Deep Learning-Based Non-Intrusive Load Disaggregation Considering Continuous Power Appliances

Non-intrusive load monitoring (NILM) can achieve device-level event detection and energy disaggregation by utilizing meter data at power entrance, paving the way for a smart energy consumption future. However, the presence of continuously variable device (CVD) poses challenges for accurate event detection and disaggregation of other devices. Therefore, this paper proposes a NILM model and method focusing on continuous power feature. It analyzes the power change intensity within sliding windows to accurately determine the switching moments of finite-state device (FSD). Subsequently, a deep learning-based load disaggregation method predicts FSD power consumption. By stepwise striping FSD and noise, the CVD power consumption is obtained. Scenario-based simulation results validate the effectiveness of the method to achieve precise event detection and all-type load energy consumption disaggregation.