Nonintrusive load disaggregation method based on graph signal processing

To solve the problems of mining load data with low sampling rate and inaccurate load prediction, a non-invasive load decomposition prediction model based on graph signal processing is proposed to realize unsupervised learning and energy efficiency analysis and prediction by integrating graph signals. The model is based on Graph signal processing (GSP), which is applied to solve adaptive threshold, event clustering and feature matching respectively. From the perspective of signal processing, the noninvasive load decomposition problem is an event detection and data classification. The RDD data set, the sliding window with margins and the event detection algorithm are combined and superposed to obtain the final prediction results. Experiment results show that the proposed graph signal model can accurately monitor the overall variation trend of electric power load and real-time electricity consumption, track and decompose electric power more effectively, and have higher monitoring accuracy.