more days.

"Changing the specified day of the week for consecutive holidays of 3 days or more, although each day of the week and holidays are divided into 8 categories" means changing the day before a consecutive holiday of 3 days or more to Friday, the first day of the same to Saturday, and The final day will be changed to Sunday. This is because depending on the water distribution district, changing the day of the week as described above may improve the prediction accuracy of water demand. For example, in the case of a three-day holiday of "Friday (holiday), Saturday, and Sunday", the previous day is Thursday, but since it is a weekday before the holiday, water demand may show a pattern similar to Friday's water demand. In addition, for example, in the case of a three-day holiday of "Saturday, Sunday, and Monday (holiday)", the last day Monday (holiday) is a holiday, but because it is a holiday before a weekday, it shows a similar pattern to the water demand on Sunday. There are cases. Therefore, changing the above days of the week will lead to improved prediction accuracy.

FIG. 5 shows an example of the prediction condition management table 133 registered by the prediction condition setting unit 124. The setting contents of the prediction condition management table 133 in FIG. 5 correspond to the setting contents when the prediction conditions are set as shown in FIG. 4.

As described above, the prediction condition setting unit 124 performs the prediction condition setting process.

Next, the water demand forecasting process (3) will be explained. The computer system 101 includes related condition data for the water demand on the forecast day (weather, maximum temperature, day of the week, latest water demand time series, etc.), the above related condition data for the water demand for each day in the most recent past predetermined period, And, using the water demand record data corresponding to the prediction target time period for each day above, prioritize the water demand record value for at least one day in the most recent past predetermined period that has the above-mentioned related condition data similar to the forecast date. It is used to predict water demand on the forecast date.

For example, the degree of similarity between the above-mentioned related condition data on the prediction date and the above-mentioned related condition data on each day of the most recent past predetermined period is calculated, and the prediction target time period of the past day that has related condition data with a high degree of similarity is calculated. A predictive calculation is performed using a model formula that gives priority to the actual demand value (for example, by taking a weighted average of the actual water demand values according to the degree of similarity of each past day). The computer system 101 uses, for example, a kernel ridge regression model as a model for predictive calculation based on the degree of similarity as described above.

と表すことができる。カーネルリッジ回帰モデルは上記の考え方に基づくものであり、（式１）における入力データと実績データとの近さを表す関数をガウスカーネル関数で表し、それに対する乗数を実績データのＹ座標値でなくパラメータで表したモデルであり、以下のように定式化される。 The kernel ridge regression model will be explained using FIG. 6. When performance data (learning data) indicated by circles on the XY plane are given, consider predicting the Y coordinate value Ynew corresponding to the input of the X coordinate value Xnew. At this time, Ynew can be expected to take a value close to the Y coordinate of track record data near Xnew, and there is no need to take much consideration to track record data far from Xnew. Therefore, when N pieces of performance data (learning data) are given, the value Ynew corresponding to the input Xnew is

It can be expressed as. The kernel ridge regression model is based on the above idea, and the function that expresses the closeness between the input data and the actual data in (Equation 1) is expressed as a Gaussian kernel function, and the multiplier for it is expressed not as the Y-coordinate value of the actual data. It is a model expressed by parameters, and is formulated as follows.

ここで、Ｎ：学習データの日数
αｉ：回帰パラメータ（ｉ＝１，・・・，Ｎ）
（Ｘｔ（ｉ），Ｙｔ＋ｎ（ｉ））：直近過去Ｎ日間における第ｉ日のモデル学習データ（予測日の入出力データと同じ時刻に対応した過去日の入出力ペアデータ）を用いて予測する。ここで、ｋ（Ｘｔ，Ｘｔ（ｉ））は、予測日の入力データＸｔと過去第ｉ日の学習用入力データＸｔ（ｉ）との近さを表すガウスカーネル関数であり、

ここで、||Ｘｔ－Ｘｔ（ｉ）||：入力データ（ベクトル）Ｘｔ、Ｘｔ（ｉ）の距離（Ｌ２ノルム、各ベクトルの各成分の残差平方和の平方根をとったもの）
β：ハイパーパラメータ（β＞０）
で定義される。ガウスカーネル関数は０～１の連続値をとり、予測日の入力データＸｔと過去第ｉ日の学習用入力データＸｔ（ｉ）との距離が近いほど１に近い値をとり、距離が遠いほど０に近い値をとる。よって予測日の入力データ（水需要量に対する関連条件データ：天候、最高気温、曜日、直近水需要時系列）に類似する（距離の近い）入力データを持つ過去日のデータほど予測値に対する影響が大きくなり、予測日の入力データに類似する（距離の近い）入力データを持つ過去日のデータを優先的に利用した予測が行われる。 Yt: water demand at time t (time increments are hourly),
Z1: Binary variable indicating whether or not it will be sunny on the predicted day (1: Yes, 0: Absent),
Z2: Binary variable representing the presence or absence of rain on the forecast day,
Z3: Binary variable representing the presence or absence of snow on the forecast day,
Z4: Maximum temperature on the predicted day,
Wi: Binary variable representing the day of the week of the prediction date (one variable, W1 only, when dividing by weekdays/holidays; eight variables, W1 to W8, corresponding to Mondays to Sundays/holidays, when dividing by Monday to Sunday/holidays)
The input data (vector) consisting of related condition data for water demand (weather, maximum temperature, day of the week, latest water demand time series) is expressed as Xt=(Yt, Yt-1,..., Yt-m+1, Z1, Z2, Z3, Z4, Wi)
Then, the water demand Yt+n at time t+n (n=1,...,24) is

Here, N: number of days of learning data
αi: Regression parameter (i=1,...,N)
(Xt(i), Yt+n(i)): Predict using model training data on the i-th day in the most recent past N days (input/output pair data of the past day corresponding to the same time as the input/output data of the prediction day) . Here, k(Xt, Xt(i)) is a Gaussian kernel function representing the closeness between the input data Xt on the prediction date and the learning input data Xt(i) on the i-th day in the past,

Here, ||Xt-Xt(i)||: Distance between input data (vectors) Xt and Xt(i) (L2 norm, the square root of the residual sum of squares of each component of each vector)
β: Hyperparameter (β>0)
Defined by The Gaussian kernel function takes a continuous value from 0 to 1, and the closer the distance between the input data Xt on the prediction date and the learning input data Xt(i) on the i-th day in the past, the closer it is to 1, and the farther the distance, the closer the value is to 1. Takes a value close to 0. Therefore, the more similar (closer) input data to the input data on the forecast day (related condition data for water demand: weather, maximum temperature, day of the week, most recent water demand time series) is, the more the data from the past will have a greater impact on the forecast value. Forecasting is performed preferentially using data from past days that have input data that is similar (close in distance) to input data on the prediction date.

In the above explanation, the input data vectors corresponding to the forecast date and past days (learning dates) are composed of the weather, maximum temperature, day of the week, and latest water demand time series, but here we will focus only on the water demand time series. However, as shown in the example of FIG. 7, it is assumed that there is water demand data corresponding to the forecast date and water demand data corresponding to two learning days 1 and 2.

At this time, it is assumed that other conditions (weather, maximum temperature, day of the week) are the same for each day. It is assumed that the learning day 1 data (the latest water demand amount) always exceeds the forecast day data by a predetermined amount, and the learning day 2 data fluctuates up or down from the forecast day data by the predetermined amount. At this time, generally, if the other conditions described above are the same, it is expected that the above relationship tends to continue to some extent even in the water demand data after the prediction start time of the prediction date and learning days 1 and 2. As mentioned above, the distance between the two input data vectors (base maintenance degree) is calculated by the L2 norm, that is, the sum of the squares of the errors of each component, so the distance between the predicted day data and the two learning days 1 and 2 data are about the same level, and as shown in Figure 7, the actual water demand during the predicted time period for learning day data 1 and 2 is likely to be reflected in the water demand forecast on the same day to the same extent (Figure 7 is an image) (This may not necessarily be the case.)

Here, if the difference value between each time of the latest water demand time series and the previous time is added as a component of the input data vector, each difference value of the latest water demand time series of the forecast day data and the learning day 1 data will be the same. Since the distance (similarity) with the predicted date data is closer to the predicted date data, the learning day 1 data is closer than the learning day 2 data, and as shown in Figure 8, A prediction result in which the actual demand is largely reflected in the water demand prediction on the prediction date, that is, a prediction result with a waveform similar to an actual value that exceeds the actual value on the prediction day by a predetermined amount, is likely to be obtained. However, in this case, since the predicted value always exceeds the actual value, the cumulative error in water demand prediction will increase. Since prediction errors in water demand appear as errors between actual and predicted water levels in water distribution reservoirs, an increase in the cumulative error may lead to deviations of water levels in water distribution reservoirs from the upper and lower limits. Therefore, for water management purposes, it may be more desirable for the predicted value of water demand to fluctuate up and down around the actual value so as not to increase the integration error, rather than for the predicted water demand result to always exceed the actual value.

Next, if the total value of the most recent water demand time series is added as a component of the input data vector, the total value of the most recent water demand time series of the forecast day data and the learning day 2 data will be of the same value, so the prediction The distance (similarity) between the learning day 2 data and the learning day 1 data is closer than that of the learning day 1 data. The prediction result is likely to be significantly reflected in the demand forecast, that is, the prediction result may fluctuate up or down by a predetermined amount from the actual value on the prediction date. In this case, since the actual value and the total water demand value are likely to be close to each other, it can be expected that the cumulative error in water demand forecasting will be reduced.

Therefore, by adding a new difference value or total value of the latest water demand time series to the input data vector corresponding to the forecast date and past day (learning date), forecast results and waveforms similar to the actual water demand can be obtained. This makes it possible to select prediction results with small cumulative errors. Addition of the above-mentioned difference value or the above-mentioned total value to this input data vector is the difference data addition when the latest water demand is selected as the related condition data for the water demand explained in the above prediction condition setting process (2), The administrator of the computer system 101 can change the above selection depending on the purpose of water demand prediction.

Here, the regression parameter αi (i=1,...,N) is the error 2 It is determined so as to minimize the evaluation function R, which is composed of a sum of products and a regularization term for preventing overfitting to the learning data.

上記評価関数Ｒは回帰パラメータαｉの関数であり、Ｒを最小化する回帰パラメータαｉは解析的に算出可能である（数式は省略）。よって予測計算を行う毎に、時刻ｔまでに得られた予測日の入力データＸｔ、過去日（学習データ日）の入出力データＸｔ（ｉ）、Ｙｔ＋ｎ（ｉ）を用いて、回帰パラメータαｉが算出でき、（式２）より予測対象時間ｔ＋ｎにおける水需要量予測値Ｙｔ＋ｎが算出できる。ｎをｎ＝１，・・・，２４まで変化させて上記計算を繰り返すことにより（パラメータαｉもｎ毎に算出）、予測対象時間帯（時刻ｔ＋１から時刻ｔ＋２４までの２４時間分）の水需要量予測値が算出できる。

The evaluation function R is a function of the regression parameter αi, and the regression parameter αi that minimizes R can be calculated analytically (the formula is omitted). Therefore, each time a prediction calculation is performed, the regression parameter αi is The predicted water demand value Yt+n at the prediction target time t+n can be calculated from (Formula 2). By repeating the above calculation by changing n to n = 1, ..., 24 (parameter αi is also calculated for each n), the water demand for the prediction target time period (24 hours from time t+1 to time t+24) can be calculated. The predicted amount can be calculated.

The regularization term includes a parameter λ (not described) that defines the degree of prevention of overfitting to the learning data. β included in the above kernel function and λ included in the regularization term above are hyperparameters (parameters that control the behavior of the algorithm). It is determined in advance using a brute force method or the like so as to minimize the sum of squares of errors between the predicted value and the actual value of demand.

Therefore, the related condition data for the water demand on the forecast day (weather, maximum temperature, day of the week, most recent water demand time series, etc.), the above related condition data for the water demand for each day in the most recent predetermined period, and each of the above. Based on the kernel ridge regression model, the water demand for the day(s) in the most recent past predetermined period that has the above-mentioned related condition data similar to the forecast date is calculated based on the kernel ridge regression model using water demand record data corresponding to the forecast time period of the day. It becomes possible to predict and calculate the water demand amount on the forecast date by preferentially using the actual amount value.

The water demand forecasting process (3) is basically performed once a day when the latest measured water demand up to the forecast start time is obtained, but as will be explained later, water demand forecast errors may occur. If the administrator of the computer system 101 determines that it has expanded, water demand prediction processing (re-forecasting) can be performed according to instructions from the administrator.

The water demand prediction unit 125 of the computer system 101 obtains the prediction start time, related conditions for water demand (weather, maximum temperature, day of the week, most recent water demand time), which are necessary for water demand prediction, from the prediction condition management table 133. series, etc.), and the number of days in the most recent past (number of days of learning data).

Then, the water demand prediction unit 125 calculates the latest water demand actual value up to the prediction start time on the prediction date and a predetermined past day (learning date) from the water demand data management table 131 and the day of the week/weather data management table 132. , weather, maximum temperature, day of the week data, and water demand actual value data in the predicted time period on a predetermined past day (learning date).

At this time, the water demand forecasting unit 125 changes the classification of the day of the week information (by weekday/holiday/by day of the week/holiday/by day of the week/holiday + consecutive holidays of 3 or more days) according to the settings made in the above process (2). ), and adding information related to water demand time series data to the above input data vector (water demand time series data alone / water demand time series data + difference data / water demand time series data + total quantity data). The water demand prediction unit 125 then obtains the hyperparameters β and λ of the kernel ridge regression model from the model parameter management table 134.

The water demand forecasting unit 125 uses the acquired data to calculate a predicted value of water demand in the predicted time period on the forecast date using the above-mentioned prediction calculation based on the kernel ridge regression model, and calculates the predicted value of the water demand in the predicted time period on the forecast date. is registered in the water demand forecast data management table 135 and distributed to the monitoring control system 103 via the network 111. As a result, the monitoring control system 103 can formulate an operation plan for the pump 105 and a planned water level for the water distribution reservoir 106 based on the distributed water demand prediction results. FIG. 10 shows an example of the water demand prediction data management table 135 registered by the water demand prediction unit 125.

As described above, the water demand prediction unit 125 performs water demand prediction processing.

Next, the display process (4) of water demand prediction results will be explained. This process (4) is executed every time the latest water demand forecast result is obtained by the above process (3), and every time the latest water demand measurement value (actual value) is obtained by the above process (1). be done. The prediction result display unit 126 of the computer system 101 acquires the latest prediction result from the water demand prediction data management table 135, acquires the latest water demand actual value from the water demand data management table 131, and displays the prediction start time. The cumulative error of water demand is calculated by taking the sum of the prediction errors (=predicted value - actual value) from (1 o'clock) to each time.

Then, the prediction result display unit 126 divides the calculated cumulative error by the cross-sectional area of the water reservoir 106 to calculate a water level conversion value of the water reservoir 106. Then, the prediction result display unit 126 registers the acquired and calculated water demand forecast date, actual value, integration error, and the water demand water level conversion value of the water demand integration error in the prediction result display data management table 136. , a prediction result display screen in which the above data is displayed in a graph is created and displayed on the display section 122.

FIG. 11 shows an example of the prediction result display data management table 136 registered by the prediction result display unit 126. Further, FIG. 12 shows an example of a prediction result display screen created by the prediction result display unit 126.

As described above, the prediction result display section 126 performs a process of displaying the water demand prediction results.

The administrator of the computer system 101 views the above forecast result display screen and determines whether the water demand forecast error is increasing. If it is determined that the water demand forecast error is increasing, the administrator of the computer system 101 updates the water demand using the latest data. Re-predictions can be made. When performing re-prediction, the administrator inputs a new prediction start time (current time) from the input unit 121, as in process (2) above, and instructs re-prediction from the above-mentioned prediction start time.

Then, the water demand prediction unit 125 calculates a predicted water demand for 24 hours from the new prediction start time using the latest water demand actual value, weather data, etc., as in the above process (3). The prediction results are registered in the water demand prediction data management table 135 and distributed to the monitoring control system 103. This makes it possible to re-forecast and calculate water demand using the latest data even if the prediction error increases.

Next, the model parameter determination process (5) will be explained. This process (5) may be performed periodically, for example, once a year or once every six months. The model parameter determination unit 127 sets the initial values (sufficiently large values) of the parameters β and λ of the kernel ridge regression model, and sets each day in the most recent past predetermined period (for example, 3 months) as the prediction date. , the water demand forecasting unit 125 is called to execute the water demand forecasting process (3) in the forecast time period of each forecast day. At this time, the conditions used for prediction are those registered in the prediction condition management table 133.

Then, the model parameter determining unit 127 obtains the actual water demand value for each forecast day from the water demand data management table 131, and calculates the prediction error of 2 between the predicted value and the actual value for all forecast time periods and forecast days. Calculate the sum of multiplications.

The model parameter determination unit 127 repeats the prediction process while decreasing the values of the parameters β and λ by a predetermined value, and optimizes the values of the parameters β and λ that minimize the sum of squares of the calculated prediction errors. It is registered in the model parameter management table 134 as a parameter value. FIG. 13 shows an example of the model parameter management table 134.

As described above, the model parameter determination unit 127 performs model parameter determination processing.

In the embodiment described above, as the model parameters continue to be corrected and the learning of the kernel ridge regression model progresses, the computer system 101 estimates the water demand based only on the relevant conditions at the time of predicting the water demand. Can be predicted.

As described above, according to the embodiment of the present invention, related condition data (weather, maximum temperature, day of the week, latest water demand time series, etc.) for the water demand on the forecast day, data on the water demand in the most recent past predetermined period, etc. Using the above-mentioned related condition data for the daily water demand and the water demand record data corresponding to the prediction target time period of each day above, the days in the most recent past predetermined period ( The water demand forecast for the forecast date is performed by preferentially using the actual water demand values of (multiple locations allowed). As a result, predictions are made with reference to water demand on past days that are close to changes in water demand, making it possible to predict water demand with high precision in accordance with changes in water demand.

100...Water distribution equipment 101...Computer system 102...Water supply system 103...Monitoring control system 104...Water purification plant 105...Pump 106...Water distribution reservoir 107...Water level meter 108...Flow meter 109...Water distribution pipe network 110...Customer 111...Network 121... Input section 122...Display section 123...Data acquisition section 124...Forecast condition setting section 125...Water demand forecast section 126...Forecast result display section 127...Model parameter determination section 131...Water demand data management table 132...Day of the week/weather data Management table 133...Prediction condition management table,
134...Model parameter management table 135...Water demand forecast data management table 136...Prediction result display data management table

Claims (7)

A water demand forecasting system that predicts water demand for each reference period for a prescribed water distribution area where water is distributed from water distribution equipment via a water distribution pipe network,
memory and
controller and
Equipped with
What the controller executes based on the water demand prediction program is:
Obtaining actual values of water demand in the predetermined water distribution area, which are continuously measured by the water distribution equipment, from a monitoring control system of the water distribution equipment, and recording them in the memory together with weather information and day of the week information. and,
The related conditions used for predicting the water demand amount are set from corresponding past information, and the related conditions are weather information, day of the week information, and The actual value of the water demand for a predetermined period,
Regarding the actual value of the water demand for the past predetermined period recorded in the memory, the corresponding past related conditions are determined for each of the reference periods, and the relevant conditions and the reference period for which the water demand is predicted are calculated. Calculating the degree of similarity with related conditions using a kernel ridge regression model;
Based on the actual value of the water demand in at least one of the periods corresponding to the plurality of reference periods in the past predetermined period, the predicted calculation of the water demand is performed using a kernel ridge regression model with reference to the degree of similarity. What to do and
transmitting the results of the predictive calculation to the supervisory control system;
including;
The predicted calculation of the water demand may be performed in such a manner that the actual value of the water demand in the reference period of the past predetermined period having a higher degree of similarity is the higher the actual value of the water demand in the kernel ridge regression model. Including preferential use for predictive calculations,
Setting the related conditions used for predicting the water demand includes:
setting a time-by-time difference value of water demand for a predetermined period before the reference period as the related condition;
adding a total value of water demand for a predetermined period before the reference period to the related conditions; and
and selecting the information on the day of the week for each day of the week and holiday, and if the day of the week is a consecutive holiday of three or more days and the previous day is a weekday but not a Friday, the previous day is a weekday. , the first day of said consecutive holidays shall be regarded as Saturday, and the last day of said consecutive holidays shall be regarded as Sunday even if it is not Sunday.
Water demand forecasting system. The predicted calculation of the water demand is performed by using the related conditions in the reference period for which the water demand is to be predicted as input data, and calculating the related conditions and water demand actual results in each reference period of the past predetermined period. Calculating the degree of similarity using a kernel ridge regression model using the value as learning data, and making a prediction from the amount of water demand for each reference period of the predetermined period in the past based on the degree of similarity;
The water demand prediction system according to claim 1. The controller calculates the predicted water demand amount and the actual value of the water demand amount, the cumulative error between the predicted water demand amount and the actual value of the water demand amount from a predetermined time, and the water distribution facility of the cumulative error. displaying at least one of the water level equivalent values of the water distribution reservoir;
The water demand prediction system according to claim 1. The water demand forecasting system is configured such that the predetermined period before the reference period and the past predetermined period can be input into the water demand forecasting system via an input device.
The water demand prediction system according to claim 1. The water demand forecasting system enables information on the day of the week as the related condition to be input into the water demand forecasting system via an input device, and selects a predetermined day of the week from the day of the week information of a plurality of weekdays, holidays, and public holidays for the day of the week. information can be selected,
The water demand prediction system according to claim 1. The water demand forecasting system allows the actual values of water demand as the related conditions to be input in chronological order through an input device, and allows the user to select a predetermined actual value from among the chronological actual values. I made it to
The water demand prediction system according to claim 1. A method in which a computer predicts water demand for each reference period for a designated water distribution area where water is distributed from water distribution equipment via a water distribution pipe network,
The computer includes:
acquiring water demand of the predetermined water distribution district, weather information of the predetermined water distribution district, and day of the week information;
As related conditions for the water demand in the reference period, weather information and day of the week information for a predetermined period before the reference period, and water demand in a predetermined range before the reference period are set;
Using the kernel ridge regression model, calculate the degree of similarity between the related conditions in each reference period of the past predetermined period and the related conditions in the reference period for which the water demand is to be predicted, and select conditions with a high degree of similarity. Preferentially using the water demand actual value for the reference period held, predicting the water demand amount for the reference period for which the water demand amount is to be predicted,
Setting the related conditions used for predicting the water demand includes:
setting a time-by-time difference value of water demand for a predetermined period before the reference period as the related condition;
adding a total value of water demand for a predetermined period before the reference period to the related conditions; and
selecting the day-of-the-week information for each day of the week and holiday; and if the day-of-the-week information is a consecutive holiday of three or more days and the previous day is a weekday but not a Friday, the day before is a weekday Friday. and the first day of said consecutive holidays shall be deemed to be Saturday, and the last day of said consecutive holidays shall be deemed to be Sunday even if it is not Sunday.
Said method.

Images