Forecasting Energy Availability in Local Energy Communities via LSTM Federated Learning

Local Energy Communities are emerging as crucial players in the landscape of sustainable development. A significant challenge for these communities is achieving self-sufficiency through effective management of the balance between energy production and consumption. To meet this challenge, it is essential to develop and implement forecasting models that deliver accurate predictions, which can then be utilized by optimization and planning algorithms. However, the application of forecasting solutions is often hindered by privacy constrains and regulations as the users participating in the Local Energy Community can be (rightfully) reluctant sharing their consumption patterns with others. In this context, the use of Federated Learning (FL) can be a viable solution as it allows to create a forecasting model without the need to share privacy sensitive information among the users. In this study, we demonstrate how FL and long short-term memory (LSTM) networks can be employed to achieve this objective, highlighting the trade-off between data sharing and forecasting accuracy.

💡 Research Summary

This paper addresses the critical need for accurate energy availability forecasting within Local Energy Communities (LECs) while respecting the privacy constraints imposed by regulations such as GDPR and CCPA. Traditional centralized machine‑learning approaches achieve high accuracy but require the collection of raw consumption and production data from individual households, raising serious privacy concerns. To overcome this dilemma, the authors propose a federated learning (FL) framework that couples a Long Short‑Term Memory (LSTM) neural network with a horizontal FL architecture, enabling collaborative model training without ever transmitting raw user data.

The methodology consists of several key components. First, each participant (prosumer or consumer) maintains a local LSTM model that ingests hourly time‑series features—energy production, consumption, and ambient temperature—and predicts the net energy surplus (production minus consumption) for the next 24 hours. Second, a central server orchestrates training rounds: it distributes the current global model parameters, receives updated local parameters after a few epochs of on‑device training, and aggregates them using a customized version of Federated Averaging (FedAvg) enhanced with a FedProx‑style regularization term to mitigate the effects of data heterogeneity across clients. This “Model‑Centric Cross‑Silo Horizontal” setup preserves the temporal dependencies inherent in the 24‑hour horizon while keeping communication overhead low, as only model weights (on the order of megabytes) are exchanged.

For empirical evaluation, the authors construct a synthetic dataset derived from real Danish household measurements (reference