Appliance Identification by Topological Data Analysis of Phase Space

Non-intrusive Load Monitoring (NILM) is a technology that identifies and analyzes electrical loads by examining changes in physical quantities such as voltage and current. With the increasing number and diversity of electrical loads today, extracting distinctive features and constructing corresponding classification models has become anincreasingly important issue for NILM. Although technological advancements have introduced deep learning algorithms to improve traditional feature extraction methods, the features extracted by these methods often lack clear mathematical significance a nd require considerable computation time. Topological Data Analysis (TDA) studies spatial properties that remain invariant under continuous deformations, which can reveal the structures and relationships within complex datasets such as networks, graphs, and manifolds. In this paper, we use TDA as a feature extraction method to quickly extract a large number of nonlinear shape features with clear mathematical significance from active power to identify loads. Experimental results on the public dataset PLAID, compared with other methods, demonstrate that our method can quickly identify loads and significantly improve t he identification re suits fo r certain complex appliances. Therefore, the proposed method in this paper is a novel NILM approach that balances computational speed, accuracy, and interpretability.