Nonintrusive Load Monitoring via Causal Series-enhanced Multiscale Mixing Network

Nonintrusive Load Monitoring (NILM) aims to accurately decompose household electricity usage into appliance-level consumption patterns based solely on the aggregate power signal, enabling precise and efficient management of smart grid demands. However, previous methods face limitations in accuracy due to insufficient decoupling of multiscale features and poor distribution adaptability in power series, leading to suboptimal performance in complex scenarios. To address this challenge, we propose a novel model named Causal Series-enhanced Multiscale Mixing Network (CS-MMN). CS-MMN effectively captures the multiscale features and inherent correlations of power series through multiscale causal-series embedding, while decoupling temporal correlations and integrating historical power information across multiscale through power decomposable mixing. Experiments on U.K.-DALE demonstrate that CS-MMN achieves the state-of-the-art performance across all five test appliances in the U.K.-DALE dataset.