Comprehensive Forecasting Method Of Monthly Electricity Consumption Based On Time Series Decomposition Method And Regression Analysis Method

As the foundation of power system operation,optimization and control,power consumption prediction is facing new challenges with the rapid development of integrated energy system and the gradual opening of power trading market.Traditional load forecasting technology has been widely studied by scholars.However,due to the influence of distributed energy,the user side power demand and load characteristics change compared with the past.At the same time,in the open and competitive power retail market environment,power consumption forecasting starts to face the power demand of small-scale users,which is more vulnerable to the disturbance of seasons,holidays and economic factors.Therefore,the traditional load forecasting method is not suitable for power consumption forecasting.In order to solve the above problems,this paper studies the comprehensive prediction method of electricity consumption based on time series decomposition method and regression analysis method.The main research results are as follows:(1)Analyzed the electricity consumption characteristics of typical customers in a city in the north and its influencing factors.Based on the load data of a city in the north,the hierarchical clustering model and the density clustering model were used to mine the data of 30 kinds of loads.(2)Considering the impact of seasonal abrupt changes and major holidays,proposed a comprehensive prediction method of monthly electricity consumption based on STL(Seasonal and Trend decomposition using Loess)decomposition model.Step 1: use the characteristics of STL decomposition model to set the change rates of seasonal components in different months,conduct personalized decomposition of the power consumption sequence of corresponding months under the influence of seasonal mutations and major holidays,and decompose the electricity consumption sequence into trend component,seasonal component and random component,so as to avoid mutual interference between the predicted components.The BP neural network model was used to predict the seasonal components in the months with abrupt seasonal changes and major holidays.For the seasonal components in the stationary months,the historical corresponding value was directly used as the predicted value.For the trend component,ARIMA(Autoregressive Integrated Moving Average)model is used for prediction.For random components,mean value prediction is used.Step 3: reconstruct the predicted values of the above three components into the predicted final electricity consumption.This paper adopts R language to compile the algorithm,and verifies and analyzes the effectiveness of the proposed method through the actual monthly electricity consumption data of a university park in north China.(3)On the basis of(2),considering the influence of economic factors,a monthly electricity consumption prediction method based on X12 and STL decomposition model is proposed.Based on the correlation between electricity consumption and economic development,the improvement measures of monthly electricity consumption prediction method are proposed.First,the STL model and the X12 model were used to decompose the power consumption sequence and GDP sequence respectively,and the trend component of GDP was taken as the influencing factor of the monthly power consumption trend component into the VAR(Vector autoregression)model for the power consumption trend prediction.Then the BP neural network was used to predict the seasonal component and the mean value method was used to predict the random component.