Residential Short Term Load Forecasting Based on Federated Learning

Load forecasting is an essential task in the power industry as an important means to assist the grid to balance supply demand. A large amount of user data monitored by smart grids can support deep learning models for load prediction, but accurate and fine-grained user data may reveal consumers' electricity consumption behaviors, which brings privacy and security issues. Federated Learning (FL) is a new type of high-efficiency machine learning between multiple participants or multiple computing nodes under the premise of ensuring information security during big data exchange and protecting the privacy of terminal data and personal data. Therefore, this paper explored a short-term residential energy demand forecasting method based on FL. The experimental data comes from the U.S. hourly residential base load. The federal forecast model was built on Pytorch, and we explored model behavior under different experimental conditions.

I. Introduction

Load forecasting, especially short term load forecasting, has gradually become an important part of the development of smart grids. Improving prediction accuracy of short term loads will assist dispatchers to improve safe and stable operation of the power system. The energy consumption of residential buildings accounts for a large part of the energy use [1]. Their loads are highly correlated with residents' behaviors, which are always fluctuating randomly [2], making prediction tasks much more challenging.