Research on AP Clustering of Weighted Load Index for Demand Response Potential Analysis

Abstract

Data mining techniques can help utilities to analyze the electricity utilization behavior and make decisions based on massive data from smart meter. In this paper, an innovative Affinity Propagation clustering algorithm is proposed and improved in load curve clustering to analyze demand response potential effectively. First, the feature indices of load curve are extracted to reduce data dimension. In the face of the weakness of subjective weighing, Entropy Weighting is introduced to objectively determine the weight of load feature indices. Then, Preference parameter of AP clustering algorithm is adjusted by local search mechanism to improve the similarity calculation, and the convergence of clustering result is ameliorated by DB index. Based on the clustering result, the feature indices are combined to analyze the demand response potential of customer. Case studies show that the proposed algorithm has good performance in both the clustering efficiency and effect, and the extracted features are beneficial to consumption behavior analysis.