Detecting the novel appliance in non-intrusive load monitoring

Results from Non-Intrusive Load Monitoring serve for energy decomposition and load identification, which would facilitate effective energy consumption management. Existing studies have focused on settings with a fixed number of electrical appliances. This differs significantly from real-world scenarios, thus largely limiting the practical application of related research. We study the pattern variations of the aggregated power sequences and separately analyze two different scenarios for new appliance introduction. Based on this, we propose a novel appliance detection method that can be implemented for low-frequency sampling data. The proposed method can determine whether new appliances are introduced without the prior information at the appliance level, thus establishing a basis for load monitoring in scenarios with dynamic changes in load topology. The experimental results show that the proposed method achieves at least a 35.93 % improvement in the metric of F1 compared to the event-based approach in the presence of a different number of unknown appliances. •We analyze low-frequency power data and present two case studies for new appliances.•Novel proximity measures are introduced to assess the similarity between appliances.•A new algorithm is proposed to detect the previously unknown appliances.