JP2017146815A - Power demand prediction device and method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a power demand prediction device and method on which social information is reflected.SOLUTION: The power demand prediction device comprises: social information feature quantity extraction means for detecting variation of a first threshold value or more for social information; prediction error determination means for detecting that difference between predicted power demand and actual power demand is a second threshold value or more; correction parameter calculation means for, when variation of the first threshold value or more occurs in the social information and a difference between the predicted power demand and actual power demand is the second threshold value or more, calculating a correction parameter for power demand; correction parameter holding means for storing the correction parameter; and power prediction correction means for outputting information to be used for correcting power demand prediction.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a power demand prediction apparatus and method, and more particularly, to a power demand prediction apparatus and method reflecting various social activities.

  An electric power company predicts future power demand and performs daily power generation operation every day according to the predicted power demand. These forecasts include power demand forecasts by time of day within 24 hours tomorrow, forecasts that reflect weather forecast information in the near future, forecasts by day of the week, forecasts by month, forecasts that reflect seasons, etc. It is created from the viewpoint.

  For example, in Patent Literature 1, for example, “Receiving weather information including a predicted temperature value, a predicted humidity value, and a predicted amount of solar radiation in a predetermined area from a weather information management server that manages weather information”. The weather information receiving means, the outdoor enthalpy coefficient representing the degree of influence of the outdoor air enthalpy corresponding to the temperature predicted value and the humidity predicted value on the power demand, and the solar radiation coefficient representing the degree of influence of the solar radiation predicted value on the power demand Based on the storage means stored respectively, the weather information received by the weather information receiving means, and the outside air enthalpy coefficient and the solar radiation coefficient stored in the storage means, the power demand which is a predicted value of the power demand in the predetermined area And a power demand predicted value calculating means for calculating a predicted value.

  Further, in Patent Document 2, “a prediction result input unit that inputs predicted values of power demand predicted by a plurality of prediction methods in a power demand prediction support apparatus that predicts future power demand in the power system; A compromise point determination means that searches for a compromise point from a plurality of prediction values input by the input means and determines one final prediction value, and a prediction result output that outputs the prediction value determined by the compromise point determination means to the outside And a power demand prediction support device characterized by comprising the above-mentioned means.

Japanese Patent Laying-Open No. 2015-097059 Japanese Patent Laid-Open No. 9-285010

  Patent Document 1 described above is a countermeasure plan that examines how to predict power demand from weather information. Cited Document 2 examines how to improve prediction accuracy. These contribute to improving the accuracy from each viewpoint, but do not take into account that the power demand reflects the activities in human society.

  For example, not only forecasting the weather, but also reflecting the bodily sensation that humans feel on the skin, the demand for warm items increases and reflected in electricity, or popular events are held and people move and this is reflected in electricity. In addition, the national team ’s national football team is scheduled to be aired late at night, and the results supported by many people are reflected in electricity.

  These activities in human society should be defined as social information in the present application, and social information directly or indirectly affects power demand. In addition, it is highly possible that this effect is not negligible when forecasting power demand. Therefore, power demand prediction is performed by reflecting the presence or absence of event events and the impact of sudden events. Furthermore, the demand for power demand, such as how people feel the heat and cold, is obtained as a numerical value, and power demand prediction is performed. It is thought that it is effective for power demand prediction.

  Accordingly, an object of the present invention is to provide a power demand prediction apparatus and method reflecting social information.

  In the present invention, social information feature amount extraction means for detecting fluctuations of the social information that are greater than or equal to the first threshold, and prediction error determination means that detects that the difference between the power demand prediction and the actual result is greater than or equal to the second threshold. A correction parameter calculating means for calculating a correction parameter for power demand when the social information fluctuates by more than a first threshold and the difference between the forecast and actual power demand is greater than a second threshold; Correction parameter holding means for storing the power, and power prediction correction means for outputting information used for power demand prediction correction.

  In addition, the present invention calculates a correction parameter for power demand for each social information when the social information fluctuates more than the first threshold and the difference between the forecast and actual power demand is greater than or equal to the second threshold, It is characterized by correcting the forecast of power demand.

  ADVANTAGE OF THE INVENTION According to this invention, the electric power demand prediction which reflected the influence by the presence or absence of an event event or a sudden event as social information can be performed.

Overall configuration diagram of embodiment 100 of power demand prediction apparatus and method The figure which shows an example of the social information feature-value J3 with a time tag. The figure which shows an example of the electric power demand information according to time. The figure which shows an example of the electric power demand information correction | amendment according to time. The figure which shows the example of the correlation of social information J1 and blurring on electric power demand. The figure which shows obtaining social information J1 from the POS system of a convenience store. The figure which shows the example of the feature-value extraction method designation | designated information in the case of a POS system. The figure which shows the example of the correction parameter classified by feature-value. The figure which shows obtaining social information J1 from a television viewpoint. The figure which shows the example of the feature-value extraction method designation | designated information in the case of television. The figure which shows the example of the feature-value extraction method designation | designated information in the case of television. The figure which shows obtaining social information J1 from the number of passengers of a station. The figure which shows the example of the feature-value extraction method designation | designated information in the case of the number of passengers in a station. The figure which shows obtaining social information J1 from traffic jam information. The figure which shows the example of the feature-value extraction method designation | designated information in the case of traffic jam information. The figure which shows the example of an oden as information for correction parameter calculation. The figure which shows the example of an event as information for correction parameter calculation. The figure which shows the example of TV program as information for correction parameter calculation. The figure which shows the example of an SNS site as information for correction parameter calculation. The figure which shows the structural example of the power demand prediction means 200 for calculating | requiring a power demand prediction value. The figure which shows the structure of the computer apparatus for implement | achieving the electric power demand prediction apparatus of this invention. The figure which shows the operation | movement flow at the time of correction parameter calculation. The figure which shows the operation | movement flow when correcting a demand forecast.

  Embodiments of the present invention will be described below with reference to the drawings.

  FIG. 1 shows an overall configuration diagram of a power demand prediction apparatus and method according to an embodiment of the present invention.

  The power demand prediction apparatus 100 is realized by software using hardware resources of a computer. When the general functions are listed, social information feature amount extraction means 1, correction parameter calculation means 2, prediction error determination means 3, The demand prediction correction means 4 and the feature-specific correction parameter holding means 5 are configured.

  Among them, the social information feature amount extraction means 1 is a social network with time tag from various social information J1 including time information and feature amount extraction method designation information J2 for extracting feature amounts used for power demand prediction therefrom. The information feature J3 is extracted and passed to the correction parameter calculation means 2 and the demand prediction correction means 4.

The prediction error determination means 3 receives the predicted value (time t) K1 and the actual value (time t) K2 predicted in advance as the time-specific power demand information K, and the time-specific power demand prediction error determination threshold K3. ) To determine that the prediction error is greater than or equal to the threshold value, and passes the result K4 to the correction parameter calculation means 2.
[Equation 1]
| Predicted value (time t) K1-Actual value (time t) K2 |> Threshold value K3 (1)
The correction parameter calculation means 2 receives the social information feature quantity J3 with time tag and the result K4 of the power demand prediction error determination for each time, calculates the correction parameter P1 for each feature quantity, and stores the correction parameter holding means for each feature quantity 5 Save to.

  The demand prediction correction means 4 receives the social information feature quantity J3 with the time tag, refers to the correction parameter holding means 5 for each feature quantity, and obtains a correction parameter P2 for power demand prediction and a log P3 as a basis for the correction. Output.

  An example of various social information J1 including time information in FIG. 1 is sales of oden at a convenience store. In the present invention, the power demand fluctuation is reflected in the power demand forecast according to the social information. I am trying to let you. Hereinafter, a specific procedure for reflecting sales of oden at a convenience store in power demand prediction will be described.

  FIG. 2 is a diagram illustrating an example of a social information feature J3 with a time tag. FIG. 2 shows oden sales (social information J1) on a certain day for each time, and shows that oden sales exceeded the threshold at time t0. In the social information feature amount extraction means 1 of FIG. 1, the fact that the sales of oden (social information J1) exceeds a threshold value is extracted as a social information feature amount J3 with a time tag. In FIG. 2, sales of oden per unit time are shown as they are, but it may be understood that the time when the threshold is exceeded is 1 and the time when the threshold is not exceeded is 0.

  FIG. 3 is a diagram illustrating an example of hourly power demand information K4. FIG. 3 shows the relationship between the power demand forecast K1 by time and the power demand record K2 by time on that day (hereinafter referred to as the day) showing the oden sales trend of FIG. 2, and the power demand record at time t1. K2 represents that the power demand forecast K1 has exceeded the threshold K3 or more. The prediction error determination means 3 determines whether the error between the predicted value (time t) K1 and the actual value (time t) K2 is equal to or greater than the threshold value K3. In FIG. 3, the predicted value (time t) K1 and the actual value (time t) K2 for each unit time are shown as they are, but even if the time exceeding the threshold K3 is 1 and the time not exceeding 0 is grasped as 0. good.

  FIG. 5 is a diagram illustrating an example of correlation between social information J1 and power demand fluctuation. As a specific example, the oden sales as social information J1 on the horizontal axis in FIG. 5 shows the sales that exceed the threshold on the day as shown in FIG. 2, and the fluctuation in power demand on the vertical axis in FIG. As shown in FIG. 3, the power demand of No. 2 shows a result that exceeds the threshold value from the predicted value on that day, and FIG. From this trend, it can be seen that oden sales are a factor in increasing the demand for electricity, and to what extent. That is, it is possible to correct the predicted value of power demand from the viewpoint of oden sales.

  For this processing, the correction parameter calculation means 2 correlates various social information feature quantities J3 with time tags individually with the result K4 of the power demand prediction error determination by time. If the correlation value between one of the social information feature quantities J3 with various time tags and the power demand prediction error determination result K4 is greater than or equal to a certain value, the social information feature quantities with time tags (oden sales) J3 Derive a function to explain the blur on demand. In addition, although it shows by linear approximation in FIG. 5, it can be set as a more accurate electric power demand forecast by approximating with a high-order function with good fit. Further, a function for explaining the power demand fluctuation may be derived from a plurality of feature amounts by multivariate analysis of the plurality of social information feature amounts J3 with time tags and the result K4 of the power demand prediction error determination.

  As described above, FIG. 4 is a diagram illustrating an example of time-dependent power demand information correction. The corrected prediction information is estimated by reflecting the degree of the upper blur in FIG. 5 in the initial prediction information. The correction parameter calculation means 2 obtains the correlation value between one of the time-tagged social information feature quantities J3 and the result K4 of the power demand prediction error determination, by shifting the time, and comparing the correlation values. You may investigate.

  The above-described power demand prediction apparatus 100 of FIG. 1 is an apparatus having two aspects: past information accumulation evaluation and reflection to the current situation. In other words, it normally performs the function of accumulating various correlation information by taking in a plurality of social information, which will be described later, and on the other hand, if it is observed that sales of oden are growing so that it exceeds the threshold, oden is accumulated. The current power demand is corrected with reference to the past information, and information on how much the correction period should be expected is given and reflected in the power demand.

  In evaluating the relationship between the oden sales trend in FIG. 2, the power demand forecast K1 by time in FIG. 3 and the power demand record K2 by time, the power demand forecast by time K1 is a forecast according to the season. It is necessary to take a method. For this purpose, the method shown in FIG.

  Next, specific examples of social information J1 and feature quantity extraction method designation information J2 to be taken into the social information feature quantity extraction unit 1 of FIG. 1 will be described.

  First, FIG. 6 shows a case where social information J1 is obtained from a POS system at a convenience store. FIG. 7 is a diagram in which the relationship with the feature quantity extraction method designation information J2 in this case is organized. In FIG. 7, products are products that can be social information. Here, oden, disposable warmers, beer and the like are listed. These are considered to increase or decrease in sales in relation to cold or hot, each product has information on the distinction of periods (year-round, winter), thresholds (μ, σ) to determine the degree of increase, decrease, It is organized and stored as feature quantity extraction method designation information J2. In order to identify the social information J1 from the POS system, id = 1 is set as the feature amount type.

  It should be noted that products that can be used as social information are not limited, and it is desirable that sales be influenced by the climate, such as oden, disposable warmers, and beer. In this sense, what can be called seasonal products is appropriate. Beer is drunk all year round, but is particularly noticeable when it is hot, so it is defined here as a seasonal product.

  As described above, FIG. 7 is an example of a data structure that designates the same analysis for products other than oden. As the designation for extracting the feature quantity of the sales of the product, [feature quantity type id = 1], the product whose sales are observed per unit time, and the period and threshold value are designated for each product. The threshold value may be extracted as a feature amount when, for example, an average of sales and a standard deviation value according to a period are specified and the average + k * standard deviation is exceeded.

  FIG. 9 shows a case where social information J1 is obtained from the viewpoint of television. 10A and 10B are diagrams in which the relationship with the feature amount extraction method designation information J2 in this case is organized. FIG. 9 shows an example of obtaining power demand prediction information from the viewpoint of a television from, for example, a television program information viewer input site. Various opinions of viewers are reflected on the TV program information viewer input site through the Internet, which can be a factor for estimating the degree of interest in future broadcast programs, and hence fluctuations in power.

  In FIG. 10a, the URL (URL1.URL2) of the TV program information viewer input site is set as the feature quantity extraction method designation information J2a from the viewpoint of the TV. In order to identify social information J1 related to the Internet, id = 2 is set as the feature type. In FIG. 10b, as the feature amount extraction method designation information J2b from the viewpoint of television, the points given to fluctuations in power demand by individual players or competitions regarding the broadcast contents are shown. Information on threshold values (μ, σ) for determining the degree of increase / decrease for individual players or competitions is organized and stored as feature amount extraction method designation information J2b.

As described above, FIGS. 9, 10a, and 10b describe examples of handling TV program information as social information. As the feature quantity extraction method designation information, [feature quantity type id = 2], the URL of the TV program information and the period are designated, and the keyword to be searched from the television program information and the keyword appearance frequency threshold at another site are designated. . [Feature quantity type id = 2] specifies the following operation. Whether or not the frequency of occurrence of a keyword on the URL1 viewer input site has increased is determined by threshold 1 and threshold 2. The threshold value is extracted as a feature amount when, for example, the appearance average and standard deviation value of unit time is specified and the appearance frequency exceeds the average + k * standard deviation. Also, the keyword tagged with the time is extracted from the TV program information on the URL2 site. In FIG. 12, for each station, FIG. 11 shows a case where social information J1 is obtained from the number of passengers at the station. FIG. 12 is a diagram in which the relationship with the feature quantity extraction method designation information J2 in this case is organized. In FIG. 12, information on threshold values (μ, σ) for determining the degree of increase / decrease in the number of passengers for each station is organized and stored as feature quantity extraction method designation information J2. 11 and 12 describe an example in which the number of passengers at the station is handled as social information. As the feature quantity extraction method designation information, [feature quantity type id = 3], a station for observing the number of passengers, and a threshold value are designated. For example, the threshold value is extracted as a feature amount when the average number of passengers and standard deviation value per unit time are specified and the number of passengers exceeds the average + k * standard deviation. Further, not only the number of passengers at the station alone, but also the number of travelers between stations may be estimated from the number of passengers at the station, and this may be compared with a threshold value to extract the feature amount. It is thought that the number of people staying in the station leads to prediction of power demand in the station.

  13 and 14 describe an example of handling traffic jam information as social information. As the time information, in addition to the traffic jam time, the time when the traffic jam information is disclosed can be used to estimate the power demand time.

  FIG. 8 is a diagram showing an example of the correction parameter for each feature amount stored in the correction parameter holding unit 5 for each feature amount in FIG. Here, correction parameters for correcting the power demand prediction for each social information described with reference to FIGS. 6, 7, and 9 to 14 are stored.

  For example, in the product information from the POS system with the feature quantity type id = 1, when the feature quantity type is oden, the time when the threshold value in FIG. 2 is exceeded is set as the correction start time, and the power demand prediction correction quantity for the period of 180 minutes thereafter. Is calculated and held. Also, from the viewpoint of the TV with the feature value type id = 2, for the player whose feature value type is XX, the time tagged with the TV program start time and the time tagged with the end time of the program are corrected. The start and end times are set, and for this period, it is calculated and held that the correction amount of the power demand prediction should be Pb. In addition, for the station number of passengers with feature quantity type id = 3, when the feature quantity type is Hitachi Station, power demand prediction is performed for a period of 30 minutes after the current time as the correction start time and 180 minutes after the end time. It is calculated and held that the correction amount should be Pc.

  In this manner, the feature-specific correction parameter holding means 5 in FIG. 1 stores information on the power demand correction timing and information on the appropriate correction amount at that time as past experience values for each of a plurality of various types of social information. ing.

  On the other hand, the demand prediction correction means 4 outputs information for reflecting in the correction of the future power demand prediction when an excess of social information is detected at the present time. Specifically, in response to the oden sales feature quantity, the feature quantity-specific correction parameter holding means 5 (FIG. 8) is referred to, and among the feature quantity type id = 1 corresponding to the sales feature quantity of the convenience store. The feature amount type 2 = oden, which is the sales feature amount of oden, is searched, and the correction amount = Pa is output as the correction parameter for power demand prediction after the correction start time = 0 minutes and the correction end time = 180 minutes. Further, as a basis for the correction, a log indicating the correlation between the sales feature amount of oden and the sales feature amount of oden and the fluctuation in power demand as shown in FIG. 5 is output.

  According to the above embodiment, even when the actual demand deviates from the demand prediction due to factors such as temperature, due to the warmth and energy consumption activities that people feel, the demand can be predicted with high accuracy.

  15 to 18 are examples in which feature amounts extracted from social information and power amount demand prediction errors are made into a database. These databases collectively store the power demand error experienced at this time in addition to the start time, end time, feature amount, information site, and the like.

  For example, FIG. 15 shows that seven odens were sold in a store in 30 minutes and the power demand error at that time was ΔP0, and FIG. 16 shows an event at K department store (product exhibition, ranger show). ) Indicates that the power demand error during the holding period was ΔP1 and ΔP2, and FIG. 17 shows that the power demand error during the airing on the TV program table of the skating athletes was ΔP3. FIG. 18 shows that the power demand error during the period of the SNS site introducing the player of interest was ΔP4.

  Since these are only data in which the power demand prediction error exceeds the threshold, the correction parameter calculation means 2 extracts data when the power demand prediction error does not exceed the threshold from social data, and performs a correlation test. Do. When this result is output as a log as a basis for the correction together with a correction parameter for power demand prediction, the user can know the basis for the correction and is useful for decision making.

  In the prediction error determination means 3 of FIG. 1, the difference between the time-specific power demand information predicted value K1 and the time-specific power demand information actual value K2 is compared using the time-specific power demand prediction error determination threshold K3 as a reference to calculate a correction parameter. It is used for whether or not. For this reason, the hourly power demand prediction error determination threshold value K3 is required to be an appropriate value. FIG. 19 shows a configuration example of the power demand prediction means 200 for obtaining a power demand prediction value reflecting an appropriate time-specific power demand prediction error determination threshold K3.

  The power demand prediction unit 200 shown in FIG. 19 calculates an appropriate threshold value using information from at least three types of databases DB1, DB2, and DB3. The database DB1 stores the actual demand curve D1 for a plurality of days in units of the actual actual demand curve, and the database DB2 stores the feature quantity D2 extracted from the climatic conditions and social information of the day, and the database DB3 stores a demand curve cluster. Further, as other inputs, information on the clustering condition D4 and the prediction date condition D5 is input from the outside.

  The power demand forecasting means 200 inputs the actual demand curve D1 for a plurality of days with the actual actual demand curve as a unit, the feature quantity D2 extracted from the climatic conditions and social information of the day, and the clustering condition value D4. Then, the actual demand curve D1 is clustered to learn the feature amount D2 extracted from the climate condition distribution and social information for each cluster. Here, the demand curve D1 is a power demand sequence for each time period such as 30 minutes to 1 hour. When the climatic conditions and social information of the day to be predicted are input, the similarity M1 is calculated and output for the clusters with reference to the feature distribution D2 extracted from the climatic condition distribution and social information for each cluster, and the parameters for each cluster are set. The demand forecast value curve M2 is output using this.

The demand curve clustering unit 201 clusters the actual demand curves using, for example, hierarchical clustering from the actual demand curves D1 and clustering condition values D4 for a plurality of days with the actual actual demand curve as a unit. For example, if the clustering condition value D4 is hierarchical clustering, the threshold value of the distance between the actual demand curves classified into the same cluster is designated. Alternatively, a threshold value of the element distance from the average value of the cluster is designated. For example, if “5% of the average demand value at each time” is designated, the error between the demand curves in each cluster of the clustering result can be 5% or less. Then, this value or a function value using this value as a parameter is set as a threshold used by the prediction error determination means 3. Thereby, the precision which extracts the case where it deviates from the demand value prediction by temperature can be improved. By clustering with the actual demand curve, you can extract similar days according to time, such as mid-winter days and warm seasons of spring and autumn,
It is possible to forecast demand according to similar days. As a result, the relationship between the sales of seasonal products and the power demand (FIGS. 2 and 3) can be derived.

As the distance between the demand curves, the Euclidean distance of the demand value at each time is used as in equation (2), but other functions may be used.
[Equation 2]
Distance between demand curves (Di (t), Dj (t)) = √ (Σ (Di (t) -Dj (t)) ² )… (2)
The per-cluster condition statistical value extraction unit 202 extracts, for each cluster, statistical values such as average and standard deviation for factors such as the temperature extracted at each time, the amount of solar radiation, and feature values extracted from social information. It is good to extract every time. Further, if a feature quantity effective only for discrimination between clusters is obtained, the discrimination rate of clusters can be improved. Reference numeral 203 denotes an inter-cluster identification condition extraction unit, and 204 denotes an appearance rate calculation unit for each demand curve cluster.

  By using this power demand prediction means 200, it is possible to perform demand prediction based on the conditions of the day to be predicted, and the fluctuation range is a value corresponding to the clustering condition value. Therefore, if the demand curve calculated from the temperature and the amount of solar radiation is more than the clustering condition value, it is considered that there was some factor. In addition to the clustering condition value, a statistical value such as an average or standard deviation for factors such as the temperature at each time, the amount of solar radiation, and a feature amount extracted from social information may be applied.

  The power demand prediction apparatus of FIG. 1 described above is actually realized using a computer apparatus. FIG. 20 shows a configuration of a computer apparatus for realizing the power demand prediction apparatus of the present invention. For example, the communication unit 304, the input unit 305, the display unit 306, the CPU 301, the memory 302, the storage device 303, and the like are signal lines. It can be realized by being connected via 307.

  Each function in the power demand prediction apparatus of FIG. 1 is processed by software according to a program stored in the memory 302. The processing procedure in this case is shown in FIGS.

  FIG. 21 shows an operation flow when calculating the correction parameter. According to the processing flow of FIG. 21, when processing is started in processing step S101, social information feature amount J3 is first extracted in processing step S104. On the other hand, the power demand prediction error by time (K1-K2) is compared with the threshold value K3 in the processing step S102. In this case, the process proceeds to processing step S103. In the process of process step S103, the correction parameter is calculated using the social information feature amount J3 extracted in process step S104. After calculating the correction parameter, the process proceeds to processing step S105, and a series of processes is terminated.

  FIG. 22 shows an operation flow when the demand prediction is corrected using the calculated correction parameter. According to the processing flow of FIG. 22, when processing is started in processing step S201, demand prediction is first performed in processing step S202, and social information feature amount J3 is extracted in processing step S203. In processing step S204, the presence / absence of the social information feature J3 is determined. If not, the processing proceeds to processing step S206 to end the series of processing. If yes, the processing proceeds to processing in step S205. In the process of process step S205, the demand prediction is corrected, and thereafter, the process proceeds to process step S206 to end the series of processes.

1: social information feature quantity extraction means 2: correction parameter calculation means 3: prediction error determination means 4: demand prediction correction means 5: correction parameter holding means by feature quantity 100: power demand prediction device 200: power demand prediction means J1: time Various social information including information J2: Feature amount extraction method designation information J3 Social information feature amount with time tag K: Power demand information by time K1: Predicted value (time t)
K2: Actual value (time t)
K3: Time-specific power demand prediction error determination threshold K4: Time-based power demand prediction error determination result P1: Feature-specific correction parameter P2: Power demand prediction correction parameter P3: Log

Claims (15)

  Social information feature amount extraction means for detecting fluctuations of social information that are greater than or equal to a first threshold, prediction error determination means for detecting that the difference between prediction and actual power demand is greater than or equal to a second threshold, and the social information Changes in the first threshold value or more, and when the difference between the predicted and actual power demand is greater than or equal to the second threshold value, the correction parameter calculating means for calculating the correction parameter for the power demand and the correction parameter are stored. A power demand prediction apparatus comprising: a correction parameter holding means for performing and a power prediction correction means for outputting information used for power demand prediction correction. The power demand prediction device according to claim 1,
The power demand prediction apparatus characterized in that the social information and the prediction and performance of the power demand are obtained as time-specific information. A power demand prediction apparatus according to claim 1 or claim 2,
The information used for the electric power demand prediction correction includes log information as a basis for the correction. A power demand prediction apparatus according to any one of claims 1 to 3,
The power demand prediction apparatus, wherein the social information is sales of seasonal products at a store. The electric power demand prediction apparatus according to any one of claims 1 to 4,
The power demand prediction device, wherein the social information is information related to a television program. A power demand prediction apparatus according to any one of claims 1 to 5,
The power demand prediction apparatus, wherein the social information is an event such as an attraction. The electric power demand prediction apparatus according to any one of claims 1 to 6,
The social information is obtained from a transportation system. The power demand prediction apparatus according to any one of claims 1 to 7,
The power demand prediction apparatus, wherein the correction parameter holding means stores information on a power demand correction period and information on a power amount to be corrected for each social information. The power demand prediction apparatus according to any one of claims 1 to 8,
The power demand prediction apparatus, wherein the first threshold for the social information is set for each of a plurality of social information. A power demand prediction apparatus according to any one of claims 1 to 9,
The power demand forecasting apparatus, wherein the social information is information in which events and times related to energy consumption activities are tagged. The power demand prediction apparatus according to any one of claims 1 to 10,
The forecast and actual power demand are the demand forecast value and demand actual value for each time, and the second threshold is information to be compared with the same time difference between the demand forecast value and the actual demand value. A power demand forecasting device. The power demand prediction apparatus according to claim 11,
The demand forecast value for each time is predicted from at least a factor using the temperature, and the second threshold is determined based on a fluctuation error of the demand forecast value for each time. . A power demand prediction apparatus according to any one of claims 1 to 12,
The social information is tagged as time, and the calculation of the correction parameter is performed by correlating the social information with an event in which the difference between the demand forecast and the actual demand exceeds the threshold value, and a difference of a certain time. If there is a correlation, if there is a correlation, a function that explains an event in which the difference between the demand forecast and the actual demand exceeds a second threshold value is calculated by the social information and used as a correction parameter. . The power demand prediction apparatus according to claim 13,
As a log about the basis for correction, when social information is extracted and corrected based on a correction parameter for power demand prediction, the correlation between the social information and an event in which the difference between the demand prediction and the actual demand exceeds a second threshold is obtained. A power demand forecasting device characterized by obtaining.   When the social information fluctuates more than the first threshold and the difference between the forecast and actual power demand is more than the second threshold, a power demand correction parameter is calculated for each social information, and the power demand A method for predicting power demand characterized by correcting a forecast.

Images