Accurate and Data-Limited Prediction for Smart Home Energy Management

Residential energy applications have become an important domain of cyber-physical systems. These applications provide significant opportunities for end-users to reduce their electricity costs and for the utilities to balance their supply and demand in the most effective way. One of the most important applications is predicting the total energy usage of a house. However, an accurate time-series prediction may require significant amount of data, e.g. per appliance energy consumption values, that need costly installations, data storage units, and computation and communication devices. In this paper, we propose a framework that uses a forward-selection-based input filtering mechanism for residential prediction applications. Our framework can effectively reduce the amount of data required for residential energy prediction without sacrificing prediction performance. We demonstrate that 94% of the houses can leverage our method, which leads to up to 80% reduction in required data, greatly reducing the system cost and overhead.