Research And Application Of The Key Technologies For Forecasting Electricity Sales Considering Curve Characteristics And Influencing Factors

Electricity sales is the main economic evaluation indicator for power grid enterprises,meanwhile monthly electricity sale forecast is an important daily work for power grid enterprise’s marketing department.So for power grid enterprise,accurate monthly electricity sale forecast will be of great importance in providing marketing decision support,making plans for expanding power supply and broadening marketing scope,conducting electric power replacement,implementing orderly power utilization schemes,and improving the quality of customer service.At present,comparative analysis method,structural analysis method,regression analysis method as well as neural network method are adopted to make monthly electricity sale forecast by the most power grid enterprise.These methods can predict the electricity sales to some extent,but the overall electricity prediction accuracy of the State Grid Corporation of China(SGCC)is not ideal.This is because the curves of electricity sales from different provinces own different characteristics,which is not considered in the study.Only one prediction algorithm is utilized to predict electricity sales of many provinces,and that will inevitably make the prediction accuracy be poor.In order to solve the aforementioned problems,two methods are proposed in this paper.One is based on the historical curve of electricity sales forecasting method.It is to cluster 27 provincial or municipal electric power companies according to the curve characters of them in time-domain and frequency-domain.For different categories of company,the forecasting methods is chosen according to the suitability between the curve characters of electricity sales and forecasting methods.For the same category of companies,the same forecasting method is adopted.Based on the forecasting methods of historical curve,the influencing factors such as weather,economic,holiday and social events are taking into consideration.Based on support vector machine(SVM)regression,the amendment model for electricity sales forecasting is built,which can further improve the prediction accuracy.Another way is to consider Chinese New Year factors of the electricity sales adjustment method.In this method,the electricity sales of the first quarter is assigned to the January,February and March according to the ratio of the monthly electricity sales to the corresponding the quarterly electricity sales.Utilizing the electricity sales data of SGCC from 2010 to 2014 as historical data,the monthly electricity sales of SGCC in 2015 is conducted by the method proposed in this paper,and the average prediction error is 1.78%.The results demonstrate that the proposed method is effective and high precision.