JP2009294969A - Demand forecast method and demand forecast device - Google Patents

Abstract

A demand prediction apparatus capable of probabilistically forecasting demand that fluctuates depending on weather is provided.
In a demand prediction device for obtaining a demand forecast that fluctuates depending on weather in a predetermined forecast period,
A correlation data generation unit 5 that generates correlation data of a similar period that is predetermined and similar to the forecast period from past weather record values and past demand record values;
Demand forecast value and demand forecast value probability distribution in forecast period are calculated from correlation data of similar period and weather forecast value of forecast period and weather forecast probability distribution data of weather forecast value probability distribution And a demand forecasting unit 9 to perform.
[Selection] Figure 1

Description

  The present invention relates to a demand forecasting method and a demand forecasting apparatus for performing probabilistic demand forecasting based on past weather past record values, past demand past record values, and weather forecast values.

  Electricity demand fluctuates with time, and the fluctuation pattern is subject to various influences such as human social activities and weather conditions. And since it is necessary to supply electric power according to this fluctuating electric power demand, electric power demand prediction is an important subject for the supply and demand plan and the supply and demand control. However, due to errors in weather forecasts and uncertainties in social activities, it is considered impossible to fully predict power demand. Many of the conventional demand forecasting methods output a power demand forecast value as a deterministic value based on past results and the weather forecast of the day.

For example, as a conventional maximum power demand prediction method, there is a method of predicting the maximum power demand using a multiple regression model based on the correlation between the weather conditions including the maximum temperature, the minimum temperature and humidity and the power demand. (For example, refer to Patent Document 1).
As another conventional total demand forecasting method, a forecast is made using a statistical model based on past weather data and total power demand data. From the above, there is one that corrects a predicted value by regression prediction or a neural network (for example, see Patent Document 2).

As another conventional power demand forecasting method, a future weather variable is predicted based on the value of the weather variable up to now, and the power demand is predicted from the weather variable, and the error range of the weather forecast value There is a method of estimating the error range of the power demand prediction value by estimating the power demand based on the weather prediction value of the prediction error range (see, for example, Patent Document 3).
As another conventional maximum power demand prediction method, there is a method of predicting a maximum power demand after predicting a time zone or daily load curve where the maximum power is generated (see, for example, Patent Document 4).

JP-A-6-105465 Japanese Patent Laid-Open No. 5-18995 JP-A-8-163778 JP 2000-270476 A

  In the conventional power demand forecasting method, a statistical model is created using a regression analysis or a neural network based on the correlation between past weather data and power demand data, and future power demand is deterministically determined based on the model. I was predicting. In this case, it is unclear to what degree the accuracy is predicted, and there is a problem that information is insufficient when the operator finally uses it as a means for determining a predicted value. In addition, as a method for estimating the prediction error, the demand prediction is performed by using several prediction values in the weather prediction error range as input data, but one power demand prediction value is output for one weather prediction value. There is a problem that it is a definite method as a method and a means of prediction. In addition, such problems are not limited to power demand, but have similar problems for demands that change due to weather.

  The present invention has been made to solve the above-described problems, and an object of the present invention is to provide a demand prediction method and a demand prediction apparatus that can perform a demand prediction of a demand that fluctuates depending on weather. .

The present invention relates to a demand prediction method for obtaining a demand forecast that fluctuates depending on weather in a predetermined forecast period.
Correlation data of a similar period that is similar to the forecast period using correlation data of past weather record values and past demand record values;
From the weather forecast probability distribution data of the weather forecast value in the forecast period and the probability distribution of the weather forecast value,
The demand prediction is performed by calculating the demand forecast value in the forecast period and the probability distribution of the demand forecast value.

In addition, the present invention provides a demand forecasting device for obtaining a demand forecast that fluctuates depending on weather in a predetermined forecast period,
A correlation data generation unit that generates correlation data of a similar period that is predetermined and similar to the forecast period from past weather record values and past demand record values;
Calculate demand forecast value and demand forecast value probability distribution in forecast period from correlation data of similar period and weather forecast probability distribution data of forecast period and forecast forecast probability distribution And a demand forecasting unit to perform.

The demand prediction method of the present invention is a demand prediction method for obtaining a demand forecast that fluctuates according to weather in a predetermined forecast period.
Correlation data of a similar period that is similar to the forecast period using correlation data of past weather record values and past demand record values;
From the weather forecast probability distribution data of the weather forecast value in the forecast period and the probability distribution of the weather forecast value,
Since the demand forecast is calculated by calculating the demand forecast value in the forecast period and the probability distribution of the demand forecast value, it is possible to forecast the demand that fluctuates depending on the weather.

Moreover, the demand prediction device of the present invention is a demand prediction device that obtains a demand prediction that fluctuates according to weather in a predetermined prediction period.
A correlation data generation unit that generates correlation data of a similar period that is predetermined and similar to the forecast period from past weather record values and past demand record values;
Calculate demand forecast value and demand forecast value probability distribution in forecast period from correlation data of similar period and weather forecast probability distribution data of forecast period and forecast forecast probability distribution Since the demand forecasting unit to perform is provided, the demand forecast of the demand that fluctuates according to the weather can be stochastically performed.

Embodiment 1 FIG.
Embodiments of the present invention will be described below. FIG. 1 is a diagram showing a configuration of a demand prediction apparatus according to Embodiment 1 of the present invention. In the figure, a demand forecasting device captures a past weather record value, a meteorological record data fetching unit 1 for fetching past meteorological record values, a meteorological record database 2 for storing past meteorological record values, and a past demand past record value. Demand actual data acquisition unit 3, demand actual data database 4 for storing past demand actual values, past weather actual values and past in a similar period predetermined to a predetermined prediction period for prediction Data generation unit 5 that generates correlation data from the actual demand values, a weather forecast data capturing unit 6 that captures the weather forecast values during the forecast period, and the weather forecast values during the forecast period and the weather results during the similar period A prediction error calculation unit 8 that calculates the probability distribution of the weather prediction value from the value and creates the weather prediction probability distribution data, and the prediction period from the correlation data and the weather prediction probability distribution data in the similar period A demand prediction unit 9 for forecast by calculating the probability distribution of the forecast value and forecast value in, and a demand prediction database 10 which stores forecast.

  Next, the demand prediction method of the demand prediction apparatus of Embodiment 1 configured as described above will be described. In the following description, the weather actual value will be described based on the temperature value, and the demand actual value will be described based on the electric power demand value. Further, a description will be given on the assumption that a predetermined forecast period for forecasting demand is a forecast date, and a similar period that is a predetermined similar period to the forecast period. First, the meteorological record data fetching unit 1 stores a meteorological record value, for example, a temperature value as temperature measurement data, in the meteorological record database 2 every unit time, for example, every 3 minutes. As described above, if it is performed every 3 minutes, 20 temperature values in one hour and 480 temperature values in one day are stored together with the measured times. Further, in the demand result data capturing unit 3, a demand result value at the same time as the weather result value stored in the weather result database 2, for example, a demand result value of power demand, is stored in the demand result database 4 every unit time. The For example, the unit time is 3 minutes, and the same number of demand value data at the same time as the temperature measurement data is stored.

  Next, the correlation data generation unit 5 extracts a weather result value and a demand result value at the same time from the weather result database 2 and the demand result database 4. The value is plotted on a space with the vertical axis, and the correlation between the actual weather value and the actual demand value is expressed stochastically. Correlation data generated by the correlation data generation unit 5 is plotted as shown in FIG. 6, for example, when the temperature value in all years is plotted on the horizontal axis and the demand value corresponding thereto is plotted on the vertical axis, as in the conventional case. become. There is a positive correlation between the air temperature value and the demand value at a certain temperature, and heating demand is a factor beyond that. Negative correlation occurs. Therefore, demand forecast cannot be expressed by simple linear regression.

  Therefore, conventionally, for example, as shown in FIG. 7, it is possible to avoid a state in which cooling demand and heating demand are mixed on the same graph by displaying the data divided into seasons and time zones. For this reason, the relationship between the temperature value and the demand value has a positive correlation in the summer and a negative correlation in the winter, and can be expressed by linear regression. Therefore, in the conventional method, the relationship between the temperature value and the demand value is represented by a straight line m, and the predicted temperature is output by inputting the predicted temperature into the regression equation.

  In the present invention, as shown in FIG. 2, the correlation data generation unit 5 generates correlation data on a predetermined past date for a predetermined predicted date. The “similar day” mentioned here is, for example, the predicted date (month, day, day of the week), the month, the week (week of the month), and the past day in which the day of the week is the same (one year or more before) Day). Note that the relationship of similar days to the forecast date is not limited to this. For example, when setting a similar day for the same month, week, and day of the week, the month and day of the week are set. Various examples are conceivable, such as setting a past day that is the same as a similar day, and setting a past day that has the same month and week. That is, it is necessary to set the day when the demand tends to be similar to the forecast date, and it is needless to say that this is set as appropriate according to the forecast demand.

  Then, on this similar day, the actual weather value and the actual demand value on that day are extracted from the actual weather database 2 and the actual demand database 4 (step S1 in FIG. 2). Next, weighting is performed on each data in order to prevent the past data far from the predicted date from being considered in the same way as the nearest data. For example, as shown in FIG. 3, a monotone decreasing function is set so that the weight of the nearest data is increased by the weighting coefficient, and the extracted data is weighted by the time difference from the predicted date (step S2 in FIG. 2). ). Next, a simultaneous distribution (correlation) is obtained from the weighted weather record values and demand record values on all similar days (step S3 in FIG. 2). The correlation obtained at this time is expressed stochastically as shown in FIG. This stochastic is shown in a plan view in FIG. 8, but is actually shown three-dimensionally as shown in the graph G when viewed from the predicted temperature line n in FIG. is there. As can be seen from FIG. 8, the actual weather value and the actual demand value are represented by a probability distribution, and the probability distribution of the actual demand value can be expressed as a conditional distribution when a certain weather forecast value is assumed.

  Next, the meteorological prediction data fetching unit 6 fetches meteorological forecast values of future meteorological data, such as temperature forecast data, by the Japan Meteorological Agency or weather companies. And the prediction error calculation part 8 calculates the error with respect to a weather prediction value by comparing this taken-in data with the weather actual value on a similar day. Specifically, the prediction error calculation unit 8 calculates the frequency distribution of the difference between the weather prediction value of the prediction date acquired by the weather prediction data acquisition unit 6 and the weather result value of the similar day extracted from the weather result database 2. Obtained (step S4 in FIG. 4). Next, from the distribution of all errors on similar days, the weather forecast value of the forecast date, the error with respect to the weather forecast value, and the probability of occurrence of the error, that is, the probability distribution of the weather forecast value are obtained, and the weather forecast probability distribution data is calculated (Step S5 in FIG. 4).

  Next, the demand prediction unit 9 uses the correlation data generated by the correlation data generation unit 5 and the weather prediction probability distribution data of the prediction error calculation unit 8 to predict the demand prediction value and the demand prediction indicated by the probability index. A probability distribution of values is calculated and stored in the demand forecast database 10 as a demand forecast. Specifically, the demand forecasting unit 9 sets the simultaneous distribution P (X, Y) in which the past weather result data generated by the correlation data generation unit 5 is X and the past demand result data is Y as well. Capture (step S6 in FIG. 5). Next, the demand corresponding to the weather forecast value is calculated as the demand forecast value (step S7 in FIG. 5). Next, the probability distribution of the error with respect to the weather prediction value calculated by the prediction error calculation unit 8, that is, the prediction probability P (X) of the weather prediction value is captured (step S8 in FIG. 5). Next, the appearance probability of the demand forecast value Y under the condition of the weather forecast value X1 becomes a conditional probability, and is obtained by P (Y | X1) = P (X1, Y) / P (X1). The appearance probability of the demand forecast value Y to be finally obtained is obtained by P (Y) = Σ {P (Xn) · P (Y | Xn)} (step S9 in FIG. 5). Then, the demand forecast is output to the demand forecast database 10 and stored.

  This relationship will be specifically described with reference to FIG. The probability distribution A of the weather forecast value B is shown on the temperature axis. And the probability distribution C of the demand forecast value with respect to the weather forecast value B is represented on the straight line Q parallel to a demand axis. This distribution is obtained by multiplying the appearance probability of the weather forecast value B by the conditional probability. The probability of the demand forecast value D under the weather forecast value B is the probability E at the point where the weather forecast value B and the demand forecast value D intersect. The probability of the demand forecast value D to be finally obtained is the sum of the probabilities obtained similarly to the probability E with respect to the probability distribution A of the weather forecast value B, that is, the area of the bold line in FIG. Become. Thus, the actual weather forecast value (predicted temperature value) always includes an error, is regarded as a random variable, and becomes a two-dimensional random variable between demand and weather.

  According to the demand forecasting apparatus and demand forecasting method of the first embodiment configured as described above, the demand forecast value of demand fluctuating due to weather and the probability distribution of demand forecast value can be output. It is possible to make a probabilistic judgment, and it becomes clearer, and it is possible to perform demand planning and demand control excellent in economy and reliability.

  In the first embodiment, as the actual weather value, the temperature is shown as an example. However, the present invention is not limited to this, and it is possible to similarly use humidity, weather, and the like that have a correlation with demand. Needless to say. In addition, although the forecast date is described as the predetermined forecast period, and the similar date is described as the predetermined similar period, the present invention is not limited thereto. The forecast month is the forecast month, the similar period is the similar month, and the forecast period is the forecast time. It is good also as similar time as a similar period, and it cannot be overemphasized that it can carry out similarly if it is a period for prediction. In addition, power has been explained as demand. This is because when the power is predicted, the distribution of errors can be used effectively in demand planning and demand control, and demand fluctuates due to other weather conditions. Needless to say, it can be done in the same way.

Embodiment 2. FIG.
FIG. 10 is a diagram showing the configuration of the demand prediction apparatus according to Embodiment 2 of the present invention. In the first embodiment, the example in which the prediction probability of the weather prediction value is calculated from the weather prediction value and the weather actual value has been shown. However, the present invention is not limited to this. For example, the prediction probability (probability distribution) is used as the weather prediction value. ) May be taken in by the probabilistic weather forecast data fetching unit 60 and the demand forecasting unit 9 may make the demand forecast in the same manner as in the first embodiment.

It is a figure which shows the structure of the demand prediction apparatus of Embodiment 1 of this invention. It is the flowchart which showed the process of the correlation data production | generation part of the demand prediction apparatus shown in FIG. It is the figure which showed the coefficient for performing the weighting when the correlation data generation part shown in FIG. 2 produces | generates simultaneous distribution. It is the flowchart which showed the process of the prediction error calculation part of the demand prediction apparatus shown in FIG. It is the flowchart which showed the process of the demand prediction part of the demand prediction apparatus shown in FIG. It is the figure which showed the simultaneous distribution of temperature actual value and demand actual value. It is the figure which showed the simultaneous distribution of the temperature actual value and demand actual value which limited the season and time slot | zone of the simultaneous distribution in FIG. It is the figure which showed the simultaneous distribution of the temperature actual value and demand actual value in the similar day in Embodiment 1 of this invention. It is the figure which showed the relationship between the weather forecast value in this Embodiment 1 and the probability distribution of a weather forecast value, and the demand forecast value and the probability distribution of a demand forecast value. It is the figure which showed the structure of the demand prediction apparatus of Embodiment 2 of this invention.

Explanation of symbols

  5 correlation data generation unit, 8 prediction error calculation unit, 9 demand prediction unit.

Claims (6)

In a demand forecasting method for obtaining demand forecasts that fluctuate due to weather in a predetermined forecast period,
The correlation data of a similar period predetermined similar to the forecast period using correlation data of past weather record values and past demand record values;
From the weather forecast value in the forecast period and the weather forecast probability distribution data of the probability distribution of the weather forecast value,
A demand prediction method, wherein demand forecast is performed by calculating a demand forecast value in the forecast period and a probability distribution of the demand forecast value. The demand forecast method according to claim 1, wherein the probability distribution of the weather forecast value is calculated from the weather forecast value and the weather actual value corresponding to the similar period. The demand forecasting method according to claim 1 or 2, wherein the meteorological record value is a temperature value, and the demand record value is a demand record value of electric power. In a demand forecasting device that obtains a demand forecast that fluctuates according to the weather in a predetermined forecast period,
A correlation data generation unit that generates correlation data of a predetermined similar period to the prediction period from a past weather record value and a past demand record value;
From the correlation data of the similar period, the weather forecast value of the forecast period and the weather forecast probability distribution data of the probability distribution of the weather forecast value, the demand forecast value and the probability distribution of the demand forecast value in the forecast period are obtained. A demand forecasting device comprising a demand forecasting unit that calculates and forecasts demand. The demand prediction device according to claim 4, further comprising a prediction error calculation unit that calculates the weather prediction probability distribution data from the weather prediction value and the weather actual value during the similar period. The demand forecasting device according to claim 4 or 5, wherein the meteorological record value is a temperature value, and the demand record value is a demand record value of electric power.

Images