A Non-Intrusive Load Monitoring Method Based on Relative Position Matrix and Residual Network

Non-Intrusive Load Monitoring (NILM) technology identifies various appliances’ types and operational status by analyzing electrical data collected from sensors in residential or commercial buildings, offering significant practical value. However, due to the limitations of existing methods for constructing appliance features, the identification performance of NILM models needs improvement. Addressing this, we propose a method for constructing appliance features based on a relative position matrix (RPM), which more accurately captures the temporal characteristics of electrical data compared to other methods. Additionally, a deep learning model based on the improved ResNet18 with multi-scale fusion is proposed for the load identification task. In this paper, comparative experiments were conducted on the PLAID dataset. We compared the RPM with three appliance feature construction methods: voltage-current trajectory (VI), markov transition field (MTF), and gramian angular field (GAF). We also compared our results with advanced methods reported in other literature. The results show that this method outperforms existing methods in terms of precision, recall, and F1 score across various evaluation metrics.