Data-driven combining forecasting method for the net demand of power retailers in distribution network

With the rapid development of China's electricity market and the commercial system for the sale, the net demand forecasting for power retailers in distribution network have attracted great concern. This paper proposes a data-driven technology, i.e., the adaptive combining algorithm, to improve the accuracy of predictive models. Using the technology of big data analysis, this paper proposes an anomaly-state recognition method based on fuzzy algorithm. Moreover, fuzzy clustering and hierarchical clustering method are used to search the optimal similar day. After that, the “Relief-Correlation Test” is executed to adaptively select input features for each sampling point. Finally, a novel variable-weight combining forecasting algorithm is proposed to predict the net demand of power retailers in distribution network. The proposed method is verified by a real distribution-network power retailer in China. Results show that the data-driven combining forecasting model proposed in this paper is more reliable and effective than other individual predictive models for the net demand forecasting of power retailers.