Non-intrusive Load Identification Method Based on Integrated Intelligence Strategy

Abstract

Existing non-intrusive load identification methods are mostly single recognition models that based on a certain feature space. It is difficult to guarantee the global generalization performance of the model in the complex feature space. In this paper, the distribution of features of common household appliances in different feature space is analyzed and a variety of feature spaces are constructed. The Extreme Learning Machine model is selected as the base learner to train the optimal Extreme Learning Machine recognition model in the corresponding feature space. The AdaBoost algorithm is used to combine the base learner and build the ensemble extreme learning machine load recognition model. The experimental results based on the measured data showed that the accuracy of the proposed non-intrusive load identification method is obviously superior to the classical Extreme Learning Machine model and the Support Vector Machine model.