Deep Sequence-to-point Learning for Electric Appliance Energy Disaggregation in Smart Building

In the modern era of smart grid and smart buildings, it is essential to have knowledge of the energy consumption of individual appliances in a building premise. Recently, appliance energy disaggregation via Non-intrusive load monitoring has emerged as an innovative technology for the above task. This paper proposes a deep sequence-to-point learning based neural architecture for disaggregating the energy consumption of appliances using the aggregate energy consumption data of the building measured at the electrical entrance alone. We have realized the proposed architecture using long short-term memory, gated recurrent unit, and their bidirectional variants. Using a publicly available dataset, disaggregation results obtained from these models are compared with other deep learning-based sequence-to-sequence and sequence-to-point models. The results confirm the superiority of the proposed architecture with at least 56% reduction in prediction errors for four major appliances.