An Adaptive Decentralised Federated Learning Framework for Load Forecasting in Dynamic Microgrid Environments
Abstract
Abstract In modern microgrids, reliable load forecasting is vital for Energy Management Systems (EMS) that navigate decentralised demand and intermittent renewable generation. Traditional deep learning models frequently struggle to capture long-term temporal dependencies across heterogeneous household profiles, while centralising raw consumption data creates privacy vulnerabilities. Although Centralised Federated Learning (CFL) protects user privacy, it introduces single-point node failures, communication bottlenecks, and memory overload on resource-constrained edge EMS hardware. To overcome these issues, we propose a Decentralised Federated Learning (DFL) enabled framework integrated with a hybrid TCN-BiLSTM architecture. In this design, Temporal Convolutional Networks (TCN) capture instantaneous time-slice fluctuations, Bidirectional LSTMs (BiLSTM) retain macro temporal dependencies, and peer-to-peer DFL eliminates central server dependencies to suit edge memory constraints. Validated on the Ausgrid and NSW-Sydney datasets, our DFL-TCN-BiLSTM framework achieves an R^2 of 0.97 and a MAPE of 0.74% (in NSW-Sydney), outperforming state-of-the-art baselines to deliver a secure, scalable, and highly reliable forecasting engine for dynamic microgrid environments.
Read the paper
Related studies
- Active Learning during Federated Learning for Object Detection
- Fairness-aware federated learning
- Decentralized Federated Learning
- Building a Personally Identifiable Information Recognizer in a Privacy Preserved Manner Using Automated Annotation and Federated Learning
- Preservation of Patient Level Privacy: Federated Classification and Calibration Models
- Efficient explainable federated learning
Go deeper in Evidence Hub
Evidence Hub adds a study data sheet (design, sample size, evidence level, key numbers), an AI summary and a chat grounded in the paper's text.