Privacy Enhanced Energy Prediction in Smart Building using Federated Learning
Sai Venketesh Dasari, Kaushal Mittal, Sasirekha GVK, Jyotsna Bapat, Debabrata Das
Abstract
Prediction of energy consumption is useful in energy budgeting of smart grids, expenditure budgeting by consumers, and comfort management of smart buildings. Essentially, building management systems of smart buildings need to manage energy, efficiently. In order to do this, energy prediction plays an important role. The energy consumption, in general, is predicted using machine learning algorithms. However, machine learning algorithms demand massive amounts of data for performing well. Acquisition of this data from data owners can lead to privacy breaches. Federated learning is a framework of distributed systems which can mitigate privacy breaches to certain extent. Federated learning has as such been developed for mobile edge devices like vehicles, phones etc. In this paper, a novel application of federated learning framework focused on the smart building energy prediction scenario is presented. The architectural details on how the federated learning framework is applied is presented. Also, the performance of this prediction model as compared to centralized method of machine learning is discussed. Deep neural networks are used with the American Society of Heating, Refrigerating and Air-Conditioning Engineers (ASHRAE) dataset to realize and evaluate this architecture.