A non-intrusive load monitoring algorithm based on multiple features and decision fusion

With the large-scale deployment of smart meters and wide application of various machine learning algorithms, non-intrusive load monitoring (NILM) has attracted the attention of academia and industry. However, machine learning algorithms often suffer from high variability in load identification performance due to different features. In view of the poor generalization ability and low accuracy of load identification using an individual feature or classifier model, this paper proposes a novel NILM method that accomplishes deep feature fusion and classifier model fusion by an improved Dempster–Shafer (D–S) evidence theory. Firstly, we extract the power features, current harmonic features, and voltage–current (V–I) trajectory features from the input signals. Then, K-nearest neighbor (KNN), random forest (RF), and convolutional neural networks (CNN) are employed to identify load appliances using three individual features. Finally, the probability estimates of each classifier are transmitted to the aggregator for aggregation to obtain the final identification results by the improved D–S evidence theory. The experimental results on the PLAID dataset show that the proposed method can significantly improve identification performances.