• Energy Research

Within smart homes, consumers could generate a vast amount of data that, if analyzed effectively, can improve the convenience of consumers and reduce energy consumption. In this paper, we propose to organize household appliance data into a knowledge graph by using the consumers’ usage habits, the periods of usage, and the location information for graph modeling. A framework, ‘DARK’ (Device Action Recommendation with Knowledge graphs), is proposed that includes three parts for enabling demand response. Firstly, a household device action recommendation algorithm is proposed that improves the knowledge graph attention algorithm to make accurate household appliance recommendations. Secondly, graph interpretable characteristics are developed in the DARK using trained graph embeddings. Finally, with the recommendation expectation, the consumers’ comfort level and appliances’ average power load are modeled as a multi-objective optimization problem in the DARK to participate in demand response. The results demonstrate that the proposed system can generate appliances’ action recommendations with an average of 93.4% accuracy and reduce power load by up to 20% while providing reasonable interpretations for the device action recommendation results on the customized UK-DALE dataset.