CSAPNet: A Cross-Scale Attention Pyramid Network for Multi-Appliance Non-Intrusive Load Monitoring

As a promising technology, Non-Intrusive Load Monitoring (NILM) aims to disaggregate the power consumption of individual appliances from the total power usage. Most existing methods focus on training separate models for each appliance, making practical application challenging. In this paper, we propose an innovative Cross-Scale Attention Pyramid Network (CSAPNet) for multi-appliance disaggregation, which integrates the pyramid structure with cross-scale attention mechanism. CSAPNet significantly improves the accuracy of multi-appliance power disaggregation by extracting features at different temporal scales and effectively integrating global and local information. In the cross-scale attention mechanism, fine features guide coarse features, thereby enhancing the interaction between features. Compared to baseline models, our model demonstrates superior performance on two public datasets. Additionally, we validate the effectiveness of the cross-scale attention mechanism through ablation experiments.