Nonintrusive load disaggregation by fusion of graph signal processing and CNN

Le résumé fourni par la source

Abstract Nonintrusive load disaggregation techniques play a pivotal role in power system planning and decision making for demand response. The sparsity of low frequency total power data features, coupled with the model’s challenge in effectively extracting and utilizing spatial and temporal correlations in diverse load data, presents obstacles to achieving high precision disaggregation in nonintrusive load disaggregation. To address these challenges, this paper proposes a disaggregation framework that integrates graph signal processing (GSP) with a convolutional neural network (CNN). In this framework, the GSP module tackles the issue of sparse load features, while the CNN disaggregation module extracts and exploits various features, such as spatial and temporal correlations between different loads. The GSP module directly provides load features to the CNN disaggregation module, assigning weights to the outputs for selection and thereby enhancing the accuracy of the disaggregation process. Our method is trained and validated using real datasets, demonstrating satisfactory performance when compared to existing methods.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.