Noise Attention-based Informer Model for Nonintrusive Load Monitoring

Nonintrusive load monitoring (NILM) has emerged as a promising and cost-effective technique for acquiring appliance-level power consumption data from aggregated data, enabling applications such as load forecasting, demand response, and grid dispatch. While deep learning (DL) presents a potential solution to enhance the accuracy and efficiency for NILM, existing research has overlooked the noise characteristics in aggregated smart meter data, limiting further performance improvements of NILM. This study presents a novel approach for distinguishing between load and noise patterns, considering both into a Noise Attention-based Informer Model (NAIM). By integrating the Informer architecture with a specialized noise attention mechanism, the proposed method enhances the accuracy of NILM. Experimental validation on the public REDD dataset demonstrates the superior performance of NAIM compared to existing state-of-the-art methods.