An Event-Detection Algorithm for Non-intrusive Load Monitoring of Residential Appliances

Nowadays, the issue of promoting supply-demand balance and energy conservation is urgent by increasingly exhausted traditional fossil energy sources and high penetration of renewable energy resources. Non-intrusive load monitoring (NILM) is one of the key techniques toward energy efficiency and conservation, which provides the guidance for operation scheduling of the power system and demand response strategies of the smart grid. A multi-feature non-intrusive load monitoring method based on event detection algorithm is proposed in this paper to profile residential consumer behavior, which focused on limited applicability and low recognition accuracy of existing NILM techniques. The event detection is conducted to locate the switching of appliances based on the phenomenon of a sliding window that tracks the statistical features of the acquired aggregated load data. To make sure that the model is high-precision, the event detection measurement indexes are applied to obtain the best event detector. On this basis, in order to achieve the comprehensive information of appliance power consumption, the dynamic adaptive particle swarm optimization (DAPSO) is utilized to identify the load type and working state based on the load feature database built by affinity propagation (AP) clustering which combines with the rich electrical characteristics of appliances. A case study using the open database of REDD dataset is applied, and the results validate the effectiveness of the method.