Accurate Action Recommendations and Demand Response for Smart Homes via Knowledge Graphs

Accurate action recommendations can enhance the convenience of daily life, such as automatically turning on the dining area lights during meals or playing music based on residential habits. Generating precise recommendations for the next household device actions is essential for future smart homes. This paper proposes an action recommendation system for household appliance scenarios by customizing the knowledge graph attention network in its sampling and aggregation, in which the usage habits, periods, and location information were used as common sense for graph modelling. The results of the recommendations can be explained by a designed method with the trained embeddings. Finally, with the recommendation expectation, appliances' comfort level and average power are modelled as a multi-objective optimization problem for participating in demand response. Simulations demonstrate that the proposed system can achieve 93.4% accuracy in recommendations and reduce the power consumption by 20% while providing reasonable explanations.