Multitask Learning-Based Approach for Integrated Load Identification, Electrical Fault Detection, and Signal Purification

In recent years, the application of deep learning techniques for load identification in nonintrusive load monitoring (NILM) has gained significant momentum. However, the presence of load faults and noise interference caused by external environment during practical load identification poses a challenge. Electrical fault detection and signal denoising are crucial aspects in the research of load identification. To address these issues, a multitask learning network (MTLNET) was proposed to address three tasks in parallel, including load identification, fault detection, and signal denoising. To extract effective information from load signals, a novel feature representation called robust variational mode extraction-instantaneous amplitude (RVME-IA) was introduced. Furthermore, a multilevel feature fusion attention (MLFFA) module is employed to capture both local and global features for different tasks. For the classification subnets, a local enhance attention module (LEAM) is designed to locally amplify crucial information in the feature. To ensure optimal performance across the different tasks, a customized gate control (CGC) module is integrated into the proposed model. The proposed approach was evaluated on a self-collected high-frequency load fault dataset and public NILM datasets. The experimental results indicate that the proposed model outperforms state-of-the-art methods in all three tasks.