Advancing Electric Load Forecasting: Leveraging Federated Learning for Distributed, Non-Stationary, and Discontinuous Time Series

In line with several European directives, residents are strongly encouraged to invest in renewable power plants and flexible consumption systems, enabling them to share energy within their Renewable Energy Community at lower procurement costs. This, along with the ability for residents to switch between such communities on a daily basis, leads to dynamic portfolios, resulting in non-stationary and discontinuous electrical load time series. Given poor predictability as well as insufficient examination of such characteristics, and the critical importance of electrical load forecasting in energy management systems, we propose a novel forecasting framework using Federated Learning to leverage information from multiple distributed communities, enabling the learning of domain-invariant features. To achieve this, we initially utilize synthetic electrical load time series at district level and aggregate them to profiles of Renewable Energy Communities with dynamic portfolios. Subsequently, we develop a forecasting model that accounts for the composition of residents of a Renewable Energy Community, adapt data pre-processing in accordance with the time series process, and detail a federated learning algorithm that incorporates weight averaging and data sharing. Following the training of various experimental setups, we evaluate their effectiveness by applying different tests for white noise in the forecast error signal. The findings suggest that our proposed framework is capable of effectively forecast non-stationary as well as discontinuous time series, extract domain-invariant features, and is applicable to new, unseen data through the integration of knowledge from multiple sources.