Enhancing Nonintrusive Load Monitoring With Targeted Adaptive Networks

Nonintrusive load monitoring (NILM) technology is a cost-effective method that infers the operating status and working conditions of each system appliance by disaggregating the overall electricity load recorded by the electric meter. This technology has numerous potential applications in the smart grid, including energy management, demand response, and fault detection. However, its effectiveness highly depends on the development of accurate disaggregation algorithms. Recently, deep learning (DL) has significantly enhanced the accuracy and computational efficiency of NILM. Despite these advancements, the generalization ability of most DL algorithms is limited due to the scarcity of training data. To address this limitation, we propose a new method named targeted adaptive network for NILM (TAN-NILM). This method features a main disaggregation network and an auxiliary network, trained using a multitask learning strategy. The auxiliary network extracts and provides feature information on individual appliances, assisting the main network in targeting specific appliances and estimating their loads. The experimental results demonstrate that TAN-NILM exhibits superior disaggregation accuracy and generalization ability compared to state-of-the-art NILM algorithms.