A mother-son model for multi-objective non-invasive load monitoring

Abstract

This paper proposes a mother-son model (MSM) to address the challenges of missing data, complex models, and limited transfer capabilities in non-intrusive load monitoring (NILM). The MSM model designs a novel transfer method based on feature inheritance to dynamically decompose multiple appliances (including unknown appliances). The mother model incorporates a six-parallel feature extraction structure to extract comprehensive features, which are then inherited by the son model. The features inherited from the mother model integrate information from multiple dimensions in the spatial domain and focus on local fine features in the temporal domain. The son model offers the capability to adjust the number of appliances and decompose unknown appliances, catering to the requirements of practical applications of NILM. Furthermore, a novel statistical metric, Ratio Squared Error (RSE), is designed to quantify the proportion of squared error in energy consumption. The experiments are carried out on five households and statistically evaluated using five statistical indicators. The results indicate that the son model exhibits superior decomposition performance and greater resilience compared to the existing state-of-the-art models.