KR20210103197A - Method and system for daily power peak load forecasting using regression models involving interaction effects - Google Patents

Abstract

The present invention relates to a method and a system for predicting a daily peak power demand using a regression model including an interaction. The method includes: a step of receiving a prediction value request message for requesting a prediction value with respect to the peak power demand of a prediction day; a step of calculating a regression coefficient with respect to first independent variable values and first interaction variable values in response to the prediction value request message and based on a first peak power consumption regarding a previous day at least one day before the prediction day, the first independent variable values as the values of first independent variables affecting the first peak power consumption, and the first interaction variable values reflecting the interaction between the first independent variables; and a step of calculating the prediction value based on the regression coefficient, second independent variable values as the values of second independent variables affecting the prediction value, and second interaction variable values reflecting the interaction between the second independent variables.

Description

METHOD AND SYSTEM FOR DAILY POWER PEAK LOAD FORECASTING USING REGRESSION MODELS INVOLVING INTERACTION EFFECTS

The following embodiments relate to a method and system for predicting the maximum daily power demand using a regression model including an interaction.

Electricity is a very important energy resource for the Korean economy, and it must be supplied exactly as needed. In addition, since electricity has a characteristic that production and consumption occur simultaneously, accurately predicting electricity demand is essential for power system operation. The reason that accurate peak power demand forecasting is important is that, in case of underestimation, a blackout may occur, which may paralyze part or the entire power system, and in case of overestimation, economic loss due to excessive power generation than necessary occurs. Because.

Due to the importance of peak power demand prediction, many studies have been conducted for peak power demand prediction. According to the prediction method, it can be largely classified into a statistical method and a study using a machine learning method. First, as AI (Artificial Intelligence) is being applied to many fields in recent years, many researchers are also applying prediction methods through machine learning in the field of power demand prediction. For example, researchers conduct research using deep learning techniques such as deep neural networks to predict power demand. Researchers also use machine learning methods such as radiative basis function neural networks, extreme learning machines (ELM), and support vector regression (SVR) to predict peak power demand.

Recently, research using machine learning techniques has been in the spotlight, but studies on power demand prediction using statistical techniques have been steadily published until recently. Representative prediction methods of statistical techniques include a prediction method using a time series model and a regression analysis model. As for the time series model, the Reg-ARIMA and Reg-AR-SGARCH models, which modified/improved the traditional ARIMA (Auto-Regressive Integrated Moving Average) model, have been proposed until recently and used to predict the maximum power demand. Recent studies based on the regression analysis model include a power demand forecasting method that reflects factors for the day of the week and special days in the model, and a power demand forecasting method that reflects seasonal factors in the model. However, since the interaction effect between each variable is not reflected at all, the accuracy of the electricity demand prediction is low.

The embodiments provide an accurate prediction of the maximum power demand using a plurality of interaction variable values in which the interaction between a plurality of independent variables, which are variables affecting the maximum power consumption per day, is considered when predicting the maximum daily power demand. It can provide the ability to calculate.

The method of predicting maximum power demand according to an embodiment comprises the steps of: receiving a forecast request message for requesting a forecast for the maximum power demand on a forecast day; A first maximum power consumption for a , a first plurality of independent variable values that are values of a first plurality of independent variables affecting the first maximum power consumption, and an interaction between the first plurality of independent variables are considered. calculating a regression coefficient for the first plurality of independent variable values and the first plurality of interaction variable values based on the first plurality of interaction variable values; Calculate the predicted value based on values of a second plurality of independent variables that are values of a second plurality of independent variables influencing and a second plurality of values of interaction variables in which an interaction between the second plurality of independent variables is considered. including the steps of

The calculating of the regression coefficient may include generating a first plurality of interaction variable values in which an interaction between the first plurality of independent variables is considered, based on the values of the first plurality of independent variables. can

The values of the first plurality of independent variables include a second maximum power consumption for a day immediately preceding the previous day of at least one day, a third maximum power consumption for a week before the previous day of at least one day or more, and a previous day of the at least one day or more. the day of the week, the temperature of the previous day of at least one day, the month to which the at least one previous day belongs, whether the at least one day or more of the previous day corresponds to a special day, and the at least one day or more It may include a first day index, which is a power consumption trend for the previous day.

The generating of the first plurality of interaction variable values may include: generating a first interaction variable value using the second maximum power consumption and a day to which the previous day of at least one day belongs; generating a second interaction variable value using the temperature of the previous day and the month to which the at least one previous day belongs; generating a third interaction variable value using the day of the week, generating a fourth interaction variable value by using the day of the week to which the at least one previous day belongs and the temperature of the previous day of at least one day may include

The calculating of the predictive value may include generating a second plurality of interaction variable values in which an interaction between the second plurality of independent variables is considered based on the values of the second plurality of independent variables. have.

The values of the second plurality of independent variables include a fourth maximum power consumption for the day immediately preceding the prediction date, a fifth maximum power consumption for one week before the prediction day, a day of the week to which the prediction date belongs, and a prediction of the prediction day. It may include a temperature, a month to which the forecast date belongs, whether the forecast date corresponds to a special day, and a second day index that is a power consumption trend for the forecast date.

The generating of the plurality of second interaction variable values may include generating a fifth interaction variable value using the fourth maximum power consumption and the day of the week to which the prediction date belongs, the temperature of the prediction day and the Generating a sixth interaction variable value using the month to which the prediction date belongs, and generating a seventh interaction variable value using whether the prediction date corresponds to the special day and the day of the week to which the prediction date belongs and generating an eighth interaction variable value using the day of the week to which the prediction date belongs and the expected temperature of the prediction day.

The special day may include a national holiday, a substitute holiday, an election day, or a temporary public holiday.

The method may further include receiving the first maximum power consumption, the first plurality of independent variable values, and the second plurality of independent variable values from a server.

The apparatus for predicting maximum power demand according to an embodiment includes a memory storing instructions for predicting maximum power demand, and a processor for executing the instructions, wherein when the instructions are executed by the processor, the processor comprises: Receiving a forecast request message for requesting a forecast for the maximum power demand of the forecast day, and in response to the forecast request message, the first maximum power consumption for at least one day or more prior to the forecast day, the first maximum power consumption Based on the values of the first plurality of independent variables that are values of the first plurality of independent variables that influence, and the first plurality of interaction variable values in which an interaction between the first plurality of independent variables is considered. compute a first plurality of independent variable values and a regression coefficient for the first plurality of interaction-action variable values; The predicted value is calculated based on a second plurality of interaction variable values in which an interaction between the variable values and the second plurality of independent variables is taken into account.

The processor may generate a first plurality of interaction variable values in which an interaction between the first plurality of independent variables is considered based on the values of the first plurality of independent variables.

The values of the first plurality of independent variables include a second maximum power consumption for a day immediately preceding the previous day of at least one day, a third maximum power consumption for a week before the previous day of at least one day or more, and a previous day of the at least one day or more. the day of the week, the temperature of the previous day of at least one day, the month to which the at least one previous day belongs, whether the at least one day or more of the previous day corresponds to a special day, and the at least one day or more It may include a first day index, which is a power consumption trend for the previous day.

The processor generates a first interaction variable value using the second maximum power consumption and a day to which the previous day of at least one day belongs, and the temperature of the previous day of at least one day and the previous day of at least one day or more. generating a second interaction variable value using the month, and generating a third interaction variable value using whether the at least one previous day corresponds to the special day and the day of the week to which the at least one or more previous day belongs, , a fourth interaction variable value may be generated using the day of the week to which the at least one previous day belongs and the temperature of the at least one or more previous days.

The processor may generate a second plurality of interaction variable values in which an interaction between the second plurality of independent variables is considered based on the values of the second plurality of independent variables.

The values of the second plurality of independent variables include a fourth maximum power consumption for the day immediately preceding the prediction date, a fifth maximum power consumption for one week before the prediction day, a day of the week to which the prediction date belongs, and a prediction of the prediction day. It may include a temperature, a month to which the forecast date belongs, whether the forecast date corresponds to a special day, and a second day index that is a power consumption trend for the forecast date.

The processor generates a fifth interaction variable value using the fourth maximum power consumption and the day of the week to which the prediction date belongs, and a sixth interaction variable using the temperature of the prediction day and the month to which the prediction day belongs. a value is generated, a seventh interaction variable value is generated using whether the prediction date corresponds to the special day and the day of the week to which the prediction date belongs, can be used to generate the eighth interaction variable value.

The special day may include a national holiday, a substitute holiday, an election day, or a temporary public holiday.

The processor may receive the first maximum power consumption, the first plurality of independent variable values, and the second plurality of independent variable values from the server.

1 is a diagram illustrating a system for predicting a maximum power demand according to an embodiment.
FIG. 2 is a diagram schematically illustrating the prediction apparatus shown in FIG. 1 .
FIG. 3 is a diagram for explaining an operation in which the prediction apparatus calculates a regression coefficient and a predicted value.
4 to 9 are diagrams for explaining a plurality of independent variables and interaction variables used by the prediction apparatus.
10 is a diagram illustrating an example of a result of analysis of variance of a prediction device.
11 is a diagram illustrating an example of calculating an adjustment determination coefficient of a prediction apparatus.
12 is a diagram for explaining a performance verification result of a prediction apparatus.
13 is a diagram illustrating a result of calculating a predicted value for the same period by a prediction apparatus and a conventional prediction method.
14 is a diagram illustrating the effect of interaction variables used by a prediction device on prediction performance.

Hereinafter, embodiments will be described in detail with reference to the accompanying drawings. However, since various changes may be made to the embodiments, the scope of the patent application is not limited or limited by these embodiments. It should be understood that all modifications, equivalents and substitutes for the embodiments are included in the scope of the rights.

The terms used in the examples are used for description purposes only, and should not be construed as limiting. The singular expression includes the plural expression unless the context clearly dictates otherwise. In the present specification, terms such as “comprise” or “have” are intended to designate that a feature, number, step, operation, component, part, or combination thereof described in the specification exists, but one or more other features It should be understood that this does not preclude the existence or addition of numbers, steps, operations, components, parts, or combinations thereof.

Although terms such as first or second may be used to describe various elements, the elements should not be limited by the terms. The terms are only for the purpose of distinguishing one element from another element, for example, without departing from the scope of rights according to the concept of the embodiment, a first element may be named as a second element, and similarly The second component may also be referred to as the first component.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiment belongs. Terms such as those defined in commonly used dictionaries should be interpreted as having a meaning consistent with the meaning in the context of the related art, and should not be interpreted in an ideal or excessively formal meaning unless explicitly defined in the present application. does not

In addition, in the description with reference to the accompanying drawings, the same components are given the same reference numerals regardless of the reference numerals, and the overlapping description thereof will be omitted. In the description of the embodiment, if it is determined that a detailed description of a related known technology may unnecessarily obscure the gist of the embodiment, the detailed description thereof will be omitted.

1 is a diagram illustrating a system for predicting a maximum power demand according to an embodiment.

The system 10 includes a user device 100 and a prediction device 200 . The system 10 may further include a server 300 .

The system 10 may accurately predict the power demand during a time period in which the power demand reaches a maximum during the prediction day, that is, on a day on which the user wants to predict the maximum power demand, that is, the peak power demand (power peak load).

The system 10 is not only an independent variable that affects the daily maximum power consumption, such as autocorrelation, day of the week, temperature, season, power consumption trend, and special, but also a seasonal temperature factor that is an interaction variable between independent variables. The maximum power demand can be predicted through a multiple regression model consisting of terms that can reflect the interaction effects such as , autocorrelation factors according to the day of the week, and special day factors according to the day of the week.

The user device 100 may generate a prediction request message in response to a user's input requesting a prediction for the maximum power demand on the prediction day. The user device 100 may transmit a first prediction value request message to the prediction device 200 . The user device 100 may be implemented integrally with the prediction device 200 or may be implemented independently.

The user device 100 may output the predicted value for the maximum power demand on the predicted day through a display or the like.

The prediction device 200 considers a plurality of independent variable values that are values of a plurality of independent variables affecting the maximum daily power consumption when predicting the daily maximum power demand in response to the prediction value request message, and the interaction between the plurality of independent variables A forecast for the maximum power demand on an accurate forecast day may be calculated using the plurality of interaction variable values.

The prediction apparatus 200 may output a prediction value for the maximum power demand on the prediction day through a display or the like. The prediction device 200 may transmit the predicted value for the maximum power demand on the prediction day to the user device 100 and/or the server 300 .

The prediction apparatus 200 may generate a data request message for requesting maximum power consumption and a plurality of independent variable values for at least one day prior to the prediction date. The prediction apparatus 200 may transmit a data request message to the server 300 .

The prediction apparatus 200 may store the maximum power consumption for at least one day prior to the prediction date, and a plurality of independent variable values.

The server 300 may transmit the maximum power consumption and a plurality of independent variable values for at least one day before the prediction date to the prediction device 200 in response to the data request message. For example, the server 300 may be a power data open portal system (bigdata.kepco.co.kr) and/or a meteorological data open portal system (data.kma.go.kr).

FIG. 2 is a diagram schematically illustrating the prediction apparatus shown in FIG. 1 .

The prediction device 200 includes a transceiver 210 , a processor 230 , and a memory 250 . The prediction device 200 may be connected to a network through the transceiver 210 and communicate with the user device 100 and the server 300 .

The transceiver 210 may be connected to a network through wireless communication or wired communication to communicate with the user device 100 and the server 300 .

The transceiver 210 may receive the prediction request message. The transceiver 210 may transmit a prediction request message to the processor 230 .

The transceiver 210 may transmit a data request message to the server 300 . The transceiver 210 may receive the maximum power consumption and a plurality of independent variable values for at least one day prior to the prediction date. The transceiver 210 may transmit the maximum power consumption and a plurality of independent variable values for at least one day before the prediction date to the processor 230 .

The transceiver 210 may transmit the predicted value for the maximum power demand on the forecast day to the user device 100 and/or the server 300 .

The processor 230 may include one or more of a central processing unit, an application processor, and a communication processor.

The processor 230 may execute an operation or data processing related to control of at least one other component of the prediction apparatus 200 . For example, the processor 230 may execute an application and/or software stored in the memory 250 .

The processor 230 may process data received by the transceiver 210 and data stored in the memory 250 . The processor 230 may process data stored in the memory 250 . The processor 230 may execute computer readable code (eg, software) stored in the memory 250 and instructions induced by the processor 230 .

The processor 230 may be a hardware-implemented data processing device having a circuit having a physical structure for executing desired operations. For example, desired operations may include code or instructions included in a program.

For example, a data processing device implemented as hardware includes a microprocessor, a central processing unit, a processor core, a multi-core processor, and a multiprocessor. , an Application-Specific Integrated Circuit (ASIC), and a Field Programmable Gate Array (FPGA).

The processor 230 is configured to respond to the prediction request message, a first plurality of independent variables that are values of a first plurality of independent variables influencing the first maximum power consumption and the first maximum power consumption for at least one day prior to the prediction date. an interaction between the values and the first plurality of independent variables is applied to the first plurality of independent variable values and the first plurality of interaction variable values based on the considered first plurality of interaction variable values. We can calculate the regression coefficient for

For example, the values of the first plurality of independent variables include a second maximum power consumption for a day immediately preceding the previous day of at least one day or more, a third maximum power consumption for a week before the previous day of at least one day or more, and a previous day of at least one day or more. The day of the week, the temperature of at least one previous day, the month in which at least one previous day belongs, whether the at least one previous day is a special day, and the history of at least one or more previous days It may include a first-day index that is a consumption trend.

The processor 230 may generate a first plurality of interaction variable values in which an interaction between the first plurality of independent variables is considered based on the values of the first plurality of independent variables. For example, the processor 230 may generate the first interaction variable value using the second maximum power consumption and the day of the week to which the previous day of at least one day or more belongs. The processor 230 may generate the second interaction variable value using the temperature of at least one day or more of the previous day and the month to which the previous day of at least one day or more belongs. The processor 230 may generate the third interaction variable value using whether the previous day of at least one day or more corresponds to a special day and the day of the week to which the previous day of at least one day or more belongs. For example, special days may include national holidays, alternative holidays, election days, temporary holidays, and the like. The processor 230 may generate the fourth interaction variable value using the day of the week to which the previous day of at least one day belongs and the temperature of the previous day of at least one day or more.

The processor 230 generates a regression coefficient, a second plurality of independent variable values that are values of a second plurality of independent variables affecting the prediction value, and a second plurality of interaction variables in which an interaction between the second plurality of independent variables is considered. Based on the values, it is possible to calculate a forecast for the maximum power demand for the forecast day.

The processor 230 may generate a plurality of second interaction variable values in which an interaction between the second plurality of independent variables is considered based on the values of the second plurality of independent variables. For example, the processor 230 may generate a fifth interaction variable value using the fourth maximum power consumption and the day of the week to which the prediction date belongs. The processor 230 may generate a sixth interaction variable value using the temperature of the prediction day and the month to which the prediction day belongs. The processor 230 may generate a seventh interaction variable value using whether the prediction date corresponds to a special day and the day of the week to which the prediction date belongs. For example, special days may include national holidays, alternative holidays, election days, temporary holidays, and the like. The processor 230 may generate an eighth interaction variable value using the day of the week to which the prediction date belongs and the expected temperature of the prediction day.

Memory 250 may include volatile and/or non-volatile memory. The memory 250 may store instructions and/or data related to at least one other component of the prediction device 200 .

The memory 250 may store software and/or a program. For example, the memory 250 may store an application and software for predicting the maximum power demand on the forecast day.

FIG. 3 is a diagram for explaining an operation in which the prediction apparatus calculates a regression coefficient and a predicted value.

The prediction device 200 may receive a prediction value request message requesting a prediction value for the maximum power demand on the prediction day ( 310 ).

The prediction device 200 responds to the prediction value request message, a first plurality of independent values that are values of a first plurality of independent variables affecting the first maximum power consumption and the first maximum power consumption for at least one day or more prior to the prediction date. the first plurality of independent variable values and the first plurality of interaction variable values based on the first plurality of interaction variable values in which the interaction between the variable values and the first plurality of independent variables is considered A regression coefficient can be calculated for 320 ( 320 ).

For example, the prediction apparatus 200 may calculate a regression coefficient through Equation 1, which is a regression model, and a least squares method.

은 예측일(t일)의 n일 이전일(t-n일)의 최대 소비 전력,

은 예측일(t일)의 n일 이전일(t-n일)의 직전일에 대한 최대 소비 전력,

은 예측일(t일)의 n일 이전일(t-n일)의 일주일 전일에 대한 최대 소비 전력,

는 예측일(t일)의 n일 이전일(t-n일)의 이주일 전일에 대한 최대 소비 전력,

는 예측일(t일)의 n일 이전일(t-n일)의 요일,

는 예측일(t일)의 n일 이전일(t-n일)의 기온,

은 예측일(t일)의 n일 이전일(t-n일)이 속한 월,

은 예측일(t일)의 n일 이전일(t-n일)이 특수일에 해당하는지 여부,

은 예측일(t일)의 n일 이전일(t-n일)에 대한 일 지수,

는 j번째 변수값의 회귀 계수(j=1, 2, …12),

는 회귀 모형의 상수,

은 회귀 모형의 오차를 의미할 수 있다. 즉, 예측 장치(200)는 제1 복수의 독립 변수값들 및 제1 복수의 교호 작용 변수값들에 대해 최소 자승법을 적용하여 회귀 계수(

내지

)를 계산할 수 있다.

is the maximum power consumption on the day (tn day) n days before the forecast date (t day),

is the maximum power consumption for the day immediately preceding the day n days before the forecast date (day t) (day tn),

is the maximum power consumption for one week before the forecast date (day t) n days before the forecast date (day t),

is the maximum power consumption for two weeks before the forecast date (day t) n days before the day (day tn),

is the day of the week n days before the forecast date (day t) (day tn),

is the temperature of the day n days before the forecast date (day t) (day tn),

is the month in which the n days before the forecast date (day t) (day tn) belongs,

is whether the date n days before the forecast date (day t) (day tn) is a special day,

is the day exponent for the day n days before the forecast date (day t) (day tn),

is the regression coefficient of the j-th variable value (j=1, 2, …12),

is the constant of the regression model,

may mean the error of the regression model. That is, the prediction apparatus 200 applies the least squares method to the first plurality of independent variable values and the first plurality of interaction variable values to obtain a regression coefficient (

inside

) can be calculated.

내지

), 각 이전일에 대한 복수의 독립 변수값들(

,

,

,

,

,

,

,

) 및 복수의 교호 작용 변수값들(

,

,

,

)에 기초하여 최소 자승법을 통해 회귀 계수(

내지

)를 계산할 수 있다.For example, assuming that the forecast date is set to February 02, 2020 and the days preceding the forecast date are set to a total of 100 days, the day before the forecast date (the day immediately preceding the forecast date) is February 2020 01, and the day before one week before the forecast date may be January 25. The prediction device 200 is the maximum power consumption (

inside

), multiple independent variable values for each previous day (

,

,

,

,

,

,

,

) and multiple interaction variable values (

,

,

,

) based on the regression coefficient (

inside

) can be calculated.

The prediction apparatus 200 may calculate a prediction value based on the regression coefficient, the second plurality of independent variable values for the prediction date, and the second plurality of interaction variable values for the prediction date ( 330 ).

For example, the prediction apparatus 200 may calculate a predicted value for the maximum power demand on the prediction day through Equation 2, which is a regression model.

는 예측일(t일)의 최대 전력 수요에 대한 예측치,

은 예측일 직전일(t-1일)의 최대 소비 전력,

예측일 1주일 전일(t-7일)의 최대 소비 전력,

는 예측일 2주일 전일(t-14일)의 최대 소비 전력,

는 예측일의 요일,

는 예측일의 예상 기온,

는 예측일이 속한 월,

는 예측일이 특수일에 해당하는지 여부,

는 예측일의 전력 소비 추세인 일 지수,

는 j번째 변수값의 회귀계수(j=1, 2, …12),

는 회귀 모형의 상수를 의미할 수 있다.

is the forecast for the maximum power demand on the forecast day (day t),

is the maximum power consumption of the day immediately preceding the forecast date (day t-1),

The maximum power consumption of one week before the forecast date (day t-7),

is the maximum power consumption two weeks before the forecast date (day t-14),

is the day of the forecast date,

is the expected temperature on the forecast day,

is the month in which the forecast date belongs,

is whether the forecast date falls on a special day,

is the daily index, which is the electricity consumption trend of the forecast day,

is the regression coefficient of the j-th variable value (j=1, 2, …12),

may mean a constant of the regression model.

는 예측일이자(t일)의 최대 전력 수요(MW 단위)로 예측하고자 하는 대상일 수 있다.That is, the value of the dependent variable

may be a target to be predicted as the maximum power demand (in MW) of the forecast date (day t).

)뿐 아니라, 자기상관도가 높게 나타난 7일전(

), 14일전(

) 최대 전력 수요도 모형에 포함시킬 수 있다.With the plurality of independent variable values, the prediction apparatus 200 may directly reflect the maximum power demand before the prediction date in the regression model to reflect the autocorrelation. The prediction device 200 determines the maximum power demand (

) as well as 7 days ago when autocorrelation was high (

), 14 days ago (

) can also be included in the model for peak power demand.

)도 모형에 포함시킬 수 있다(

= {월, 화, 수, 목, 금, 토, 일}, 범주형 변수값). 예측 장치(200)는 앞서 자기상관성을 반영하기 위해 도입된 독립 변수값

과

를 통해 회귀 모형에 요일 요인을 중복해서 반영할 수 있다. 예측 장치(200)는 예측일의 예상 기온(

)으로 평균 기온인 서울지역의 일 평균 기온 자료를 이용할 수 있다.Prediction device 200 is the day of the week (

) can also be included in the model (

= {Mon, Tue, Wed, Thu, Fri, Sat, Sun}, categorical variable value). The prediction device 200 is an independent variable value introduced to reflect the autocorrelation.

class

Through this, the day factor can be reflected in the regression model in duplicate. The prediction device 200 is the predicted temperature (

), the daily average temperature data of the Seoul area, which is the average temperature, can be used.

={1월, 2월, 3월, 4월, 5월, 6월, 7월, 8월, 9월, 10월, 11월, 12월}, 범주형 변수). 예측 장치(200)는 예측일의 특수일 여부는 이진 변수

로 반영할 수 있다(t일이 특수일이면,

=1; 그렇지 않으면,

=0).The prediction device 200 shows the power demand differently depending on the season, and since it is further subdivided and also appears differently depending on the month, the regression model can include which month the prediction date belongs to as an independent variable value (

={January, February, March, April, May, June, July, August, September, October, November, December}, categorical variables). The prediction device 200 determines whether the prediction date is a special day or not is a binary variable.

(If day t is a special day,

=1; Otherwise,

=0).

변수로 반영할 수 있다. 예를 들어, 샘플 크기가 N개라면, 예측일의 일 지수

는 N+1값을 가질 수 있다. N개 샘플 중 예측일과 가장 멀리 떨어져 있는 t-N일의 일 지수

은 1 값을, 그리고 예측일과 가장 가까운 t-1일의 일 지수

은 N 값을 가질 수 있다.The prediction device 200 adds a work index to the regression model to reflect the power consumption trend factor.

It can be reflected as a variable. For example, if the sample size is N, the work index of the forecast day

may have a value of N+1. The day exponent of the tN days farthest from the predicted date out of N samples.

is a value of 1, and the day exponent of day t-1 closest to the forecast date.

may have N values.

을 포함할 수 있다. 예측 장치(200)는 회귀 모형에 기온이 전력 수요에 미치는 영향을 계절에 따라 다르게 반영하기 위해, 즉 예측일이 속한 달에 따라 기온의 영향을 다르게 반영하기 위해 교호 작용 변수값으로

를 포함할 수 있다. 예측 장치(200)는 특수일도 요일에 따라 그 영향이 다를 수 있기 때문에 이를 반영하기 위해 회귀 모형에 교호 작용 변수값으로

를 포함할 수 있다. 예측 장치(200)는 요일에 따른 수요의 등락이 요일만의 원인인지 기온의 영향도 포함되어 있는지를 좀 더 세분하게 반영할 수 있도록 회귀 모형에 교호 작용 변수값으로

를 포함할 수 있다.With a plurality of interaction variable values, the prediction device 200 converts the demand of the previous day into an interaction variable value so that the demand of the previous day can be differently reflected in the regression model according to the day of the week to which the previous day belongs.

may include. The prediction device 200 is an interaction variable value in order to reflect the effect of temperature on power demand differently according to seasons in the regression model, that is, to reflect the effect of temperature differently depending on the month to which the prediction date belongs.

may include. Since the prediction device 200 may have different effects depending on the special day and the day of the week, it is used as an interaction variable value in the regression model to reflect this.

may include. The prediction device 200 is used as an interaction variable value in the regression model to reflect in more detail whether the fluctuations in demand according to the day of the week are the cause of only the day of the week or whether the effect of temperature is also included in the regression model.

may include.

4 to 9 are diagrams for explaining a plurality of independent variables and interaction variables used by the prediction apparatus.

The maximum daily power demand can be a major factor influencing autocorrelation, day of the week, whether there is a special day, season, power consumption trend, temperature, etc.

1. An example of a plurality of independent variables

(1) autocorrelation

An autocorrelation of daily peak power demand may mean that today's demand is not significantly different from yesterday's demand. Electricity and energy used by the whole nation every day will inevitably be greatly affected by the demand of the previous day, unless the domestic population or industrial scale changes overnight. This autocorrelation can also be confirmed statistically.

4 shows an example of an auto-correlation function (ACF) graph shown using daily maximum power demand data from 2006 to 2016. FIG. That is, the daily maximum power demand has a high correlation with the demand one day ago, and may also show a high correlation with the demand 7 days ago and 14 days ago on the same day.

Accordingly, in order to calculate the regression coefficient, the prediction device 200 calculates the regression coefficient of the day immediately preceding (tn-1) and/or 7 days before the previous day (tn-7) of at least one day before the previous day (tn) of the prediction date (t). Maximum power consumption is available. In addition, the prediction apparatus 200 may use the maximum power consumption 14 days before (t-n-14) of the previous day (t-n) and/or the maximum power consumption 21 days before (t-n-21) of the previous day (t-n).

The prediction device 200 calculates a forecast value for the maximum power demand on the forecast day, the maximum of the day (t-1) immediately before the forecast date (t) and/or the maximum of 7 days before the forecast day (t) (t-7). power consumption is available. Also, the prediction device 200 may use the maximum power consumption 14 days before the prediction date t (t-14) and/or the maximum power consumption 21 days before the prediction date t (t-21).

(2) Day of the week and whether it is a special day

As of 2017, more than half or 56% of domestic electricity sales may have been sold for industrial use. For this reason, the pattern of peak power demand can be strongly influenced by industrial power demand. In other words, the peak demand for electricity may have a large difference between weekdays when workers' work is concentrated and weekends when they are not.

5 shows an example of a bar graph showing the average value of the daily maximum power demand from 2006 to 2016 for each day of the week. As can be seen in FIG. 5 , the peak power demand on weekdays is similarly at around 61,000 MW, while the demand on weekends is around 53,000 MW, which is clearly lower than weekdays. In addition, the peak power demand may show a significant difference even between Saturday and Sunday, which are the same weekend.

Accordingly, the prediction apparatus 200 may use a day of the week of the previous day (t-n) of at least one day before the prediction date (t) to calculate the regression coefficient. The prediction device 200 may use the day of the week of the prediction day to calculate a prediction value for the maximum power demand on the prediction day.

In order to calculate the regression coefficient, the prediction apparatus 200 may use whether or not the prediction date t is a special day of at least one day or more before the previous day t-n. The prediction device 200 may use whether the prediction day t is a special day in order to calculate a prediction value for the maximum power demand on the prediction day. Special days may include all days when many businesses are closed, such as national holidays, alternative holidays, election days, and temporary holidays.

On special days, like weekends, demand for electricity may decrease. The average peak power demand for a special day could be around 52,000 MW for 11 years from 2006 to 2016. This may be at a level similar to the daily maximum power demand on Sunday shown in FIG. 5 . If the forecast date does not reflect whether the forecast date is a special day or not, the forecast for the special day may be highly likely to deviate significantly. In addition, when a sample including a special date is used for estimating the regression coefficient, an error in the estimation of the regression coefficient may increase if it is not recognized whether the sample is a special date.

(3) season

The daily peak power demand varies greatly depending on the season, and the peak power demand can be particularly high in summer and winter.

Accordingly, in order to calculate the regression coefficient, the prediction apparatus 200 may use which season belongs to at least one day (t-n) prior to the prediction date (t). The prediction device 200 may use which season the prediction day t belongs to in order to calculate a prediction value for the maximum power demand on the prediction day.

6 is a view showing the average value of the daily maximum power demand from 2006 to 2016 by month, and FIG. 7 is a diagram showing the average value of the daily maximum power demand from 2006 to 2016 by spring/summer/autumn/winter. The prediction apparatus 200 may classify the composition of the season into spring/summer/autumn/winter. The prediction apparatus 200 may classify the composition of the season by subdividing it into monthly or 24 seasons.

Referring to FIG. 6 , it can be seen that there is quite a difference in the maximum power demand for each month even in the same season. Referring to FIG. 7 , it can be seen that the maximum power demand is large in the order of winter, summer, autumn, and spring. That is, the prediction apparatus 200 may calculate an accurate regression coefficient and a predicted value by using which season or month the prediction date belongs.

(4) Power consumption trend

Forecasts for a short period of time, such as forecasting daily peak power demand, may generally not take into account trend factors. This may be because demand close to the forecast date is directly reflected.

However, the prediction device 200 may use a work index that is a power consumption trend of at least one day prior to the prediction day t. The prediction device 200 may use a work index, which is a power consumption trend, in order to calculate a predicted value for the maximum power demand on the forecast day. For example, when the size of the sample used increases by a certain level or more, the prediction apparatus 200 may use a factor of the power consumption trend to calculate a predicted value for the maximum power demand on the prediction day.

If, when estimating the regression coefficient using samples of 2 or 3 years, if the power consumption trend is not reflected, the maximum power consumption of the day before the forecast date or the maximum power consumption 2 or 3 years away from the forecast date or the same weight is used in the calculation of the regression coefficient, which may lead to inaccurate results.

FIG. 8 is a view showing the maximum power consumption for the day in chronological order by finding the day on which the maximum daily maximum power consumption occurs for each month for 11 years from 2006 to 2016. FIG. As can be seen from FIG. 8 , it can be seen that the demand for power consumption steadily increases year by year. Therefore, the prediction device 200, for example, when using the maximum power consumption per day (that is, about 700 to 1200 samples) for about 2 to 3 years, the difference in the average demand level that occurs in the time difference between the maximum power consumption Regression coefficients and forecasts can be calculated using the power consumption trend representing

(5) temperature

Temperature may be the most well-known factor influencing peak power consumption. In recent years, with the spread of air conditioners that use electricity, it is not difficult to find news that the maximum power consumption has been updated in the summer (or winter) when heat waves (or extreme cold) are continuous.

9 is a diagram illustrating a scatter diagram of daily average temperature and daily maximum power consumption from 2006 to 2016. FIG. 9 may have been prepared using daily average temperature data in the Seoul area. Referring to FIG. 9 , it can be seen that there are many days when the maximum power consumption is also high when the average daily temperature is approximately 30 degrees Celsius and below zero. As such, it can be seen that the temperature has a direct effect on the maximum power consumption. However, it may be noted that the effect of temperature varies depending on the season in which the forecast date belongs. That is, if the temperature is high in summer, the maximum power consumption increases, but if the temperature is high in winter, the maximum power consumption decreases.

The prediction apparatus 200 may accurately calculate a regression coefficient and a predicted value by using the influence of the temperature by dividing the seasons.

2. An example of multiple interaction variables

An interaction effect in which the effect of one factor depends on the level of another factor may appear among the plurality of independent variables affecting the maximum power demand described above with reference to FIGS. 4 to 9 .

In other words, it can be seen that the direction of the influence changes depending on which time of year the forecast date belongs to, even for temperature, among the plurality of independent variables. Accordingly, the prediction apparatus 200 may calculate a regression coefficient and a predicted value using a plurality of interaction variable values in which an interaction between a plurality of independent variables is considered.

For example, the prediction apparatus 200 uses the maximum power consumption of the day immediately preceding the prediction date to use the autocorrelation among the plurality of independent variables, and the effect of the autocorrelation factor varies depending on the day of the week to which the prediction date belongs. can be considered. That is, in the prediction device 200, if the prediction day (t) and the day immediately preceding the prediction day (t-1) both belong to weekdays, the level of maximum power consumption will not be significantly different, but the prediction day (t) is Monday and If the day immediately preceding (t-1) of the forecast date is a Sunday, considering that the immediately preceding day (t-1) is a holiday, the demand on the forecast day (t) is higher than that of the previous day (t-1). can That is, when using autocorrelation to calculate the regression coefficient and the predicted value, the prediction apparatus 200 may reflect the characteristics of the day of the week when calculating the regression coefficient and the predicted value if the prediction date is Monday, Sunday, or Saturday.

Since the reflection effect on the special day may also vary depending on which day of the week the special day belongs to, the prediction device 200 may have a special day on a weekend, if it is a holiday because it is Friday or Monday, if it is located in the middle of a weekday, etc. In order to reflect the effect of a special day differently, the interaction between whether a special day and the day of the week factors can be reflected in the prediction value calculation.

10 is a diagram illustrating an example of a ANOVA result of a prediction apparatus, and FIG. 11 is a diagram illustrating an example of calculating an adjustment determination coefficient of the prediction apparatus.

For the analysis of variance, a total of 4,018 observations from January 1, 2006 to December 31, 2016 were used based on the dependent variable. 10 may show a result of analysis of variance for the prediction value calculated by the prediction apparatus 200 .

It can be seen from FIG. 10 that all variables except the temperature factor show very small significance probabilities of less than 0.0001. The significance probability of the temperature factor may also be very low. That is, it can be said that the plurality of interaction variable values and the plurality of independent variable values that are interaction factors used by the prediction apparatus 200 are all statistically significant. It can be confirmed that the prediction apparatus 200 exhibits high explanatory power through the adjusted determination coefficient (Adjusted R ^{2 ) value of 0.9722.}

11 may show an example of calculating each adjustment determination coefficient from 2006 to 2016 in order to check whether the prediction apparatus 200 shows high explanatory power regardless of the year. It can be seen that the high explanatory power of the prediction apparatus 200 is consistent in that the adjusted coefficient of determination for each year shown in the table of FIG. 11 shows an adjusted coefficient of determination of 0.93 or more in all years.

12 is a diagram for explaining a performance verification result of a prediction apparatus, and FIG. 13 is a diagram illustrating a result of calculating a prediction value for the same period in the prediction apparatus and the conventional prediction method.

As an index for comparing the performance of the prediction apparatus 200, the mean absolute percentage error (MAPE), which is most used in demand prediction research, may be used.

In order for the prediction apparatus 200 to predict the maximum power demand on a prediction day, when calculating a regression coefficient, how much a sample size is to be designated may be specified. For performance validation, the regression coefficients can be calculated using the 800 observations immediately before the prediction date as samples.

The first conventional prediction method predicted the maximum daily electricity demand based on the regression model, and may present the MAPE for the daily maximum electricity demand prediction result for 7 years from 2010 to 2016. For comparison of the performance of the prediction apparatus 200 and the performance of the first existing prediction method, the prediction value calculation result was derived, and the MAPE for the prediction value calculation result may be summarized in the table of FIG. 12 . As shown in the table of FIG. 12 , it can be seen that the prediction apparatus 200 not only shows better performance than the first conventional prediction method over all periods, but also shows a very low MAPE of less than 2% over all years. .

Recently, a number of studies using deep learning techniques to predict daily maximum power have been published. The second conventional prediction method may predict the maximum daily power demand in 2016 using a radiated basis function neural network.

The third conventional prediction method predicts the maximum daily power demand for three months from October 1 to December 31, 2016 using LSTM (Long Short Term Memory), which is known to be suitable for time series data estimation among deep learning techniques. can The third conventional prediction method may provide the MAPE value showing the best result by repeatedly performing the LSTM five times.

The fourth conventional prediction method may predict the maximum power for days excluding weekends and public holidays during the winter period from December 21, 2016 to February 28, 2017 by using a deep neural network (DNN). The fourth conventional prediction method may have applied the DNN 25 times and provided the average value of the MAPE.

Referring to the table of FIG. 13 , the second conventional prediction method showed a very high prediction performance of 2.16% MAPE, and it can be seen that the prediction apparatus 200 shows a prediction performance improved by 0.2% in terms of MAPE. . In addition, it can be seen that the third conventional prediction method shows an excellent performance of 1.64%, which is improved by 1% or more in terms of MAPE, whereas the MAPE is 2.74%.

The MAPE of the prediction apparatus 200 may show a very slight difference of 0.01% from the fourth conventional prediction method. However, the prediction device 200 can be applied for any period regardless of the season of the year, regardless of weekdays and holidays, whereas the fourth conventional prediction method shows almost the same performance as the deep learning technique specialized only on weekdays in the winter season of a specific year. In this respect, the superiority of the prediction apparatus 200 can be confirmed. In addition, it is worth noting that the error rate of the fourth conventional prediction method is the average MAPE obtained after 25 applications. In addition, in order for the fourth conventional prediction method to perform DNN deep learning, it is necessary to set various hyperparameters such as the number of layers, the number of neurons per layer, learning rate, and an activation function. Therefore, it can be seen that the prediction value calculation of the prediction apparatus 200 has a strength in terms of execution power compared to the fourth conventional prediction method.

14 is a diagram illustrating the effect of interaction variables used by a prediction device on prediction performance.

,

,

,

)가 각각 어느 정도 효과가 있는지 확인하기 위해 각 교호 작용 변수를 제외한 동작으로 예측한 결과를 이용할 수 있다The prediction apparatus 200 may exhibit a high level of predictive power by including the interaction factor in the regression model. An interaction analysis experiment for interaction variables is an example of a plurality of interaction variables used by the prediction device 200, which is an interaction variable (

,

,

,

) can be used to determine how much effect each

The table of FIG. 14 shows the MAPE of the prediction apparatus 200 for each year from 2006 to 2016, and the MAPE from which each interaction variable is excluded. It can be seen that the overall MAPE of the prediction device 200 including all interaction variables shows the highest predictive power of 1.66%. The last column of the table of FIG. 14 shows the prediction results excluding all interaction variables. Since there is a difference of about 0.4% in terms of MAPE and the prediction apparatus 200 including all interaction variables, it can be considered that a plurality of interaction variables contributed to the improvement of the error rate.

변수가 제외된 결과가 오차율이 가장 크게 증가한 것을 볼 수 있다. 이는

변수의 예측 공헌도가 교호 작용 중에는 가장 크다는 것을 의미하며 기온 요인을 모형에 반영할 때 월별 또는 계절별로 다르게 고려할 수 있도록 교호 작용 변수를 포함시키는 것이 필요하다는 것을 알 수 있다. 마찬가지로, 오차율의 증가량을 비교하면 나머지 교호 작용 변수들 중

,

,

순으로 예측 공헌도가 크다는 것을 알 수 있다.Among the four interaction variables,

It can be seen that the result of excluding the variable has the largest increase in the error rate. this is

It means that the predictive contribution of the variable is the largest during the interaction, and it can be seen that it is necessary to include the interaction variable so that it can be considered differently by month or season when the temperature factor is reflected in the model. Similarly, if we compare the increase in the error rate, among the remaining interaction variables,

,

,

It can be seen that the prediction contribution is large in that order.

The prediction device 200 considers temperature factors, autocorrelation factors, day factors, and seasonal factors, as well as differences in temperature sensitivity according to seasons and differences in autocorrelation according to days of the week, as interaction factors. The prediction device 200 calculates the predicted value for the maximum power demand for 4,018 days for 11 years from 2006 to 2016 and compares the prediction value derived from the existing prediction methods with the MAPE. As a result, the prediction device 200 is excellent. It can be seen that the predictive performance is shown.

That is, the prediction apparatus 200 includes the interaction between the plurality of independent variables as interaction variable values in the prediction value calculation, so that the interaction between the plurality of independent variables is effective in lowering the prediction error rate through empirical analysis. can be checked

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc. alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the embodiment, or may be known and available to those skilled in the art of computer software. Examples of the computer-readable recording medium include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic such as floppy disks. - includes magneto-optical media, and hardware devices specially configured to store and carry out program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include not only machine language codes such as those generated by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the embodiments, and vice versa.

The software may comprise a computer program, code, instructions, or a combination of one or more thereof, which configures a processing device to operate as desired or is independently or collectively processed You can command the device. The software and/or data may be any kind of machine, component, physical device, virtual equipment, computer storage medium or device, to be interpreted by or to provide instructions or data to the processing device. , or may be permanently or temporarily embody in a transmitted signal wave. The software may be distributed over networked computer systems, and stored or executed in a distributed manner. Software and data may be stored in one or more computer-readable recording media.

As described above, although the embodiments have been described with reference to the limited drawings, those skilled in the art may apply various technical modifications and variations based on the above. For example, the described techniques are performed in a different order than the described method, and/or the described components of the system, structure, apparatus, circuit, etc. are combined or combined in a different form than the described method, or other components Or substituted or substituted by equivalents may achieve an appropriate result.

Therefore, other implementations, other embodiments, and equivalents to the claims are also within the scope of the following claims.

Claims (18)

Receiving a forecast request message requesting a forecast for the maximum power demand of the forecast day;
In response to the prediction request message, a first maximum power consumption for at least one day or more prior to the prediction date, a first plurality of independent variable values that are values of a first plurality of independent variables affecting the first maximum power consumption and the first plurality of independent variable values and the first plurality of interaction variable values based on a first plurality of interaction variable values in which an interaction between the first plurality of independent variables is considered. calculating a regression coefficient for and
a second plurality of independent variable values that are values of a second plurality of independent variables affecting the regression coefficient, the predictive value, and a second plurality of interaction variable values in which an interaction between the second plurality of independent variables is considered calculating the predicted value based on
A method of forecasting peak power demand, including
According to claim 1,
Calculating the regression coefficient comprises:
generating a first plurality of interaction variable values in which an interaction between the first plurality of independent variables is considered based on the values of the first plurality of independent variables;
A method of forecasting peak power demand, including
3. The method of claim 2,
The values of the first plurality of independent variables are
a second maximum power consumption for a day immediately preceding the at least one day before the previous day, a third maximum power consumption for a week before the at least one day before the previous day, a day of the week to which the at least one previous day belongs, The temperature of the at least one or more previous days, the month to which the at least one or more previous days belong, whether the at least one or more previous days correspond to special days, and a first power consumption trend for the at least one or more previous days containing the day exponent
How to forecast peak power demand.
4. The method of claim 3,
The generating of the first plurality of interaction variable values comprises:
generating a first interaction variable value using the second maximum power consumption and a day of the week to which the at least one previous day belongs;
generating a second interaction variable value using the temperature of the at least one previous day and the month to which the at least one previous day belongs;
generating a third interaction variable value using whether the at least one previous day corresponds to the special day and a day to which the at least one or more previous day belongs; and
generating a fourth interaction variable value using the day of the week to which the at least one previous day belongs and the temperature of the at least one previous day
A method of forecasting peak power demand, including
According to claim 1,
Calculating the predicted value comprises:
generating a second plurality of interaction variable values in which an interaction between the second plurality of independent variables is considered based on the values of the second plurality of independent variables;
A method of forecasting peak power demand, including
6. The method of claim 5,
The values of the second plurality of independent variables are
The fourth maximum power consumption for the day immediately preceding the forecast date, the fifth maximum power consumption for one week before the forecast date, the day of the week to which the forecast date belongs, the expected temperature of the forecast day, the month to which the forecast date belongs, the Whether the forecast date corresponds to a special day, and a second day index that is a power consumption trend for the forecast date
How to forecast peak power demand.
7. The method of claim 6,
The generating of the second plurality of interaction variable values comprises:
generating a fifth interaction variable value using the fourth maximum power consumption and a day of the week to which the predicted date belongs;
generating a sixth interaction variable value using the temperature of the prediction day and the month to which the prediction date belongs;
generating a seventh interaction variable value using whether the prediction date corresponds to the special day and the day of the week to which the prediction date belongs; and
Generating an eighth interaction variable value using the day of the week to which the prediction date belongs and the expected temperature of the predicted day
A method of forecasting peak power demand, including
7. The method of claim 3 or 6,
The above special days include national holidays, substitute holidays, election days, and temporary holidays.
How to forecast peak power demand.
According to claim 1,
Receiving the first maximum power consumption, the first plurality of independent variable values, and the second plurality of independent variable values from a server
Maximum power demand forecasting method further comprising.
a memory for storing instructions for predicting peak power demand; and
a processor for executing the instructions
including,
When the instructions are executed by the processor, the processor
receiving a forecast request message requesting a forecast for the maximum power demand on a forecast day;
In response to the prediction request message, a first maximum power consumption for at least one day or more prior to the prediction date, a first plurality of independent variable values that are values of a first plurality of independent variables affecting the first maximum power consumption and the first plurality of independent variable values and the first plurality of interaction variable values based on a first plurality of interaction variable values in which an interaction between the first plurality of independent variables is considered. Calculate the regression coefficient for
a second plurality of independent variable values that are values of a second plurality of independent variables affecting the regression coefficient, the predictive value, and a second plurality of interaction variable values in which an interaction between the second plurality of independent variables is considered to calculate the predicted value based on
Peak power demand forecasting device.
11. The method of claim 10,
The processor is
generating a first plurality of interaction variable values in which an interaction between the first plurality of independent variables is considered based on the values of the first plurality of independent variables;
Peak power demand forecasting device.
12. The method of claim 11,
The values of the first plurality of independent variables are
a second maximum power consumption for a day immediately preceding the at least one day before the previous day, a third maximum power consumption for a week before the at least one day before the previous day, a day of the week to which the at least one previous day belongs, The temperature of the at least one or more previous days, the month to which the at least one or more previous days belong, whether the at least one or more previous days correspond to special days, and a first power consumption trend for the at least one or more previous days containing the day exponent
Peak power demand forecasting device.
13. The method of claim 12,
The processor is
generating a first interaction variable value using the second maximum power consumption and a day to which the previous day of at least one day belongs;
generating a second interaction variable value using the temperature of the at least one previous day and the month to which the at least one previous day belongs,
generating a third interaction variable value using whether the previous day of at least one day or more corresponds to the special day and a day to which the previous day of at least one day or more belongs;
generating a fourth interaction variable value using the day of the week to which the at least one previous day belongs and the temperature of the at least one previous day
Peak power demand forecasting device.
11. The method of claim 10,
The processor is
generating a second plurality of interaction variable values in which an interaction between the second plurality of independent variables is considered based on the values of the second plurality of independent variables;
Peak power demand forecasting device.
15. The method of claim 14,
The values of the second plurality of independent variables are
The fourth maximum power consumption for the day immediately preceding the forecast date, the fifth maximum power consumption for one week before the forecast date, the day of the week to which the forecast date belongs, the expected temperature of the forecast day, the month to which the forecast date belongs, the Whether the forecast date corresponds to a special day, and a second day index that is a power consumption trend for the forecast date
Peak power demand forecasting device.
16. The method of claim 15,
The processor is
generating a fifth interaction variable value using the fourth maximum power consumption and the day of the week to which the predicted date belongs;
A sixth interaction variable value is generated using the temperature of the forecast day and the month to which the forecast date belongs,
A seventh interaction variable value is generated using whether the prediction date corresponds to the special day and the day of the week to which the prediction date belongs,
generating an eighth interaction variable value using the day of the week to which the forecast date belongs and the expected temperature of the forecast day
Peak power demand forecasting device.
16. The method of claim 12 or 15,
The above special days include national holidays, substitute holidays, election days, and temporary holidays.
Peak power demand forecasting device.
11. The method of claim 10,
The processor is
receiving the first maximum power consumption, the first plurality of independent variable values, and the second plurality of independent variable values from a server
How to forecast peak power demand.

Images