A Stacking Ensemble Learning Method for Non-Intrusive Appliance Load Monitoring

Non-Intrusive Load Monitoring (NILM) collects the overall power consumption information from the power panel with advanced metering infrastructure, and obtains the operating states and power consumption information of each electrical appliance through load decomposition algorithms, so as to come up with appropriate strategies to reduce energy consumption and improve energy efficiency, which is of great significance for energy conservation and reduction. In recent years, NILM methods based on deep learning algorithms have received much attention and achieved good results, but the type of features extracted by a single deep learning model is limited, resulting in the model’s inability to overcome the bottleneck of accurate decomposition. To address the above problems, this paper proposes a deep ensemble learning framework for NILM, which takes the total active power sequence as input and synthesizes the advantages of multiple deep learning neural network models to further improve the performance of appliance identification and power disaggregation. Experiments are designed and conducted on the UK-DALE dataset to verify the effectiveness of the proposed deep ensemble learning method for NILM. The experimental results show that the method proposed in this paper has more advantages and is more promising in terms of various metrics compared to single deep learning approaches.