Non-intrusive Load Identification Method Based on Feature Optimization and Bayesian Algorithm

Abstract

With the gradual improvement of virtual power plant marketization and demand side response mechanism,the demand for accurate identification of user load power data is also increasing.The current research method lacks the discussion of the comprehensiveness and redundancy of load feature combination.This paper establishes a set of non-intrusive load monitoring and identification methods for this problem,including event detection,feature extraction,feature selection and load identification.First,the CUSUM event monitoring algorithm based on the sliding window was improved to detect the load switching events and isolate the single load information.A total of 84-dimensional load features were extracted,including steady-state current,power,harmonic features,U-I curve shape features,and transient features.Measure the correlation between features and categories and the redundancy between features.Then the Bayesian classification algorithm model is established,with the selected features as input,and the load category identification is completed.Using the publicly available dataset WHITED,the present method selects the optimal feature subset,analyzes the influence of different feature dimensions on the accuracy rate,with fast identification speed and high accuracy rate,and realizes all the functions from event monitoring to load identification.