JP2015139283A - Device, method and program for prediction of power demand peak - Google Patents

Abstract

PROBLEM TO BE SOLVED: To more accurately predict a peak value or peak time zone of power demand.SOLUTION: A device 1 for prediction of power demand peak comprises: a past data storage part 12 for storing past data including use amount of electricity for each time in a past day and feature information on the past day in association with each other; a feature information acquisition part 11 for acquiring feature information on a prediction target day; a past data extraction part 13 for extracting, from the past data storage part 12, one or more pieces of past data including feature information which is the same as or similar to the feature information on the prediction target day; a peak identification part 14 for identifying a power demand peak value or a power demand peak time zone with respect to each of the past data extracted by the past data extraction part 13; a prediction part 15 for calculating a predicted value of a power demand peak value or power demand peak time zone in the prediction target day, on the basis of the power demand peak value or power demand peak time zone of each past data; and an output part 16 for outputting the predicted value calculated by the prediction part 15.

Description

  The present invention relates to a power demand peak prediction device, a power demand peak prediction method, and a power demand peak prediction program that predict a peak value or peak time zone of power demand.

  In order to reduce the carbon use and improve the efficiency of energy use in the private sector, it is expected that distributed power sources such as photovoltaic power generation and fuel cells will be used effectively in multiple households such as apartment houses. For that purpose, it is important to statistically evaluate the fluctuation characteristics of the power / heat demand, and appropriately predict the magnitude and change of the demand based on the result (see Non-Patent Document 1 below). So far, the present inventors have developed a method for predicting a 30-minute value at 24 hours the next day for the total power demand of a plurality of households (see Non-Patent Documents 2 and 3 below).

  In the method described in Non-Patent Document 3 below, in order to take into account the long-period trend of power demand fluctuation, the day is divided into five time zones, and the time of the next day is determined based on the expected temperature of the next day. The average power demand is predicted. Moreover, in the said method, in order to consider the short cycle trend of a demand fluctuation, the ratio for every 30 minutes with respect to the average value of each time division is estimated. As a result, the average absolute error rate (% MAE) of daily peak demand prediction is reduced by 1 to 2% in the method described in Non-Patent Document 3 below compared to the method described in Non-Patent Document 2. Yes.

鄭, Iwafune; quantification of the habituation effect of power consumption of houses based on measured data, 29th Energy System / Economic / Environmental Conference, 14-5 (2013.3.1) Matsumoto, Kato, Suzuki; A study on prediction of power demand pattern for households in the next day, 29th Energy System / Economic / Environmental Conference, No. 14-3 (2013.3.1) Kato, Matsumoto, Suzuoki; A study on the prediction of the previous day of 30-minute total power demand for multiple households, 32nd Annual Meeting of the Society of Energy and Resources, No. 3-1 (20133.6)

  However, in the above method, paying attention only to the periodic trend of power demand fluctuation, the forecast value of power demand is calculated by taking the average of power demand data for multiple days with similar trend patterns. Yes. If the trend pattern as a whole is averaged in this way, the predicted value of the power demand peak will be estimated lower than the actual peak value because the peak value of the power demand for each day is accustomed. There is. That is, there is a problem that the peak value of power demand cannot be accurately predicted. Moreover, there is a similar problem with respect to the predicted time zone of the power demand peak.

  Therefore, an object of the present invention is to provide a power demand peak prediction device, a power demand peak prediction method, and a power demand peak prediction program that can predict a peak value or peak time zone of power demand more accurately.

  A power demand peak prediction apparatus according to the present invention includes a past data storage unit that stores past data in which electricity usage for each time in the past period and feature information indicating characteristics of the past period are associated, and a prediction target period. Feature information acquisition means for acquiring feature information indicating features, and one or more past data including feature information similar to the feature information of the prediction target period acquired by the feature information acquisition means or similar to a predetermined criterion, are stored as past data The past data extraction means extracted from the means, and for each past data extracted by the past data extraction means, the power demand peak value indicating the peak of electricity usage in the past period, or the electricity usage amount in the past period is a predetermined threshold or more The peak specifying means for specifying the power demand peak time period and the peak of each past data specified by the peak specifying means Or based on peak hours, comprising prediction means for calculating a predicted value of the power demand peak value or the power demand peak time zone of the prediction period, and output means for outputting a prediction value calculated by the prediction means.

  Moreover, the power demand peak prediction apparatus according to the present invention includes a feature information acquisition unit that acquires feature information indicating a characteristic of a prediction target period, and a feature information indicating a feature of the past period and the amount of electricity used for each time in the past period. One or more past data including feature information that is the same as or similar to the feature information of the prediction target period acquired by the feature information acquisition unit or similar to the feature information is extracted from the past data storage unit that stores past data associated with For the past data extraction means and the past data extracted by the past data extraction means, the power demand peak value indicating the peak of electricity usage in the past period or the power demand for which the electricity usage exceeds a predetermined threshold in the past period Peak specifying means for specifying the peak time zone, and the power demand peak value of each past data specified by the peak specifying means or A prediction means for calculating a power demand peak value in a prediction target period or a prediction value of a power demand peak time period based on a power demand peak time zone; and an output means for outputting a prediction value calculated by the prediction means. It may be a configuration.

  In the power demand peak prediction apparatus according to the present invention, past data including feature information that is the same as or similar to the feature information (for example, weather information) of the prediction target period is extracted by the past data extraction unit. That is, only the past data of the past period estimated to have a feature similar to the prediction target period is extracted. Further, the peak specifying means specifies the power demand peak value or the power demand peak time zone of each past data extracted by the past data extracting means. Then, the prediction means calculates the power demand peak value or the power demand peak time zone predicted value for the prediction target period based on the power demand peak value or power demand peak time zone of each past data. The predicted value calculated in this way is output by the output means. Thus, according to the power demand peak prediction apparatus according to the present invention, paying attention to the power demand peak value or the power demand peak time zone of the past data of the past period estimated to have characteristics similar to the prediction target period. It is possible to calculate the power demand peak value in the prediction target period or the power demand peak time predicted value. Therefore, the peak value or peak time zone of the power demand in the prediction target period can be predicted with higher accuracy compared to the case where the prediction is made by paying attention only to the periodic trend of the power demand fluctuation.

  In the power demand peak prediction apparatus, the feature information may be at least one of a maximum temperature, a day of the week, weather, humidity, sunshine duration, precipitation, cloud cover, and wind conditions. There seems to be a certain correlation between electricity demand and maximum temperature, day of the week, weather, humidity, sunshine duration, precipitation, cloud cover and wind conditions. Therefore, by using such feature information, it can be expected to improve the prediction accuracy of the peak value or peak time period of the power demand in the prediction target period. Further, when a plurality of such feature information is used in combination, the past data extraction unit can extract past data of a past period whose features are more similar to those of the prediction target period. Thereby, it can be expected that the prediction accuracy of the peak value or peak time of the power demand in the prediction target period is further improved.

  In the power demand peak prediction apparatus, the peak specifying means specifies at least two of the start time, end time, and length of the power demand peak time zone of each past data, and the prediction means is specified by the peak specifying means. In addition, by executing statistical processing based on at least two of the start time, end time, and length of the power demand peak time zone of each past data, the predicted value of the power demand peak time zone of the forecast period is calculated. Also good. For example, the result of performing statistical processing (for example, expected value, probability distribution calculation processing, etc.) for each of the start time and length of the power demand peak time period of each past data (for example, the value that maximizes the expected value and probability) Etc.) can be used as the predicted value of the start time and length of the power demand peak time period in the prediction target period. Thus, the power demand peak time zone of the prediction target period can be predicted easily and accurately by statistical processing based on at least two of the start time, end time and length of the power demand peak time zone of the past data.

  In the power demand peak prediction apparatus, the peak specifying unit specifies the power demand peak value of each past data, and the prediction unit executes statistical processing based on the power demand peak value of each past data specified by the peak specifying unit. Thus, the predicted value of the power demand peak value in the prediction target period may be calculated. For example, the result of statistical processing (for example, expected value or probability distribution calculation processing) for the peak power demand value of each past data (for example, the expected value, the value with the highest probability, etc.) It can be a predicted value of the peak value. As described above, the power demand peak value in the prediction target period can be predicted easily and accurately by the statistical processing based on the power demand peak value of the past data.

  In the power demand peak prediction apparatus, the characteristic information includes numerical information indicated by numerical values, the peak specifying means specifies the power demand peak value of each past data, and the predicting means is each specified by the peak specifying means. By calculating the regression equation based on the combination of the peak power demand value of the past data and the numerical information included in the past data, and substituting the numerical information of the forecast target period into the regression formula, the power demand peak in the forecast target period A predicted value of the value may be calculated. The numerical information is, for example, the maximum temperature. By substituting numerical information (for example, maximum temperature) of the prediction target period into the regression equation thus calculated, the power demand peak value of the prediction target period can be predicted easily and accurately.

  In the power demand peak prediction device, for each of a plurality of different extraction periods, the peak specifying means, the prediction based on the past data extracted by the past data extraction means from the past data including the past period in the extraction period And a power demand peak value or a power demand peak value for the forecast target period output by executing each process by the output means, and a power demand peak value or a power demand peak for the forecast target period. Predicted value / actual value input means for inputting an actual value in a time zone, and error calculation means for calculating an error between the actual value and each predicted value for each extraction period input by the predicted value / actual value input means Based on the error for each extraction period calculated by the error calculation means, the prediction of the power demand peak value or the power demand peak time period in the new prediction target period Extraction period determining means for determining a prediction extraction period used to calculate the power consumption, and the past data extracting means calculates a power demand peak value or a power demand peak time period predicted value for a new prediction target period. In this case, feature information that is the same as or similar to the feature information acquired by the feature information acquisition unit is selected from past data in which the past period is included in the prediction extraction period determined by the extraction period determination unit. One or more past data may be extracted.

  According to the power demand peak prediction apparatus, the predicted value obtained based on a plurality of different extraction periods and the measured value in the prediction target period are input by the predicted value / measured value input means. Then, the error calculation means calculates an error between the predicted value and the actual measurement value for each extraction period. Then, the extraction period determining means uses the extraction period for prediction used for calculating the power demand peak value of the new prediction target period or the prediction value of the power demand peak time zone based on the error (for example, extraction with the smallest error). Period etc.) is determined. On the other hand, the past data extraction unit extracts past data in the prediction extraction period when calculating the power demand peak value in the new prediction target period or the prediction value in the power demand peak time zone. Thereby, based on the past data in the extraction period for prediction, it becomes possible to predict the power demand peak value or the power demand peak time zone in the prediction target period more accurately.

  In the power demand peak prediction apparatus, the extraction period determining means determines a prediction extraction period for each time period in which the prediction target period is included in a predetermined cycle based on the error for each extraction period calculated by the error calculation means. The past data extraction means is used for prediction corresponding to a time when a new prediction target period is included in a predetermined cycle when calculating a power demand peak value of a new prediction target period or a prediction value of a power demand peak time zone. One or more pieces of past data including feature information that is the same as or similar to the feature information acquired by the feature information acquisition unit or similar to a predetermined standard may be extracted from past data that includes the past period in the extraction period.

  According to the power demand peak prediction apparatus, the extraction period determination unit determines the prediction extraction period for each period (for example, a specific month such as “April”) in a predetermined period (for example, a period such as “1 year”). Is done. Accordingly, it is possible to predict the power demand peak value or the power demand peak time zone of the prediction target period more accurately using an appropriate prediction extraction period in accordance with the time when the new prediction target period is included in the predetermined cycle. It becomes possible.

  By the way, the present invention can be described as the invention of the power demand peak prediction apparatus as described above, and as an invention of the power demand peak prediction method and the power demand peak prediction program executed by the power demand peak prediction apparatus as follows. Can also be described. Since these inventions are substantially the same inventions only in different categories, they exhibit the same operations and effects.

  That is, the power demand peak prediction method according to the present invention is a power demand comprising past data storage means for storing past data in which the amount of electricity used at each time in the past period and feature information indicating characteristics of the past period are associated with each other. A power demand peak prediction method executed by a peak prediction device, the feature information acquisition step for acquiring feature information indicating the characteristics of the prediction target period, and the feature information of the prediction target period acquired in the feature information acquisition step Alternatively, a past data extraction step for extracting one or more past data including feature information similar to each other in a predetermined standard from past data storage means, and for each past data extracted in the past data extraction step, Peak electricity demand peak value or electricity usage in the past period exceeds a predetermined threshold Prediction of power demand peak value or power demand peak time zone of forecast period based on peak specification step for specifying power demand peak time zone and peak value or peak time zone of each past data specified in peak specification step A prediction step of calculating a value, and an output step of outputting the prediction value calculated in the prediction step.

  In addition, the power demand peak prediction program according to the present invention includes a past data storage unit that stores past data in which a computer is associated with feature information indicating characteristics of the past period and the amount of electricity used at each time in the past period. Feature information acquisition means for acquiring feature information indicating the characteristics of the prediction target period, and one or more past information including feature information that is the same as or similar to the feature information of the prediction target period acquired by the feature information acquisition means For past data extracted by past data extraction means for extracting data from past data storage means, and for each past data extracted by the past data extraction means, the electric power consumption peak value indicating the peak of electric usage in the past period or the electric usage in the past period A peak identification unit that identifies a peak time period for power demand that exceeds a predetermined threshold, and a peak identification unit Based on the peak value or peak time zone of each past data, a prediction means for calculating a power demand peak value or a power demand peak time zone prediction value for the prediction target period, and a prediction value calculated by the prediction means are output. Function as output means.

  According to the present invention, the peak value or peak time zone of power demand can be predicted with higher accuracy.

It is a block diagram which shows the function structure of the electric power demand peak prediction apparatus which concerns on 1st Embodiment. (A) is a figure which shows the example of the estimated temperature data memorize | stored in the weather forecast database, (b) is a figure which shows the example (highest temperature) of the characteristic information of the prediction object period acquired by the characteristic information acquisition part It is. (A) is a figure which shows the example of the past data memorize | stored in the past data storage part, (b) is a figure which shows the example of the past data extracted by the past data extraction part. It is a figure for demonstrating the process which extracts past data using several feature information. It is a figure which shows the electric power demand peak value and electric power demand peak time slot | zone in past data. It is a figure which shows the example of the starting point of an electric power demand peak time slot | zone for every past data, an end point, and length. It is a figure which shows the example of the electric power demand peak value for every past data. It is a figure which shows the regression formula of an electric power demand peak value and maximum temperature. It is a flowchart for demonstrating operation | movement in case an electric power demand peak prediction apparatus estimates the electric power demand peak time zone of a prediction object day. It is a flowchart for demonstrating operation | movement in case an electric power demand peak prediction apparatus estimates the electric power demand peak value on the prediction object day. It is a flowchart for demonstrating operation | movement in case an electric power demand peak prediction apparatus estimates the electric power demand peak value of a prediction object day based on a regression equation. It is a figure which shows the module structure of an electric power demand peak prediction program. It is a figure which shows the generation frequency according to magnitude | size of an electric power demand peak value. It is a figure which shows the probability that the time slot | zone for every 30 minutes will be contained in an electric power demand peak time slot | zone. It is a figure which shows the comparison with the predicted value and actual value of an electric power demand peak value. It is a figure which shows the relationship between the similar range of the highest temperature at the time of extracting past data, and the prediction error of a power demand peak value. It is a figure which shows the generation frequency according to the magnitude | size of the prediction error of an electric power demand peak time slot | zone. It is a figure which shows the prediction error regarding the expected value of an electric power demand peak time slot | zone. It is a figure which shows the relationship between the daily minimum temperature, the afternoon peak demand, and the morning peak demand. It is a block diagram which shows the function structure of the electric power demand peak prediction apparatus which concerns on 2nd Embodiment.

  An embodiment of a power demand peak prediction apparatus, a power demand peak prediction method, and a power demand peak prediction program according to the present invention will be described with reference to the drawings. If possible, the same parts are denoted by the same reference numerals, and redundant description is omitted.

[First Embodiment]
FIG. 1 shows a functional configuration of a power demand peak prediction apparatus according to the first embodiment of the present invention. The power demand peak prediction apparatus 1 according to the present embodiment is an apparatus for predicting a power demand peak (power demand peak value or power demand peak time zone) on a prediction target date (prediction target period). More specifically, the power demand peak prediction device 1 includes the day of the week, the weather, the maximum temperature, the humidity, the sunshine time, the precipitation, among the data (past data) indicating the past power usage record value in one or a plurality of households. Past data associated with a past date (past period) in which feature information (explanatory variables) such as cloud amount and wind condition coincides with the prediction target date is extracted. The power demand peak prediction device 1 predicts the power demand peak on the prediction target day based on the power demand peak specified for each past data extracted in this way. In order to execute such a prediction process, the power demand peak prediction device 1 includes a feature information acquisition unit 11, a past data storage unit 12, a past data extraction unit 13, a peak identification unit 14, a prediction unit 15, and an output unit 16. Is provided. In the present embodiment, the prediction target period and the past period are both “one day”, but the prediction target period and the past period are not limited to this, for example, “half day (am / pm)” or “1 Week "etc. may be sufficient.

  The power demand peak prediction device 1 physically includes one or more CPUs, RAMs and ROMs as main storage devices, input devices such as keyboards and mice as input devices, output devices such as displays, The computer system includes a communication module that is a data transmission / reception device such as a network card, and an auxiliary storage device such as a hard disk drive and a semiconductor memory. The power demand peak prediction device 1 may be physically configured as a single device, or may be configured by a plurality of devices that operate in cooperation with each other.

  The function of the power demand peak prediction device 1 is to operate the input device, the output device, and the communication module under the control of the CPU by, for example, reading predetermined computer software on the hardware such as the CPU and RAM. At the same time, it is realized by reading and writing data in the RAM and the auxiliary storage device. Hereinafter, each function of the power demand peak prediction device 1 will be described.

  The feature information acquisition unit 11 is a feature information acquisition unit that acquires feature information indicating the feature of the prediction target day. Here, the prediction target date is set in advance by an operator or the like, for example. The feature information acquisition unit 11 may specify the next day (the date of the next day) by referring to the hardware system time of the power demand peak prediction device 1, for example, and may acquire the next day as the prediction target day. The characteristic information is information indicating characteristics of the prediction target day (an explanatory variable for explaining the characteristics of the prediction target day), and is, for example, a day of the week, weather, maximum temperature, and the like. The feature information acquisition unit 11 acquires the feature information of the prediction target date from a system different from the power demand peak prediction device 1 (for example, the weather forecast database 2 disclosed by the Japan Meteorological Agency), for example. As shown in FIG. 2A, the weather forecast database 2 stores, for example, predicted temperature data indicating the predicted temperature every 30 minutes on the prediction target day for each region (area) and each prediction target day. Here, FIG. 2A shows predicted temperature data on December 6, 2013. FIG.

  For example, let us consider a case where the prediction target date is December 6, 2013. In this case, the feature information acquisition unit 11 refers to the weather forecast database 2 by designating the prediction target date and the prediction target area, thereby predicting the predicted prediction date and area predicted temperature data (FIG. 2 (a ) See). And the feature information acquisition part 11 acquires the maximum value (maximum temperature) of the predicted temperature of the acquired predicted temperature data as the feature information of the prediction target day. In this example, the feature information acquisition unit 11 extracts “7.0 ° C.” at 14:30 as the highest temperature (feature information) on the prediction target day (see FIG. 2B).

  The above-mentioned weather forecast database 2 stores data in which the prediction target date is associated with information indicating the prediction value of the prediction target day (prediction weather), humidity, sunshine duration, precipitation, cloud cover, wind conditions, and the like. May be. In this case, the feature information acquisition unit 11 can acquire the weather, humidity, sunshine duration, precipitation, cloud cover, wind conditions, and the like of the prediction target day by referring to the weather forecast database 2. The feature information acquisition unit 11 may acquire the day of the prediction target day by inputting the date of the prediction target day into a predetermined program that inputs the date and outputs the day of the week corresponding to the date, for example. Such a program may be provided as one function of the feature information acquisition unit 11 or may be another function available from the feature information acquisition unit 11. Note that the method by which the feature information acquisition unit 11 acquires the feature information of the prediction target date is not limited to the above-described method. For example, the feature information acquisition unit 11 may acquire the feature information (day of the week, weather, maximum temperature, etc.) of the prediction target day by executing another system or a predetermined program as described above, or by an operator or the like. You may acquire the day of the week, the weather, the highest temperature, etc. that have been input. In addition, the feature information acquisition unit 11 refers to an external system that stores calendar information in which a specific date and information on attributes of the date (such as holidays and Bon festival) are associated with each other, thereby predicting the prediction target date. Attributes may be input as feature information.

  The past data storage unit 12 is a past data storage unit that stores past data in which the amount of electricity used for each time on the past day is associated with feature information indicating characteristics of the past day. As shown in FIG. 3A, the past data storage unit 12 includes, for example, the date of the past date, the total electricity usage (for example, in units of kilowatts “kWh”) for one or more households on the past day, In addition, past data in which the highest temperature (feature information) of the past day is associated is stored for each past day. In this example, past data of “November 25, 2013” and “November 26, 2013” are stored. Here, the past data stored in the past data storage unit 12 may be divided for each region (area) and for each building type (detached house, apartment house, etc.). Further, the past data stored in the past data storage unit 12 may be associated with a plurality of pieces of feature information (here, the maximum temperature and the day of the week as an example) as shown in FIG. The past data stored in the past data storage unit 12 is stored in advance by data input by an operator or the like, for example. The past data stored in the past data storage unit 12 may be added sequentially by being periodically input by an operator or the like. In addition, the past data is periodically transmitted to the past data storage unit 12, for example, data related to the power consumption automatically measured by a power meter such as a building or a data measuring device (data logger). Thus, it may be stored in the past data storage unit 12.

  The past data extraction unit 13 extracts from the past data storage unit 12 one or more past data including feature information that is the same as or similar to the feature information of the prediction target date acquired by the feature information acquisition unit 11. Data extraction means. That is, the past data extraction unit 13 extracts past data relating to past days (corresponding days) that are considered to have characteristics similar to the prediction target days. The past data extraction unit 13 may extract past data whose past date is included in a past predetermined period (for example, the past 30 days, the past 90 days, etc.) based on the prediction target date, for example. The predetermined period is predetermined by an operator or the like, for example. In addition, the past data extraction unit 13 may extract past data according to a prediction target area or a building type preset by an operator or the like together with the prediction target date, for example. Thereby, the past data for predicting a power demand peak more accurately according to the area and the building type can be extracted.

  For example, when using a variable that changes continuously, such as maximum temperature, as feature information, the past data extraction unit 13 extracts feature information that falls within a predetermined range from the feature information of the prediction target day. You may extract the past data containing. That is, the past data extraction unit 13 determines that the feature information included within the predetermined range from the feature information of the prediction target day is the same or similar feature information to the feature information of the prediction target day. May be extracted. For example, when the maximum temperature (feature information) on the prediction target day is 7.0 ° C. and the predetermined range is preset as ± 0.5 ° C., the past data extraction unit 13 determines that the maximum temperature is Past data that is 6.5 ° C. or higher and 7.5 ° C. or lower may be extracted from the past data storage unit 12. As a result, as shown in FIG. 3B, for example, past data with the maximum temperatures of “7.4 ° C.” and “6.9 ° C.” is extracted.

  As shown in FIG. 4A, when the past data storage unit 12 stores past data associated with a plurality of feature information (maximum temperature, day of the week), the past data extraction unit 13 May extract past data using a plurality of feature information. For example, consider a case where the highest temperature on the prediction target day is 7.0 ° C. and the day of the prediction target day is Thursday. In this case, the past data extraction unit 13 receives from the feature information acquisition unit 11 the maximum temperature and day of the week to be predicted acquired by the feature information acquisition unit 11. Subsequently, the past data extraction unit 13 extracts past data (see FIG. 4B) having a maximum temperature of 6.5 ° C. or more and 7.5 ° C. or less from the past data storage unit 12, and the prediction target date. Past data (see FIG. 4C) for the same Thursday as the day of the week is extracted. The past data extraction unit 13 matches the past data extracted based on the maximum temperature in this way with the past data extracted based on the day of the week, and extracts past data (see FIG. 4D) included in both. Also good. In other words, the past data extraction unit 13 may extract past data corresponding to a past date that is the same as the prediction target day or similar in a predetermined criterion for both the day of the week and the highest temperature.

  Here, the distribution of the power demand peak value or the power demand peak time zone of the extracted past data is the actual value of the forecast target day (the actual measurement of the power demand peak value or the power demand peak time zone actually measured on the forecast target date). Value) and the distribution width is narrower, the prediction accuracy is considered to be higher. By adjusting the combination of feature information used for extracting past data for each prediction item (power demand peak value, power demand peak time zone), it is considered possible to construct a distribution with high prediction accuracy. Therefore, by using a combination of a plurality of feature information as described above, the past data extraction unit 13 can extract past data of a past date that is more similar in characteristics to the prediction target date. Thereby, it can be expected that the prediction accuracy of the power demand peak on the prediction target day by the processing of the peak specifying unit 14 and the prediction unit 15 described later is further improved.

  Further, the past data extraction unit 13 may extract the past data extracted for each feature information (see FIG. 4B and FIG. 4C) as past data. In other words, the past data extraction unit 13 may extract past data corresponding to a past date that is the same as the prediction target day or similar in a predetermined standard at any day of the week and highest temperature. In order to appropriately execute the processing of the peak specifying unit 14 and the prediction unit 15 described later, it is necessary that a past number of past data is extracted by the past data extracting unit 13. Accordingly, by extracting past data as described above, it is possible to more reliably extract a certain number of past data.

  The peak specifying unit 14 is a peak specifying unit that specifies a power demand peak value or a power demand peak time zone for each past data extracted by the past data extracting unit 13. FIG. 5 is a graph showing the amount of electricity used every 30 minutes as a bar graph for a certain past date, with the horizontal axis representing time and the vertical axis representing electricity usage. As shown in FIG. 5, the power demand peak value P is a value indicating the peak of electricity usage in the past day. The power demand peak time zone T is a time zone in which the amount of electricity used is equal to or greater than a predetermined threshold value Dh in the past day. More specifically, the power demand peak time zone T indicates a time zone including the time when the power demand peak value P is taken and the amount of electricity used is continuous at a threshold value Dh or more. Here, the threshold value Dh can be arbitrarily determined in advance by an operator or the like. For example, the threshold value Dh is 1.0 times to 1.3 times the average value Da on the basis of the average daily demand (average value of electricity usage) Da, for example. It can be set to double range.

  The prediction unit 15 calculates a power demand peak value or a power demand peak time zone prediction value on the prediction target day based on the power demand peak value or the power demand peak time zone of each past data specified by the peak specifying unit 14. It is a prediction means to do. Hereinafter, a specific example of a series of processes by the peak identification unit 14 and the prediction unit 15 will be described.

(First example)
First, the first example will be described. As shown in FIG.5 and FIG.6, the peak specific | specification part 14 is at least 2 of the start point time T1, end point time T2, and length (end point time T2-start point time T1) of the electric power demand peak time zone T of each past data. Calculate (specify) one. The prediction unit 15 performs statistical processing based on at least two of the start point time, the end point time, and the length of the power demand peak time zone of each past data specified by the peak specifying unit 14, thereby Calculate the predicted value for the peak power demand period. Here, the reason for executing the statistical processing based on at least two of the start point time, end point time, and length of the power demand peak time zone is that the start point time, end time, and length of the power demand peak time zone This is because if at least two are specified, the power demand peak time zone is uniquely specified.

  Specifically, the prediction unit 15 determines, for example, the start time and length of the power demand peak time zone based on the start time and length of the power demand peak time zone of each past data specified by the peak specification unit 14. Calculate each expected value. The expected value is calculated by the average value of the actual measurement values (start time and length) of each past data. The prediction unit 15 calculates the expected value of the start point time and the length calculated in this way as the predicted value (the predicted value of the start point time and the length) of the power demand peak time zone on the prediction target day.

  Here, the prediction unit 15 calculates the start point time based on the past data extracted by the past data extraction unit 13 instead of calculating the expected value of the start point time and the length of the power demand peak time period of each past data. And the probability distribution of each length may be calculated. The probability distribution here means the occurrence frequency (probability) for each predetermined range. That is, the probability distribution here is a value obtained by dividing the number of past data including the start time in each range by the total number of extracted past data. For example, for the start time, the prediction unit 15 sets a range (for example, “12: 0 to 12:30”) for each predetermined time interval (for example, “30 minutes”), and the past includes the start time in each range. By dividing the number of data by the total number of extracted past data, a probability distribution (frequency distribution (occurrence frequency in each range)) at the start point time can be calculated. The probability distribution can be calculated for the length by the same method. As a result, the prediction unit 15 determines the start point time and length range (or values representing the range) at which the occurrence frequency (probability) is maximum as the start point time and length of the power demand peak time zone on the prediction target day. Can be calculated as a predicted value.

  As described above, the predicting unit 15 executes statistical processing (expected value, probability distribution calculation processing, etc.) based on at least two of the start point time, the end point time, and the length of the power demand peak time period of the past data. The power demand peak time zone of the prediction target date can be predicted easily and accurately.

(Second example)
Subsequently, a second example will be described. As shown in FIG. 5, the peak specifying unit 14 specifies the power demand peak value P of each past data. Specifically, as shown in FIG. 7, the peak specifying unit 14 specifies the maximum value (peak value) from the amount of electricity used every 30 minutes for each past data (that is, each past date). And the prediction part 15 calculates the predicted value of the power demand peak value of a prediction object day by performing the statistical process based on the power demand peak value of each past data specified by the peak specific | specification part 14. FIG. More specifically, the prediction unit 15 may calculate an expected value (average value) of the power demand peak value of each past data, and calculate the expected value as a predicted value of the power demand peak value on the prediction target day. it can. In addition, the prediction unit 15 sets a range (for example, “7 kW to 8 kW”) for each predetermined amount of electricity usage (for example, “1 kW”), for example, and past data including a power demand peak value in each range The probability distribution (frequency distribution) of the power demand peak value is calculated by dividing the number by the total number of extracted past data, and the power demand peak value with the highest occurrence frequency is predicted for the power demand peak value on the prediction target day It may be calculated as a value.

  In this way, the prediction unit 15 executes statistical processing based on the peak power demand value of past data (calculation processing of expected value and probability distribution (frequency distribution), etc.), so that power demand on the prediction target day can be easily and accurately. Peak value can be predicted.

(Third example)
Subsequently, a third example will be described. In this example, the feature information includes numerical information indicated by numerical values (here, “maximum temperature” as an example). Moreover, the peak specific | specification part 14 specifies the electric power demand peak value of each past data by the method similar to a 2nd example. And as shown in FIG. 8, the prediction part 15 is based on the set A of the electric power demand peak value of each past data specified by the peak specification part 14, and the highest temperature (numerical information) contained in the said past data. The regression equation F is calculated. More specifically, the prediction unit 15 calculates the regression equation F by using a known method such as a least square method for the set A of the peak power demand value and the maximum temperature of each past data. Here, the regression equation F may be linear (linear equation) or a curve equation as shown in FIG. The prediction unit 15 calculates the predicted value Pe of the power demand peak value of the prediction target day by substituting the maximum temperature Te of the prediction target day into the regression equation F calculated in this way.

  In this way, the prediction unit 15 calculates the regression equation F, and substitutes the maximum temperature Te for the prediction target day into the calculated regression equation F, so that the power demand peak value (prediction value) of the prediction target day can be simply and accurately. Pe) can be predicted. For example, in a period in which the maximum temperature increases (or decreases) every day at the change of season, when the maximum temperature is used as characteristic information, the maximum temperature is lower (or higher) than the prediction target day by the past data extraction unit 13. ) There is a high possibility that a lot of past data for the past day will be extracted. Then, there is a possibility that sufficient prediction accuracy cannot be obtained in the second example in which the predicted value of the power demand peak value of the prediction target day is calculated by the statistical processing of the power demand peak value of the past data. In other words, the prediction error may increase due to prediction based on past data of the past day that is biased in one direction (the direction in which the maximum temperature is low or the direction in which the maximum temperature is high) based on the maximum temperature on the prediction target day. high. On the other hand, when predicting based on the regression equation as in the third example, a value close to the power demand peak value on the past day when the maximum temperature close to the maximum temperature on the prediction target day is observed is obtained as the predicted value. Probability is high. Therefore, the prediction of the power demand peak value according to the third example is considered to be particularly effective at the turn of the season as described above. Specifically, when the temperature rises, for example, in early spring, a high temperature such as 15 ° C. may be predicted for the prediction date even if the temperature has been around 10 ° C. so far. In this case, even if there is no past similar date (past date similar to the prediction target date) in the first place, or the temperature range of the past similar date is ± 5 ° C, only data around -5 ° C is collected and distributed. May cause problems such as creating Therefore, it is expected that such a problem can be solved and the prediction accuracy can be improved by estimating the data around the predicted temperature on the prediction target day using the regression equation. The feature information used to extract past data by the past data extraction unit 13 and the feature information (numerical information) used to calculate the regression equation by the prediction unit 15 are the same as described above (both are the highest temperature). May be different or different. For example, the day of the week may be used for extracting past data, and the highest temperature may be used for calculating the regression equation.

  The output unit 16 is an output unit that outputs the predicted value calculated by the prediction unit 15. Although the output method of the predicted value by the output part 16 is not specifically limited, For example, the output part 16 may output a predicted value by displaying a predicted value as a prediction result on the display mentioned above. In addition, the output unit 16 may output the predicted value by transmitting the predicted value to another device that performs processing using the predicted value via the communication module described above.

  In the power demand peak prediction apparatus 1 according to the present embodiment, the past data extraction unit 13 extracts past data including feature information that is the same as or similar to the feature information (for example, maximum temperature) of the prediction target day. That is, only the past data of the past date estimated to have a feature similar to the prediction target date is extracted. Further, the peak specifying unit 14 specifies the power demand peak value or the power demand peak time zone of each past data extracted by the past data extracting unit 13. And the prediction part 15 calculates the predicted value of the power demand peak value or the power demand peak time zone of the prediction target day based on the power demand peak value or the power demand peak time zone of each past data. The predicted value calculated in this way is output by the output unit 16.

  As mentioned above, according to the power demand peak prediction apparatus 1 which concerns on this embodiment, paying attention to the power demand peak value or the power demand peak time zone of the past data of the past date estimated to have a characteristic similar to the prediction target day. Thus, it is possible to calculate the power demand peak value or the power demand peak time predicted value on the prediction target day. Therefore, the peak value or the peak time zone of the power demand on the prediction target day can be predicted with higher accuracy than in the case where the prediction is made by paying attention only to the periodic trend of the power demand fluctuation.

  Then, operation | movement of the electric power demand peak prediction apparatus 1 containing the electric power demand peak prediction method which concerns on one Embodiment of this invention is demonstrated using FIGS. 9-11. 9 to 11 are diagrams showing processing flows corresponding to the first to third examples described above.

  A processing flow corresponding to the first example will be described with reference to FIG. First, the feature information acquisition unit 11 acquires the feature information (maximum temperature, etc.) of the prediction target day (step S11, feature information acquisition step). Subsequently, the past data extraction unit 13 extracts one or more past data including feature information that is the same as or similar to the feature information of the prediction target day from the past data storage unit 12 (step S12, past data). Extraction step). Subsequently, the peak specifying unit 14 specifies at least two of the start point time, the end point time, and the length (here, the start point time and the length as an example) of the power demand peak time zone of each past data (step S13). , Peak identification step). Subsequently, based on the start point time and length of the power demand peak time zone of each past data specified by the peak specifying unit 14 by the prediction unit 15, the start time and length of the power demand peak time zone of the prediction target day. Is calculated (step S14, prediction step). Specifically, as described above, the prediction unit 15 calculates the predicted value by calculating an expected value (or probability distribution (frequency distribution)) of the start time and length of each past data. Subsequently, the output unit 16 outputs the predicted value calculated by the prediction unit 15 (step S15, output step).

  Next, a processing flow corresponding to the second example will be described with reference to FIG. First, the feature information acquisition unit 11 acquires the feature information (maximum temperature, etc.) of the prediction target day (step S21, feature information acquisition step). Subsequently, the past data extraction unit 13 extracts one or more past data including feature information that is the same as or similar to the feature information of the prediction target day from the past data storage unit 12 (step S22, past data). Extraction step). Subsequently, the peak specifying unit 14 specifies the power demand peak value of each past data (step S23, peak specifying step). Subsequently, based on the power demand peak value of each past data specified by the peak specifying unit 14, the prediction unit 15 calculates the predicted value of the power demand peak value on the prediction target day (step S24, prediction step). . Specifically, as described above, the prediction unit 15 calculates the predicted value by calculating the expected value (or probability distribution (frequency distribution)) of the power demand peak value of each past data. Subsequently, the output unit 16 outputs the predicted value calculated by the prediction unit 15 (step S25, output step).

  Next, a processing flow corresponding to the third example will be described with reference to FIG. First, the feature information acquisition unit 11 acquires the feature information (maximum temperature, etc.) of the prediction target day (step S31, feature information acquisition step). Subsequently, the past data extraction unit 13 extracts one or more past data including feature information that is the same as or similar to the feature information of the prediction target day from the past data storage unit 12 (step S32, past data). Extraction step). Subsequently, the peak specifying unit 14 specifies the power demand peak value of each past data (step S33, peak specifying step). Subsequently, the prediction unit 15 calculates a regression equation based on the combination of the peak power demand value and the maximum temperature of each past data specified by the peak specifying unit 14 (step S34). Subsequently, the prediction unit 15 calculates the predicted value of the power demand peak value on the prediction target day by substituting the maximum temperature of the prediction target day into the regression equation (step S35). Here, the process step which combined step S34 and step S35 comprises a prediction step. Subsequently, the output unit 16 outputs the predicted value calculated by the prediction unit 15 (step S36, output step).

  Then, the electric power demand peak prediction program P1 which concerns on one Embodiment of this invention is demonstrated using FIG. The power demand peak prediction program P <b> 1 is a program for causing a computer to function as the power demand peak prediction device 1. FIG. 12 is a block diagram showing modules of the power demand peak prediction program P1. As shown in FIG. 12, the power demand peak prediction program P1 includes a feature information acquisition module P11, a past data storage module P12, a past data extraction module P13, a peak identification module P14, a prediction module P15, and an output module P16. The functions realized by executing the feature information acquisition module P11, the past data storage module P12, the past data extraction module P13, the peak identification module P14, the prediction module P15, and the output module P16 are respectively the power demand peak predictions described above. The function is the same as that of the feature information acquisition unit 11, the past data storage unit 12, the past data extraction unit 13, the peak identification unit 14, the prediction unit 15, and the output unit 16 of the device 1.

  The power demand peak prediction program P1 configured as described above is stored in a recording medium such as a CD-ROM, DVD, and HD, and is executed by a computer used as the power demand peak prediction device 1. Specifically, the computer includes a recording medium reading unit such as a CD-ROM drive, a DVD drive, and an HDD. When the recording medium is set in the recording medium reading unit, the computer can access the power demand peak prediction program P1 stored in the recording medium from the recording medium reading unit. Then, by causing the computer to execute the power demand peak prediction program P1, the computer can be operated as the power demand peak prediction device 1.

  The power demand peak prediction program P1 may be provided via a network as a data signal superimposed on a carrier wave.

  Subsequently, an example of the power demand peak prediction device 1 according to the first embodiment will be described with reference to FIGS. 13 to 19 and an evaluation result of the prediction result will be described.

  First, feature information used by the past data extraction unit 13 to extract past data will be described. In this embodiment, the day of the week, the weather, and the maximum temperature are assumed as the characteristic information. When using the day of the week as the feature information, the two types of “weekdays” and “Saturdays, Sundays, and holidays” are used as values indicating the days of the week so that the number of past data extracted by the past data extraction unit 13 is a certain number or more. . And the past data extraction part 13 shall extract the past data corresponding to the said past day, when the day of prediction day and the day of the past day are the same classification | category.

  When the weather is used as the feature information, the weather on the prediction target date and the past day is determined as follows. If at least one rain is predicted in the weather forecast for each time zone from 3:00 to 12:00 on the prediction target day, or if the precipitation probability is 50% or more, the weather on the prediction target day is set to rain. In other cases, the forecasted weather (for example, “sunny”, “cloudy”, etc.) for the weather forecast for each time zone targeting 9 to 12:00 on the forecast target day was used as the forecast target day weather. For past data stored in the past data storage unit 12, if any of the past days at 9 o'clock, 15 o'clock, or 21 o'clock (actually observed weather) is raining, the past day's weather is displayed. It was raining. In other cases, the weather at 15:00 of the past day (for example, “sunny”, “cloudy”, etc.) was used as the weather of the past day. And the past data extraction part 13 shall extract the past data corresponding to the said past day, when the weather of the prediction object day and the weather of the past day are the same.

  When using the maximum temperature as the feature information, the past data extraction unit 13 determines that the maximum temperature (measured value) of the past day is within ± α ° C. with respect to the maximum temperature (forecast value) of the prediction target day. The past data corresponding to the past day is extracted. Here, it is known that there is a certain correlation between the maximum temperature and the power demand. Therefore, by reducing α, it is expected that the past day whose power demand peak value is close to the prediction target date is extracted. However, if α is reduced, the number of days in the past extracted is reduced. Therefore, the sensitivity was analyzed by changing α in 1 ° C. units within a range of 1 to 10 ° C. The result of the sensitivity analysis will be described later.

  In this example, the prediction accuracy of the prediction result was evaluated for the case where the above-described feature information (day of the week, weather, maximum temperature) was used independently. The past data extraction unit 13 extracts the corresponding past data from the past data associated with the past days included within the past 90 days with the prediction target date as a reference.

(About usage data)
In this example, the 30-minute average value (hereinafter simply referred to as “electricity consumption”) of the electricity usage (total power demand) of 16 households living in an apartment house in Yokohama was used as the prediction target. The period of the past date (hereinafter referred to as “examination target period”) from which past data is extracted by the past data extraction unit 13 is from August 2012 to July 2013. In the past data during the examination period, the maximum value of electricity usage was 18.9 kW, the minimum value was 1.6 kW, the average value was 5.2 kW, and the standard deviation was 2.6 kW. The feature information acquisition unit 11 acquires information on the predicted maximum temperature and the weather forecast described in the evening edition of the Asahi Shimbun Kanto edition the day before the prediction target date as the maximum temperature and weather of the prediction target date by input by an operator or the like. It was supposed to be. In addition, when the day before the prediction target date is a closed day of the evening publication, the feature information acquisition unit 11 uses the predicted maximum temperature and the weather forecast information described in the morning newspaper on the prediction target day as the highest prediction target date. Acquired as temperature and weather. In addition, the prediction target date in which the evening publication on the day before the prediction target date and the morning edition on the prediction target day are both closed, was excluded from the prediction target. Moreover, the actual temperature (actual value) related to the maximum temperature and the weather on the prediction target day was confirmed by information disclosed on the JMA website.

(Characteristics of peak power demand)
FIG. 13 shows the frequency of occurrence of each power demand peak value in the examination target period. FIG. 13A shows the frequency of occurrence of each power demand peak value in the entire past data in the period to be examined. As shown to Fig.13 (a), in the past data of the examination object period, the electric power demand peak value varies widely in the range of 5.4 kW to 18.9 kW depending on the day. The average value of the power demand peak value in the entire past data in the examination period was 9.6 kW, and the standard deviation was 2.3 kW. These average value and standard deviation are considered to be the expected value of the daily power demand peak value and its mean square error. Therefore, if a distribution having a smaller variation than that of FIG. 13A can be created from past data extracted by the past data extraction unit 13 using the above-described feature information (day of the week, weather, maximum temperature), the extracted data is extracted. The prediction based on the past data can be expected to improve the prediction accuracy of the power demand peak value on the prediction target day.

  13 (b), 13 (c), and 13 (d) are extracted using feature information that the maximum temperature is 25 to 35 ° C., the weather is “rain”, and the day of the week is “Saturdays, Sundays, and holidays”. In addition, the frequency of occurrence of power demand peak values obtained from past data is shown. As shown in FIG. 13 (b), the power demand peak value of the past data extracted using the maximum temperature as the characteristic information has a high probability of occurring in the range of 8 to 11 kW, and is more variable than FIG. 13 (a). It became a small distribution. Moreover, as shown in FIG.13 (c), the average value of the power demand peak value of the past data extracted using the weather as characteristic information was 8.4 kW. Moreover, as shown in FIG.13 (d), the average value of the power demand peak value of the past data extracted using the day of the week as the characteristic information was 9.4 kW. Thus, the average value of the power demand peak value of the past data extracted using the weather or the day of the week as the characteristic information is not much different from the average value of the power demand peak value of the entire past data in the examination target period, 9.6 kW, The variation in peak power demand was similar to the variation in the entire past data during the study period. From these results, it can be seen that, in the case of the present embodiment, the prediction accuracy of the power demand peak value on the prediction target day can be expected to be improved by using the maximum temperature as the feature information.

(Characteristics of peak demand hours)
In FIG. 14A, when the threshold value Dh is equal to the average value Da, the probability that the power demand peak time zone includes each time zone in which one day is divided every 30 minutes with respect to the entire past data in the examination target period. (Frequency). As shown to Fig.14 (a), the probability (frequency) that the start time of an electric power demand peak time zone will be 16: 30-17: 00, and end time will be 21: 0 to 22:30 was high. As described above, on most days, the peak power demand period occurs in the evening, but the day when the peak power demand period occurs in the morning was 33 days (13.9% of the entire study period). In addition, the probability (frequency) that the length of the power demand peak time zone is 4 to 6 hours is high regardless of the occurrence time of the power demand peak time zone.

  14 (b), 14 (c), and 14 (d) are extracted using feature information that the maximum temperature is 25 to 35 ° C., the weather is “rainy”, and the day of the week is “Saturdays, Sundays, and holidays”. In addition, regarding the past data, the probability (frequency) that each time zone obtained by dividing one day every 30 minutes is included in the power demand peak time zone is shown. As shown in these figures, the probability (frequency) that each time zone is included in the power demand peak time zone is the same regardless of the feature information used to extract the past data.

  In addition, the probability distribution of the start time and end time of the power demand peak time zone in the entire study period, that is, the occurrence frequency of the start time and end time in each time zone divided into 30 minutes every day was calculated. As a result, the probability (frequency) that the starting point time was 16:00 to 17:00 was high. On the other hand, the probability (frequency) that the end point time is 21:30 to 23:30 is high, and it was confirmed that the variation depending on the day is slightly larger than the start point time. Further, it was confirmed that when the threshold value Dh for defining the power demand peak time period is increased, the variation in the end point time is increased while the variation in the start point time is relatively small. Therefore, in the case of the past data used in the present embodiment, it is considered that the start time of the power demand peak time zone has very little difference depending on the day throughout the year. However, as described above, during the 33 days of the year, the time when peak demand occurs (time when the power demand peak value is taken) occurs in the morning. Therefore, the power demand peak prediction device 1 may be configured to more appropriately predict the power demand peak generated in the morning by dividing the prediction target period into half-day units.

(Forecast results for peak power demand)
The case of predicting the power demand peak value on the prediction target day based on the expected value of the power demand peak value of the past data among the methods shown in the above “second example” is expressed by the following formula (1). The prediction accuracy was evaluated using the average absolute error rate (% MAE).

In the above formula (1), d _max indicates the total number of days of the evaluation target date (the prediction target date on which the power demand peak value is predicted by the power demand peak prediction device 1). _DF-d indicates the predicted value of the power demand peak value on day d (d = d _{1 to} d _max ), which is the evaluation target day. D _O-d indicates an actual measurement value of the power demand peak value on d (d = d _{1 to} d _max ) day, which is the evaluation target date. Thus,% MAE is a value obtained by dividing the sum of the prediction error rates on each evaluation target day by the total number of days on the evaluation target date, and indicates the size of the prediction error per day on the evaluation target date. Therefore, it can be said that the smaller the% MAE, the smaller the prediction error per day for the evaluation date, and the higher the prediction accuracy.

In the case of predicting the power demand peak value on the prediction target day based on the probability distribution (frequency distribution) of the power demand peak value of the past data among the methods shown in the above “second example”, the following formula (2 ) CRPS (Continuous Rank Probability Score) represented by an evaluation index (reference: Hans Hersbach; “Decomposition of the Continuous Ranked Probability Score for Prediction Systems”, American Meteorological Society, 15, 5, (2000), pp. 559-570).

In the above formula _{(2), P F-d} (D i) is the power demand peak value of the past data extracted by using the feature information at 0.1kW units until 0kW~20kW (i = 0~200) Indicates the accumulated probability. P _o (D _{o- d} ) represents a step function that changes from 0 to 1 with the actual measurement value D _o-d on the prediction target day. Therefore, CRPS indicates the probability of the probability distribution PF _-d predicted for the prediction target date, and the smaller the value, the higher the accuracy of the prediction probability distribution of the evaluation target date.

  Past data of the past date (corresponding day) extracted based on the feature information is extracted from the past 90 days based on the prediction target date, and the expected value (average value) of the peak power demand value of the past data is predicted. The prediction result when it is assumed to be a value will be described. When the day of the week, weather, and maximum temperature were used alone as characteristic information, the% MAEs were 23.0%, 23.0%, and 14.3%, respectively. Here, when the maximum temperature is used as the feature information, the past data is extracted with a day within ± 5 ° C. of the maximum temperature forecast value of the prediction target day as the corresponding day. On the other hand,% MAE when the expected value (average value) of the peak power demand value of the past data for 7 to 78 days randomly extracted from the past 90 days without using the feature information is about 23% there were. Therefore, when using the maximum temperature as the feature information, it was confirmed that the% MAE was improved by about 9% and the prediction accuracy was improved as compared with the case where the past data was extracted at random.

  On the other hand, when the power demand peak value on the prediction target day is predicted based on the regression equation between the maximum temperature and the power demand peak value by the method shown in the above “third example”, the% MAE is 14. 1%. As described above, in this example, the prediction accuracy was slightly improved in the case of using the regression equation than in the case where the expected value (average value) of the power demand peak value of the past data is used as the predicted value. FIG. 15 shows a correlation diagram between the actual measurement value and the predicted value of the power demand peak value on the prediction target day when the regression equation is used. In FIG. 15, the horizontal axis (x-axis) and the vertical axis (y-axis) indicate the measured value magnitude (kW) and predicted value magnitude (kW) of the power demand peak value on the prediction target day, respectively. As shown in the figure, a regression line represented by “y = 0.95884x” was obtained by performing regression analysis based on a plurality of sets of actually measured values and predicted values.

  Further, regarding the prediction accuracy when the predicted value of the power demand peak value is calculated using the probability distribution (frequency distribution), the above formula (2) is used when the day of the week, the weather, and the maximum temperature are used as feature information for extracting past data. The CRPS calculated by) were 16.0, 15.6, and 10.2, respectively. As described above, even when the power demand peak value is predicted based on the probability distribution (frequency distribution), it is effective to increase the prediction accuracy by using the maximum temperature as the feature information as compared with the day of the week or the weather. I understood it. Similarly to the% MAE evaluation, the CRPS when the expected value (average value) of the peak power demand value of the past data for 7 to 78 days randomly extracted from the past 90 days is 18.7 is 18.7. Met. As described above, in this embodiment, the prediction accuracy when the power demand peak value is predicted based on the probability distribution (frequency distribution) of the power demand peak value of the past data is 45% by using the maximum temperature as the feature information. Even when the day of the week and the weather are used as characteristic information, it was confirmed that the level improved by about 15%.

  In order to evaluate the effect of the forecast error of the maximum temperature on the forecast target day on the forecast result of the peak power demand value, the past day when the maximum temperature within ± 5 ° C was observed from the actual measured maximum temperature on the forecast date The past data of (corresponding day) was extracted, and the power demand peak value was predicted based on the past data. As a result,% MAE when forecasting based on the expected value (average value) of the peak power demand value of past data is 14.1%, and% MAE when forecasting using the regression equation is 13.5%. Compared to the forecast error related to the maximum temperature, it improved by 0.2% and 0.6%, respectively. The absolute value of the forecast error related to the daily maximum temperature was about 1.2 ° C on average during the target period for which the peak demand for power demand was predicted. It was confirmed that there was a forecast error of 1 ℃ or more. Therefore, when past data is extracted using the maximum temperature as feature information, it can be expected that the prediction accuracy of the power demand peak value is improved by improving the prediction accuracy of the maximum temperature on the prediction target day.

  The range of the maximum temperature of the past data to be extracted by the past data extraction unit 13 is changed from ± 1 ° C to ± 10 ° C from the predicted maximum temperature on the prediction target day, and the expected value of the power demand peak value of the past data ( % MAE when the power demand peak value is predicted based on the average value) and the regression formula, and CRPS when the power demand peak value is predicted based on the probability distribution (frequency distribution) of the power demand peak value of the past data. Calculated. FIG. 16 shows the result. In FIG. 16, the horizontal axis indicates the range of the maximum temperature of the past data to be extracted (the range based on the maximum temperature forecast value on the prediction target day), and the vertical axis indicates the prediction error (average absolute error (% MAE), CRPS). Here, when the range of the maximum temperature was set to ± 1 ° C. from the predicted maximum temperature on the prediction target day, the corresponding day might not exist, so it was excluded from consideration. In addition, when the regression equation is used, if the maximum temperature range is set to ± 3 ° C. or less from the maximum temperature forecast value on the prediction target day, it may be excluded from consideration because there may be only one day of the corresponding day. As shown in the figure, CRPS for predicting based on% MAE and probability distribution (frequency distribution) when predicting based on an expected value indicates the range of maximum temperature from the maximum temperature predicted value on the prediction target day. It was confirmed that the temperature was minimized when the temperature was 3 ° C. (prediction accuracy was high). In this case, the past data to be extracted may exist only for one day, but it is considered that the prediction accuracy has been improved by using only the day with the measured value of the maximum temperature close.

  Further, sensitivity analysis was performed by changing the period (extraction period) in which past data was extracted by the past data extraction unit 13. As a result, except in March, the number of past data to be extracted is reduced, but when the extraction period is relatively short (for example, when the past 30 days from the prediction target day is the extraction period), the prediction error is greater. It was confirmed to be smaller. On the other hand, in March, it was confirmed that the prediction error becomes smaller when past data is extracted retroactively to around December in the past 90 days as the extraction period. However, in this case, the power demand peak value in March is predicted by the actual measurement value around December, so it was also confirmed that% MAE is about 6.8% larger than in other seasons. . Therefore, it is considered possible to improve the prediction accuracy by using the feature information other than the maximum temperature together and optimizing the extraction range of the past day according to the prediction target date.

(Forecast results for peak power demand hours)
The start point, end point, and length of the power demand peak time zone of the prediction target day based on the expected value of the start point, end point, and length of the power demand peak time zone of the past data by the method shown in the “first example” above. In the case of predicting the error, the prediction accuracy was evaluated by using an average error (ME: Mean Error) that is a value obtained by adding the error between the predicted value and the actual measurement value. Further, for the prediction accuracy of the entire power demand peak time zone, an index NCAE (Normalized Cumulative Absolute Error) representing the absolute error of the probability distribution between the predicted value and the actually measured value represented by the following formula (3) was defined and used. .

In the above formula (3), _{PFd, N} is the corresponding day (past date) for each time zone N (N = 1 to 48) obtained by dividing one day (24 hours) every 30 minutes. The probability (frequency) indicating whether the time zone N is included in the power demand peak time zone. P _{o-d, N} is a probability (0 or 1) indicating whether or not each time zone _N of the prediction target date d (d = 1 to d _max ) is included in the power demand peak time zone, When the time zone N is included in the power demand peak time zone, “1” is taken, and when it is not included, “0” is taken. The smaller the prediction error in the entire evaluation target date, the smaller the NCAE. In addition, as described above, since the time period in which the amount of electricity used includes the time for taking the power demand peak value and is continuous at the threshold value Dh or more is defined as the power demand peak time period, the power demand peak is increased by increasing the threshold value Dh. The time zone becomes shorter. Therefore, in order to compare the prediction accuracy regardless of the difference in the threshold value Dh, the NCAE represented by the above formula (3) is standardized by the length of each daily power demand peak time zone. That is, since NCAE does not depend on the threshold value Dh (the length of the power demand peak time zone), it can also be used as an evaluation index when comparing prediction accuracy when prediction is performed using different threshold values Dh.

  FIG. 17 shows the frequency of occurrence of prediction errors in the power demand peak time zone when the power demand peak time zone is defined with the threshold Dh as “Dh = Da”. 17A, 17B, and 17C are graphs in which the horizontal axis represents the prediction error (h: time) and the vertical axis represents the occurrence frequency (%). FIGS. 17A, 17 </ b> B, and 17 </ b> C show graphs related to the expected value of the start time, end time, and length of the power demand peak time period, respectively. As shown in FIG. 17 (a), the prediction error of the start time is ± 64% of the entire prediction target day as the evaluation target, regardless of which day of the week, weather, or maximum temperature is used as the feature information. Within 30 minutes. As described above, since the end point time is more different depending on the day than the start point time, the prediction error of the end point time is larger than the prediction error of the start point time as shown in FIG. The number of days with many prediction errors within ± 30 minutes was 50.1% of the entire prediction target day. As shown in FIG. 17C, the prediction error related to the length of the power demand peak time zone is the largest among the start point time, end point time, and length, and the day when the prediction error is within ± 30 minutes is the prediction target day. It was 47.5% of the whole. This is considered because the error between the start point time and the end point time may be in the opposite direction (one is a positive value and the other is a negative value). However, in the present embodiment, as described above, it is not possible to predict the morning power demand peak time zone that occurs from autumn to winter, and therefore, the start point time and the end point time have a very large error of 10 to 12 hours. There is a case. However, in such a case, both the start point time and the end point time occur in the morning. Therefore, with regard to the length of the peak time zone, a large prediction error does not occur by not considering the morning power demand peak time zone.

  FIG. 18 shows the expected value in the power demand peak time zone when the threshold value Dh is “Dh = Da”, “Dh = 1.1 Da”, “Dh = 1.2 Da”, and “Dh = 1.3 Da”. The prediction error (average error) is shown. 18A, 18B, and 18C are graphs in which the horizontal axis indicates the threshold value Dh and the vertical axis indicates the average error (h: time). FIGS. 18A, 18B, and 18C show graphs relating to the expected start point time, end point time, and expected length of the power demand peak time period, respectively. Here, the average error is calculated excluding a total of 33 days when the peak power demand period occurs in the morning. As shown in FIG. 14, since the probability that the starting point time of the power demand peak time zone is around 17:00 is very high, the prediction error is about 40 minutes regardless of the threshold value Dh. On the other hand, in the case of the end point time where the difference by day is larger than the start point time, the prediction error is about 40 minutes when the threshold value Dh is “Dh = Da”, and the threshold value Dh is “Dh = 1.3 Da”. In this case, it takes about 80 minutes. The length also has the same tendency as the end point time, but the length prediction error is generally about 10 minutes larger than the end point time prediction error. However, it was confirmed that the prediction errors of the start point time, the end point time, and the length are not much different from the case where the past data of the past 7 to 78 days are randomly extracted without using the feature information. That is, in order to improve the prediction accuracy regarding the start point time, end point time, and length of the power demand peak time zone in which the difference by day is about one hour throughout the year, similar days with these differences of several tens of minutes are extracted. It is considered necessary.

  Supplementary information on the peak power demand time zone (morning peak time zone) that occurs in the morning. Predicting the morning peak hours is considered effective for efficient power supply. As a first-stage study on forecasting morning peak hours, the ratio of morning peak demand (number of days when power demand peak hours occur in the morning) to afternoon peak demand (number of days when power demand peak hours occur in the afternoon) ( The correlation between morning and afternoon peak demand ratio) and various feature information was evaluated. FIG. 19 shows the relationship between the minimum temperature and the morning / afternoon peak demand ratio as an example. FIG. 19 is a graph in which data of a plurality of past days are plotted with the horizontal axis being the daily minimum temperature (° C.) and the vertical axis being the morning / afternoon peak demand ratio (morning peak demand / afternoon peak demand). According to the figure, the ratio of the morning / afternoon peak demand ratio to 1 or more, that is, the morning peak time zone is high on the day when the minimum temperature is 5 ° C. or lower or 25 ° C. or higher. Therefore, if the minimum temperature on the prediction target day such as the next day is 5 ° C. or lower or 25 ° C. or higher, the possibility of the morning peak time zone occurring may be considered.

[Second Embodiment]
The power demand peak prediction apparatus 100 according to the second embodiment will be described with reference to FIG. The power demand peak prediction apparatus 100 according to the present embodiment has a function of determining a period (extraction period for prediction) in which past data is extracted according to the prediction target date, and thus the power demand peak prediction according to the first embodiment. Mainly different from the apparatus 1. As illustrated in FIG. 20, the power demand peak prediction apparatus 100 includes a feature information acquisition unit 101, a past data storage unit 102, a past data extraction unit 103, a peak identification unit 104, a prediction unit 105, an output unit 106, a predicted value / actual measurement. A value input unit 107, an error calculation unit 108, an extraction period determination unit 109, and an extraction period storage unit 110 are provided.

  The power demand peak prediction apparatus 100 according to the present embodiment has a year (predetermined) from a plurality of different extraction periods (for example, “the past 30 days from the prediction target day”, “the past 90 days from the prediction target day”, etc.). A first phase for determining a prediction extraction period for each period (for example, a specific “month” such as “April”), and a prediction target day using the prediction extraction period determined in the first phase A second phase for predicting a power demand peak (power demand peak value or power demand peak time zone) is executed.

  For example, the power demand peak prediction apparatus 100 determines a prediction extraction period to be used for prediction in the next fiscal year and after by executing the first phase process for a predetermined period (for example, one year). In this case, the power demand peak prediction device 100 executes a prediction process (second phase) using the prediction extraction period determined in the first phase when predicting a new prediction target day in the next fiscal year or later. can do. Here, for the next year and thereafter, the processing of the first phase may be executed together with the second phase in order to determine the extraction period for prediction to be used for the prediction after the following year.

  In addition, the power demand peak prediction device 100 performs the first phase process based on the past data stored in the past data storage unit 102 and uses the past date as the prediction target date, thereby predicting the future prediction. You may determine the extraction period for prediction used for prediction of an object day (for example, the next day etc.). In this case, the power demand peak prediction apparatus 100 can perform prediction (second phase) of the prediction target date (for example, the next day) using the prediction extraction period determined in the first phase.

  Hereinafter, processing executed by each functional element in each phase will be described. Of the functions of the feature information acquisition unit 101, the past data storage unit 102, the past data extraction unit 103, the peak identification unit 104, the prediction unit 105, and the output unit 106, the feature information acquisition unit 11 according to the first embodiment. Detailed description of the same functions as those of the past data storage unit 12, the past data extraction unit 13, the peak identification unit 14, the prediction unit 15, and the output unit 16 will be omitted.

  First, processing performed by each functional element in the first phase will be described. As described above, the first phase calculates and evaluates the prediction error for each of a plurality of different extraction periods, so that the prediction extraction period corresponding to the prediction target day (or the time when the prediction target day is included) is calculated. It is a phase to decide.

  For each of a plurality of different extraction periods, the past data extraction unit 103 is the same as the feature information of the prediction target date acquired by the feature information acquisition unit 101 from the past data including the past date in the extraction period, or One or more past data including feature information similar to each other according to a predetermined standard is extracted from the past data storage unit 102. Here, the plurality of different extraction periods are different periods such as “the past 30 days from the prediction target date” and “the past 90 days from the prediction target date”, for example. Further, the extraction period may be a period retroactive to the past based on the prediction target date as described above, or may be a period retroactive to the past based on a date other than the prediction target date. The extraction period may be a period that is continuous in time, or may be a period in which a plurality of periods that are discontinuous in time are bundled. A plurality of different extraction periods are set in advance by an operator or the like, for example.

  The peak specifying unit 104, the prediction unit 105, and the output unit 106 execute each process described above based on past data extracted by the past data extraction unit 103 for each of a plurality of extraction periods. Thereby, the predicted value of the power demand peak of the prediction target day is calculated and output for each of the plurality of extraction periods.

  The predicted value / actual value input unit 107 inputs the predicted value of the power demand peak output for each of the plurality of extraction periods as described above and the actual measured value of the power demand peak of the predicted target date. It is a predicted value / actual value input means for inputting. In the example of FIG. 20, the predicted value / actual value input unit 107 refers to the actual measurement value of the power demand peak on the prediction target date stored in the past data storage unit 12 after the prediction target date has elapsed. The actual measurement value of the power demand peak on the prediction target day is input.

  The error calculation unit 108 is an error calculation unit that calculates an error between each predicted value for each extraction period input by the predicted value / measured value input unit 107 and the measured value. The error calculation method is not particularly limited, but the error calculation unit 108 may calculate, for example, the absolute value (absolute error) of the difference between the predicted value and the actual measurement value as the error.

  The extraction period determining unit 109 is an extraction that determines a prediction extraction period used to calculate a predicted value of the power demand peak for a new prediction target day based on the error for each extraction period calculated by the error calculation unit 108. It is a period determining means. As described above, the extraction period determination unit 109 may determine a prediction extraction period for each time period in which a prediction target day is included in one year (predetermined cycle). The “time when the prediction target day is included in one year” indicates a specific period that periodically visits, such as “month” or “season” including the prediction target day.

  The extraction period determination unit 109, for example, for each extraction period, each of the prediction target days (for example, “April 1” to “April 30”) included in a predetermined evaluation target period (for example, “April”). Based on the prediction error for, the prediction error (e.g., the above-described indices such as% MAE, CRPS, and NCAE) in the entire evaluation target period is calculated. Subsequently, the extraction period determining unit 109 compares the prediction errors for each extraction period to determine the extraction period that minimizes the prediction error as the prediction extraction period corresponding to the evaluation target period (time). Can do. Further, the extraction period determination unit 109 may determine a prediction extraction period corresponding to the evaluation target period, for example, as follows. That is, the extraction period determination unit 109 specifies an extraction period in which a prediction error is the smallest among a plurality of extraction periods for each prediction target day included in the evaluation target period. Subsequently, the extraction period determination unit 109 determines the extraction period that is identified most frequently, that is, the extraction period that has the largest number of prediction target days with the smallest prediction error as the prediction extraction period corresponding to the evaluation target period. can do.

  The extraction period for prediction determined by the extraction period determination unit 109 is stored in the extraction period storage unit 110. As described above, when the prediction extraction period is determined for each time period including the prediction target day, the extraction period storage unit 110 stores the time period (for example, “April”) and the prediction extraction period (for example, “past 90” Day)) in association with each other.

  Next, processing executed by each functional element in the second phase will be described. As described above, the second phase is a phase for predicting a power demand peak on a new prediction target day by actually using the prediction extraction period determined in the first phase.

  The past data extraction unit 103 calculates the prediction value of the power demand peak for the new prediction target day, and the prediction extraction period determined by the extraction period determination unit 109 (the prediction stored in the extraction period storage unit 110). The past data is extracted using the extraction period. In other words, the past data extraction unit 103 is the same as or similar to the feature information (for example, the maximum temperature) of the new prediction target day from the past date past data included in the prediction extraction period. Extract one or more past data including. Thereby, based on the past data in the extraction period for prediction, it becomes possible to predict the power demand peak on the prediction target day more accurately.

  When the extraction period storage unit 110 stores the time and the extraction period for prediction in association with each other, the past data extraction unit 103 calculates the predicted value of the power demand peak for the new prediction target day. Even if past data is extracted using a prediction extraction period (for example, “past 90 days”) corresponding to a period (for example, “April”) in which the new prediction target day is included in one year (predetermined period) Good. Thereby, it becomes possible to predict the power demand peak of the prediction target day more accurately based on the appropriate extraction period for prediction according to the time when the prediction target day is included in the predetermined cycle.

  In the second embodiment described above, in the first phase, a result (predicted extraction period obtained) learned from the prediction result of the same period (same “month”, etc.) of last year or last year in a cycle of one year. It has been described that the prediction in the second phase is executed using. However, the processes of the first phase and the second phase are not limited to this case. For example, in the first phase, learning may be performed from the prediction result of the most recent period of the same year (the past week, the past month, etc.). Further, the power demand peak prediction apparatus determines feature information that can be expected to improve the prediction accuracy of the power demand peak on the prediction target day based on the same idea as the method for determining the prediction extraction period described above. May be.

  Further, in the above-described embodiment, the case where the prediction is performed using the maximum temperature, the weather, and the day of the week as the characteristic information has been mainly described. The prediction may be performed using the amount, the cloud amount, the wind condition, and the like as feature information.

  Moreover, in the above-mentioned embodiment, as shown in FIG.1 and FIG.20, the structure where a power demand peak prediction apparatus is provided with a past data storage part (For example, the past data storage part is built in the apparatus which comprises a power demand peak prediction apparatus. Explained). However, the past data storage unit may be provided in another device that can be externally connected to the power demand peak prediction device via USB or the like. Alternatively, the past data storage unit may be provided in another device that can be accessed from the power demand peak prediction device via the network. In addition, when the power demand peak prediction device is configured not to include the past data storage unit as described above, the past data extraction unit is configured to store the past data stored in a device accessible via an external connection or a network. What is necessary is just to extract past data from a part. Further, when the computer is caused to function as a power demand peak prediction apparatus that does not include a past data storage unit, the above-described past data storage module is not required in the power demand peak prediction program.

  DESCRIPTION OF SYMBOLS 1,100 ... Electric power demand peak prediction apparatus, 11, 101 ... Feature information acquisition part, 12, 102 ... Past data storage part, 13, 103 ... Past data extraction part, 14, 104 ... Peak identification part, 15, 105 ... Prediction , 16, 106 ... output unit, 107 ... predicted value / actual value input unit, 108 ... error calculation unit, 109 ... extraction period determination unit, 110 ... extraction period storage unit, P11 ... feature information acquisition module, P12 ... past data Storage module, P13 ... past data extraction module, P14 ... peak identification module, P15 ... prediction module, P16 ... output module.

Claims (10)

Past data storage means for storing past data in which the amount of electricity used at each time in the past period and the feature information indicating the characteristics of the past period are associated with each other;
Feature information acquisition means for acquiring feature information indicating the characteristics of the prediction target period;
Past data extraction means for extracting one or more past data including feature information that is the same as or similar to the feature information of the prediction target period acquired by the feature information acquisition means from the past data storage means;
For each past data extracted by the past data extracting means, a power demand peak value indicating a peak of the electricity usage amount in the past period or a power demand peak time zone in which the electricity usage amount exceeds a predetermined threshold in the past period. Peak identification means to identify;
Prediction for calculating a power demand peak value or a power demand peak time zone predicted value for the forecast period based on the power demand peak value or the power demand peak time zone of each past data specified by the peak specifying means Means,
Output means for outputting a predicted value calculated by the prediction means;
A power demand peak prediction apparatus comprising:   The power demand peak prediction apparatus according to claim 1, wherein the characteristic information is at least one of maximum temperature, day of the week, weather, humidity, sunshine duration, precipitation, cloud cover, and wind conditions. The peak specifying means specifies at least two of a start point time, an end point time, and a length of the power demand peak time zone of each past data,
The prediction unit performs statistical processing based on at least two of a start point time, an end point time, and a length of the power demand peak time period of each past data specified by the peak specifying unit, and thereby the prediction target Calculate the predicted value of the peak power demand period for the period,
The power demand peak prediction apparatus according to claim 1 or 2. The peak specifying means specifies the power demand peak value of each past data,
The predicting unit calculates a predicted value of the power demand peak value in the prediction target period by executing a statistical process based on the power demand peak value of each past data specified by the peak specifying unit.
The electric power demand peak prediction apparatus as described in any one of Claims 1-3. The feature information includes numerical information indicated by numerical values;
The peak specifying means specifies the power demand peak value of each past data,
The prediction means calculates a regression equation based on a set of the power demand peak value of each past data specified by the peak specification means and the numerical information included in the past data, and the prediction equation is calculated as the prediction formula By substituting the numerical information of the target period, the predicted value of the power demand peak value of the prediction target period is calculated.
The electric power demand peak prediction apparatus as described in any one of Claims 1-4. For each of a plurality of different extraction periods, the peak specifying means, the prediction means, and the based on past data extracted by the past data extraction means from past data that includes the past period in the extraction period The output means inputs the power demand peak value or the power demand peak time period predicted value during the prediction target period output by executing each process, and the power demand peak value or power demand peak time during the prediction target period. A predicted value / measured value input means for inputting the measured value of the belt,
Error calculation means for calculating an error between each of the predicted values for each extraction period input by the predicted value / actual value input means and the actual value;
Based on the error for each extraction period calculated by the error calculation means, a prediction extraction period used to calculate a power demand peak value for a new prediction target period or a power demand peak time period prediction value is determined. An extraction period determining means,
The past data extraction unit is configured to calculate the power demand peak value of the new prediction target period or the prediction value of the power demand peak time zone in the prediction extraction period determined by the extraction period determination unit. One or more past data including feature information similar to the feature information acquired by the feature information acquisition unit or similar to a predetermined standard is extracted from the past data including
The electric power demand peak prediction apparatus as described in any one of Claims 1-5. The extraction period determining means determines the prediction extraction period for each time period in which the prediction target period is included in a predetermined cycle based on the error for each extraction period calculated by the error calculation means,
The past data extraction means corresponds to a time when the new prediction target period is included in the predetermined cycle when calculating a power demand peak value or a power demand peak time period prediction value of the new prediction target period. Extracting one or more past data including feature information that is the same as or similar to the feature information acquired by the feature information acquisition unit from past data in which the past period is included in the prediction extraction period ,
The power demand peak prediction apparatus according to claim 6. Feature information acquisition means for acquiring feature information indicating the characteristics of the prediction target period;
From the past data storage means for storing the past data in which the amount of electricity used at each time in the past period and the feature information indicating the feature of the past period are associated, the prediction target period acquired by the feature information acquisition means Past data extraction means for extracting one or more past data including feature information that is the same as or similar to the feature information in a predetermined criterion;
For each past data extracted by the past data extracting means, a power demand peak value indicating a peak of the electricity usage amount in the past period or a power demand peak time zone in which the electricity usage amount exceeds a predetermined threshold in the past period. Peak identification means to identify;
Prediction for calculating a power demand peak value or a power demand peak time zone predicted value for the forecast period based on the power demand peak value or the power demand peak time zone of each past data specified by the peak specifying means Means,
Output means for outputting a predicted value calculated by the prediction means;
A power demand peak prediction apparatus comprising: A power demand peak prediction method executed by a power demand peak prediction device comprising past data storage means for storing past data in which the amount of electricity used at each time in the past period is associated with feature information indicating the characteristics of the past period Because
A feature information acquisition step of acquiring feature information indicating characteristics of the prediction target period;
Past data extraction step of extracting one or more past data including feature information similar to or the same as the feature information of the prediction target period acquired in the feature information acquisition step from the past data storage unit;
For each past data extracted in the past data extraction step, a power demand peak value indicating a peak of the electricity usage amount in the past period or a power demand peak time zone in which the electricity usage amount becomes a predetermined threshold or more in the past period. A peak identification step to identify;
Prediction for calculating a power demand peak value or a power demand peak time zone prediction value for the forecast period based on the power demand peak value or the power demand peak time zone of each past data identified in the peak identifying step Steps,
An output step of outputting the predicted value calculated in the prediction step;
Power demand peak prediction method including. Computer
Past data storage means for storing past data in which the amount of electricity used at each time in the past period and the feature information indicating the characteristics of the past period are associated with each other;
Feature information acquisition means for acquiring feature information indicating the characteristics of the prediction target period;
Past data extraction means for extracting one or more past data including feature information that is the same as or similar to the feature information of the prediction target period acquired by the feature information acquisition means from the past data storage means;
For each past data extracted by the past data extracting means, a power demand peak value indicating a peak of the electricity usage amount in the past period or a power demand peak time zone in which the electricity usage amount exceeds a predetermined threshold in the past period. Peak identification means to identify;
Prediction for calculating a power demand peak value or a power demand peak time zone predicted value for the forecast period based on the power demand peak value or the power demand peak time zone of each past data specified by the peak specifying means Means,
A power demand peak prediction program for functioning as output means for outputting a predicted value calculated by the prediction means.