Energy Disaggregation with Federated and Transfer Learning

Authors:

Abstract
Energy disaggregation is the decomposition of a home or facility’s total energy consumption into individual equipment-level readings. Currently, non-intrusive load monitoring (NILM) is performed by constructing a regression model for different equipment using either data-driven or statistical methods, which requires the collection of equipment consumption data to train the model. However, this could potentially compromise privacy since users’ activity could be inferred by an adversary from the equipment working pattern. This paper proposes a new framework called Distributed Federated Non-Intrusive Load Monitoring (DFNILM) based on federated learning and transfer learning, which can achieve optimal user privacy while preserving good performance. Besides, the transferability of our framework between different dataset domains is explored using transfer learning. We implement our framework and conduct extensive experiments to evaluate the performance of two public NILM datasets. The results show that our framework can achieve comparable performance to state-of-the-art NILM algorithms while providing strong privacy protection, which is valuable for deploying the distributed algorithm to the IoT (Internet of Things) devices in a household.

This publication has 33 references indexed in Scilit: