Short-term Load Prediction Based on User Behaviors Analysis

Short-term load prediction is the foundation of dispatch and decision for urgent demand response of power grid. However, traditional load prediction approaches feature not so good adaption due to high random behaviors of individual household loads. In this paper, user home appliance data within extremely short time interval are acquired with help of non-intrusive load identification technique. By in-depth analysis of household home appliance operation characteristics, differentiate parameters and relevant edge values of various states are determined, resulting in state space set of specific home appliance. Time-domain probability distribution function and density function of specific home appliance are derived based on user behavior analysis model construction using continuous-time Markov model. State transition probability matrix and rate matrix are obtained, too. User short-term load prediction procedures are constructed by applying solution of Kolmogorov forward equations. Finally, 100 typical households electricity consumption data of Jiangsu Province in China from July to September are sampled for analysis. The accuracy and effectiveness of proposed method is verified by analysis and comparison results with traditional load prediction method.