Non-Intrusive Adaptive Load Identification Based on Siamese Network

The traditional non-intrusive load monitoring (NILM) algorithms are mostly based on classification models, which have several deficiencies. Firstly, a large amount of labeled data is required to train the model. Secondly, these cannot identify unknown devices that frequently encountered in practical application. Finally, models poor performance versatility, means they only adapt trained data. These shortcomings greatly influence practicality NILM algorithms. To tackle problems, this paper has proposed adaptive identification model Siamese network, uses both V-I trajectory and active power as signatures. network utilized calculate similarity trajectory, realized by matching signature with feature library. Through adding new features library dynamically, can be realized. In addition, typical for few-shot learning, thus small number samples achieve ideal recognition effect. At last, validity versatility verified PLAID dataset COOLL dataset.