The Electricity Sales Forecasting Based on Leading Analysis and Factor Compensation

Electricity sales forecasting is a fundamental work for State Grid Corporation of China. Currently, after data preprocessing, X13 seasonal adjustment is applied to decompose electricity sales series into trend, season and random items. The decomposed items are predicted respectively to produce accurate forecasting results. On the basis of it, a new electricity sales forecasting method based on leading analysis and factor compensation is proposed in this paper. First of all, the leading factors are introduced to improve the ability in identifying sudden change of trend item. Then, the impact of the Spring Festival in random item is eliminated and the prediction of random item is categorized into normal month and abnormal month. Under normal month, the random item is forecasted by combining Mixture of Experts model with factor data and processed by Kalman Filter; under abnormal month, the prediction results are obtained by the method of factor elimination and compensation. Finally, combined with the impact of the Spring Festival, the adjustment of electricity sales forecasting result on January to March is achieved. The actual validation shows that the prediction accuracy and robustness of the proposed model is greatly improved.