Research And Analysis On The Forecasting Of The Monthly Electricity Sales Based On The Optimal Combination Method

Electricity sale is the main product of the power supply enterprise's operation,is the core standard for the economic benefits of power supply enterprises.The electricity sale data reflect the prospect of the community economy.The calculating of a series of indicators,such as electricity sales price,sales profits and line loss,are deeply influenced by it.Its increasing and decreasing trend is influenced by many direct and indirect factors,such as region,climate,major festivals and holidays,iron and steel price index,distributed energy interconnecting to the grid,national policies,internal strategies of power supply enterprises,etc.The forecast of the electricity sales is also affected by the data information and forecasting models,which produce forecasting errors and reduce the accuracy.Therefore,enough attention should be paid to the analysis and forecast of electricity sales and to improve the accuracy of forecast.It involves the upgrading of Standard Target System indicators such as electricity charge,electricity price,measurement and power demand side management,also concerns the safety of power system dispatching,the operation mode of power grid,co-ordination of supply and demand,the strategic decision and deployment of marketing strategy and the formulation of purchase and sale electricity plan.It can even provide practical guidance for power plants.It is of great significance to promote the vigorous development of the power market.This paper analyzes monthly electricity sale data for 60 months from 2013 to 2017 by means of diagrams,form and so on.BP neural network forecasting model and RBF neural network forecasting model are used to get the predicted values of 12 months in 2017,respectively.The predicted values based on the two forecasting models are compared with the actual electricity sales,then corresponding weights are allocated according to the error variance countdown method.By the optimal combination forecasting model,we get the preliminary forecasting data.Considering the influence of the Spring Festival,National Day and other reasons on electricity sales as well as the corresponding hysteresis,the electricity sales in March and December with larger forecasting errors are corrected by the monthly-to-quarterly ratio method.That is,forecast the ratio between the electricity salesin January and the quarter electricity sales,then multiply the ratio by the forecasting quarter electricity sales.The result is the predicted value in March.Using the same method,we can get the predicted value in December.This paper also analyzes various reasons that affect the electricity sales and forecast accuracy,including PV generation,electric energy substitution and policies.Concerning the actual situation,it explains and puts forward suggestions and corrected formula for optimizing and adjusting the forecasting model.The prediction shows that the minimum relative error of the 12 months in 2017 is 0.17%,the maximum is 4.46%,the mean relative error is 1.37% and the forecasting accuracy is98.63%.Compared with the separate use of BP neural network forecasting model and the RBF neural network forecasting model,the prediction accuracy is greatly improved,which shows that the method has a high accuracy and reliability for the forecasting of the monthly electricity sales produced by the existing load characteristics in Qinhuangdao area.