JP2007199862A - Energy demand predicting method, predicting device, program and recording medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an energy demand predicting method in consideration of the irregular action of a client while coping with a single demand pattern. <P>SOLUTION: Prediction conditions are read, and sample data to be used for regression analysis or learning are prepared from a database 15. Also, a prediction model determined from prediction conditions is configured, and initialization processing is executed. An analyzed basic prediction model is prepared from regression analysis or the learning of a neural network by using the prediction model and all sample data. Then, the correction processing of the prediction model is repeated by analyzing the prediction model again after selecting sample data, so that the prediction model is determined. Weather information, calendar information and demand data corresponding to a time zone to be predicted are input to the analyzed prediction model determined as mentioned above, and a demand prediction amount integrated value after the lapse of each time is calculated. The transition of the time series demand amount integrated values is predicted from those values. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an apparatus and method for predicting energy demand in a home or the like.

  In a system for supplying energy demand such as electric power and heat load using a distributed energy device such as a fuel cell, prediction of electric power and heat load demand is performed as in the system and control method described in Patent Document 1. There is a method of performing control of the distributed energy device based on an optimal operation plan calculated using those predicted values. Even in the control of household fuel cells, it is possible to perform efficient or economical operation by calculating a predicted value of energy demand and performing operation control based on an optimal plan.

  The demand for energy such as electric power / hot water supply for each household in a general household depends on the irregular living behavior of the consumer, so it is difficult to predict with high accuracy. The demand for hot water supply is more difficult to predict because there are many times when there is no demand and a demand peak occurs once. However, since a normal hot water storage tank is installed, if the prediction error of the hot water supply integrated value is small, the influence of the hot water supply demand on the optimum operation is small.

Conventionally, there is a prediction method using a multiple regression analysis or a neural network for a power demand for a large area. Multiple regression analysis is capable of calculating predicted values at high speed, is being established as a prediction method, and is widely used as various prediction models. On the other hand, neural networks have been proposed as prediction models for complex time-series data because of their nonlinear characteristics, and their effectiveness has been demonstrated.
JP 2005-102364 A

  About the hot water supply demand which becomes the single demand pattern mentioned above, it is difficult to approximate by a continuous type, or to apply to a regression model type | formula. In addition, since the energy demand for each household in a general household greatly depends on the living behavior of the consumer, there is a problem that the neural network is overfitted with sample data when the lifestyle pattern becomes irregular.

  An object of the present invention is to provide an energy demand prediction method and a prediction device that can cope with a single demand pattern and take into account the irregular behavior of consumers.

  In order to achieve the above object, the present invention uses weather information, calendar information, and actual demand in the past specific time period as an explanatory variable with respect to the prediction target time period, and after a specific time has elapsed from the prediction time point. Sample data consisting of the explanatory variable and the data of the objective variable in the past fixed period in the prediction model, comprising a prediction model in which the specific time in the objective variable is set in time series, with the demand amount integrated value as the objective variable Is used to analyze the relationship between the explanatory variable and the objective variable, and the value of the objective variable obtained by inputting the explanatory variable data at the prediction target date and time into the analyzed prediction model It is characterized by predicting the transition of the demand integrated value after elapse of a specific time by using the predicted value.

  According to the present invention, since the transition of the demand amount integrated value is targeted for prediction, in hot water supply demand prediction in a system having a fuel cell and a hot water tank, a prediction that has an error that has little influence on the optimal operation of the fuel cell is performed. Can do.

  Next, embodiments of the present invention will be described with reference to the drawings.

  FIG. 1 shows the configuration of an energy control device and a consumer using the energy demand prediction device of the present invention. This embodiment is comprised from the energy control apparatus 1 and the customer 2 which were mutually connected by the communication line. The consumer 2 includes a power demand 21, a hot water tank 22, a hot water supply demand 23, and a fuel cell 24, and the fuel cell 24 is connected to the power system 3 by a power line. The energy control device 1 includes a meteorological information reception unit 13, an energy demand measurement unit 14, a database 15, a prediction model calculation unit 16, an energy demand prediction device 11, an optimum operation plan creation unit 12, and the like. The weather information receiving unit 13 receives forecasts and results such as temperature and weather, and accumulates them in the database 15. The energy demand measuring unit 14 measures electric power demand and hot water supply demand and accumulates them in the database 15. The prediction model calculation unit 16 predicts electric power and hot water supply demand from the weather information and energy demand database 15, and includes a prediction model configuration unit 16A, a prediction model analysis unit 16B, and a demand amount integrated value prediction side unit 16C. The optimum operation plan creation unit 12 creates an operation pattern of the fuel cell 24 that minimizes the energy cost, using the predicted value created by the prediction model calculation unit 16. Based on this operation pattern, the start and stop of the fuel cell 24 and output control are performed. By executing the demand prediction and the optimum operation plan at regular intervals while shifting the start time, the plan can be corrected in accordance with the deviation of the forecast and optimum control can be performed.

  Next, an outline flow of energy demand prediction will be described with reference to FIG.

  First, the prediction model construction unit 16A reads prediction conditions such as a target date and time and a period of sample data to be used, creates sample data used for regression analysis and learning from the database 15 (step 101), and is determined from the prediction conditions. The prediction model is configured and initialization processing is performed (step 102).

  Next, in the prediction model analysis unit 16B, first, a basic prediction model that has been analyzed by regression analysis or neural network learning is created using the prediction model and all sample data (step 103), and then sample data After the selection process, the prediction model correction process is repeated by analyzing the prediction model again to determine the prediction model (steps 104 to 107). Details of this will be described later with reference to an embodiment.

  Finally, in the demand amount integrated value prediction unit 16C, weather information, calendar information, and demand data corresponding to the time zone to be predicted are input to the analyzed prediction model, and the demand amount integrated value until after each time has elapsed. Calculate (step 108). From these values, the transition of the demand amount integrated value in time series is predicted (step 109). Moreover, the time series prediction value which is the demand amount of each time slot | zone is created by subtracting the prediction value of the previous time.

  In FIG. 3, the 1st example of the prediction model which performs the hot water supply demand prediction in this embodiment is shown. In the present embodiment, 24 hours later using a plurality of prediction models with one objective variable, which is a demand amount from the start of the prediction period until a certain time (3 hours, 6 hours,... 24 hours) has elapsed. Predict the trend of accumulated demand up to Since there is one objective variable, a multiple regression model may be used, or a neural network may be used. As shown in FIG. 3, the explanatory variable includes calendar information, weather information, and a demand actual value before the prediction target time zone. The calendar information includes a flag that sets 0 to 1 when the prediction target day is a holiday. As an example of the weather information, the average temperature on the prediction target day is used, but other maximum / minimum temperatures may be used, and information on humidity and weather may be added. As an example of the demand amount actual value before the prediction target time zone, the demand data for 3 hours immediately before the prediction target time zone, as well as the demand data for the previous 6 hours, the previous 12 hours, and the previous 24 hours are used. Sample data corresponding to the above explanatory variables and objective variables is created from the database 15, and analysis by regression analysis or neural network learning is performed.

  FIG. 4 shows a second example of a prediction model that performs demand prediction using a neural network. In this prediction model, the objective variable, which is the amount of demand from the beginning of the prediction period to a certain time (3 hours, 6 hours,... 24 hours), constitutes 8 prediction models, and until 24 hours later. Predict the trend of accumulated demand. The explanatory variable includes calendar information, weather information, and a demand actual value before the prediction target time zone. As the calendar information, in addition to the holiday flag, a flag whose day to be predicted is each day of the week is provided. Sample data corresponding to the above explanatory variables and objective variables is created from the database 15, and learning of the neural network is performed. From the learned prediction model, a demand amount integrated value is calculated every three hours, and a demand amount prediction value for each time zone is obtained.

  The time width and time increment specified in the explanatory variable and the objective variable in the prediction model described above are not limited to the time described above, and can be freely set as necessary.

  Next, an embodiment related to a prediction model correction method using sample data selection processing will be described and described.

  FIG. 5 shows a first example of a sample data selection method and a prediction model correction method. An error rate is calculated between the actual value of the objective variable in each sample data and the calculated value of the objective variable based on the analyzed prediction model (step 201). The sample data with the maximum error rate is excluded (step 202), and the prediction model is analyzed again or analyzed by regression analysis (step 203). The above processing is repeated until the end condition is satisfied (step 204). Details of the termination condition will be described later.

  FIG. 6 shows a second example of the sample data selection method and the prediction model correction method. One of the sample data is provisionally excluded, and the prediction model is analyzed repeatedly (step 301). For all cases, the correlation coefficient between the actual value of the objective variable in the sample data and the calculated value of the objective variable based on the analyzed prediction model is calculated (steps 302 and 303). Exclusion processing is performed so as to obtain a set of sample data having the maximum correlation coefficient (step 304). The above is repeated until the end condition is satisfied (step 305). In this embodiment, since the number of iterations increases, a prediction model based on multiple regression analysis that can be processed at high speed is suitable.

  In this way, the sample model is analyzed by excluding the sample data by the error or correlation coefficient between the actual value of the target variable in the prediction model and the calculated value of the target variable by the analyzed prediction model. Therefore, since the day when the consumer's daily behavior is irregular is not used, the prediction accuracy of the day when the regular daily behavior is taken can be improved.

  Next, examples of end conditions in FIGS. 5 and 6 will be described. The end condition can be when the increase in the correlation coefficient between the actual value of the objective variable and the calculated value of the objective variable based on the analyzed prediction model becomes a certain value or less. However, in this end condition, an upper limit may be set by the sample data exclusion rate.

  The function of the energy demand prediction apparatus described above is executed by recording a program for realizing the function on a computer-readable recording medium, causing the computer to read the program recorded on the recording medium. It may be a thing. The computer-readable recording medium refers to a recording medium such as a flexible disk, a magneto-optical disk, a CD-ROM, or a hard disk device built in a computer system. Further, the computer-readable recording medium is a medium that dynamically holds the program for a short time (transmission medium or transmission wave) as in the case of transmitting the program via the Internet, and in the computer serving as a server in that case Such as a volatile memory that holds a program for a certain period of time.

In one Embodiment of this invention, it is a figure which shows the structural example of the energy control apparatus using an energy demand prediction apparatus, and a consumer. It is a flowchart which shows the process in the energy demand prediction apparatus of FIG. It is a figure which shows the 1st example of the prediction model used by one Embodiment of this invention. It is a figure which shows the 2nd example which applied the neural network to the prediction model used by one Embodiment of this invention. FIG. 3 is a flowchart illustrating a first example of a prediction model correction method using the sample data selection process in FIG. 2. FIG. It is a flowchart which shows the 2nd example of the correction method of the prediction model using the selection process of the sample data in FIG.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1 Energy control apparatus 2 Consumer 3 Electric power system 11 Energy demand prediction apparatus 12 Optimal operation plan preparation part 13 Weather information reception part 14 Energy demand measurement part 15 Database 16 Prediction model calculation part 16A Prediction model structure part 16B Prediction model analysis part 16C Demand Quantity integrated value prediction unit 21 Electric power demand 22 Hot water storage tank 23 Hot water supply demand 24 Fuel cells 101-109, 201-204, 301-305 steps

Claims (8)

A method for predicting energy demand in an energy consumer performed in an energy demand prediction apparatus having a communication means, a measurement means, a prediction model construction means, a prediction model analysis means, and a demand amount integrated value prediction means,
The predictive model construction means describes the meteorological information obtained by the communication means, calendar information, and the actual demand in the past specific time zone obtained by the measuring means in the past specific time zone. And a step of configuring a prediction model in which the specific time in the objective variable is set in time series, with the demand amount integrated value after the specific time elapses from the prediction time point as an objective variable;
The predictive model analyzing means analyzing the relationship between the explanatory variable and the objective variable using sample data consisting of the explanatory variable and the objective variable data in the past fixed period in the predictive model;
The demand amount integrated value predicting means sets the value of the objective variable obtained by inputting the explanatory variable data at the prediction target date and time to the analyzed prediction model as the predicted value of the demand amount integrated value after the lapse of a specific time. An energy demand prediction method comprising: predicting a transition of an integrated demand amount after a specific time has elapsed.   The prediction model configuration means adds the value of the objective variable in the prediction model in which the specific time before the specific time set in the objective variable is set as the explanatory variable in the prediction model, and configures the prediction model. Item 2. The energy demand prediction method according to Item 1.   The energy demand prediction method according to claim 1, wherein the prediction model constituting unit uses a neural network whose objective variable is a demand amount integrated value from a prediction time point to a plurality of specific times after the prediction model.   The prediction model analysis means excludes sample data in which the error of the calculated value of the objective variable based on the analyzed prediction model with respect to the actual value of the objective variable in the prediction model is greater than or equal to a certain value from the sample data. The energy demand prediction method according to claim 1, wherein analysis is performed.   The prediction model analysis means increases the correlation coefficient between the actual value of the objective variable and the calculated value of the objective variable based on the analyzed prediction model by not using the sample data in the analysis of the prediction model. The energy demand prediction method according to any one of claims 1 to 3, wherein sample data is excluded and a prediction model is analyzed. An energy demand prediction device for predicting energy demand in an energy consumer,
Communication means;
A measuring means for measuring energy demand;
The meteorological information obtained by the communication means, the calendar information, and the actual demand in the past specific time zone obtained by the measurement means, as the explanatory variables, are used as explanatory variables, and the specific time from the prediction time point. A prediction model constituting unit that constitutes a prediction model in which a specific time in the objective variable is set in time series, the demand amount integrated value after elapse being an objective variable;
Prediction model analysis means for analyzing the relationship between the explanatory variable and the objective variable using sample data consisting of the explanatory variable and the objective variable data in the past fixed period in the prediction model;
Accumulation of demand after the elapse of a specific time by setting the value of the objective variable obtained by inputting the explanatory variable data at the prediction target date and time to the analyzed forecast model as the predicted value of the demand accumulation after the elapse of the specific time An energy demand forecasting device comprising: a demand quantity integrated value forecasting means for forecasting a change in value.   A program for causing a computer to execute the energy demand prediction method according to claim 1.   The computer-readable recording medium which recorded the program for making a computer perform the energy demand prediction method of Claim 7 of Claim 1-5.

Images