A Spatio-Temporal Customer Baseline Load Estimation Approach Based on LAD-LASSO Regression

Abstract

Customer baseline load (CBL) estimation plays a crucial role for customer compensation calculation in demand response (DR) process. Most current customer baseline load estimation methods only consider one of the spatial information in DR event day and the temporal information of DR participants. CBL estimation methods based on spatial information shows instability when load patterns between DR participants and CONTROL group customers are not similar. And CBL estimation methods based on temporal information show poor accuracy when customers' load pattern changes in DR event day. In order to improve the accuracy of CBL estimation in complex scenes, a spatio-temporal approach for CBL estimation is proposed in this paper. First, all customers are clustered by K-means algorithm, the non-DR customers in the same cluster is regarded as similar customers. Second, the spatio-temporal feature vectors are extracted from history load data of DR participants and similar customers' load data in DR event day. Third, for each DR customer, a linear function between feature vector and CBL is fitted by LAD-LASSO regression model, by which the CBL is estimated. A comparison with six well-known CBL estimation methods using a dataset of 450 residential customers indicates that the proposed approach has the best accuracy and robustness performance than other current CBL estimation methods.