A non-intrusive load event detection method based on time-domain analysis features

Accurately detecting the complete transient process of an event can improve the performance of non-intrusive load monitoring （NILM）. However， existing event detection methods struggle to maintain high performance across multiple scenarios. To address this issue， a non-intrusive load event detection method based on time-domain analysis features is proposed. First， adaptive thresholds are calculated for each power sampling point， which are then used to preliminarily locate event positions. Next， a sliding window search method is employed to repeatedly move through steady-state periods before and after the event， determining the time span of the event. The start and end times of the event are identified through change point detection （CPD）. Subsequently， a post-processing step is designed to recognize and remove fluctuation events based on waveform characteristics. Finally， comparative experiments with three other detection methods are conducted using both private and publicly available BLUED datasets. The experimental results validate the effectiveness of the proposed method.