A Multi-Task Learning Framework for Load Identification and Signal De-Noising

Non-intrusive Load Monitoring (NILM) represents a cost-efficient technological advancement in meticulously monitoring electricity consumption patterns. However, external noise interference in practical load identification poses challenges for accurate monitoring. To address this, a multi-task learning framework is proposed, concurrently tackling load identification and signal de-noising. To extract effective information from load signals, a novel feature representation called variational mode decomposition-instantaneous amplitude (VMD-IA) was introduced. To ensure optimal performance across the different tasks, a multi-gate mixture of expert architecture (MMOE) module is integrated into the proposed model. The proposed approach was evaluated on a self-collected load dataset, as well as public NILM datasets. The experimental results indicate that the proposed model outperforms state-of-the-art methods in all tasks.