EXFEDL: Explainable Clustered Federated Learning for Multivariate Residential Load Forecasting

Puntos clave

• The research aims to enhance the accuracy and interpretability of residential energy load forecasting using a novel federated learning framework.
• Developed EXFEDL, an explainable clustered federated learning framework.
• Grouped households based on temporal similarity for model training.
• Utilized a temporal convolutional network with global attention for learning cluster-specific models.
• Employed Shapley Additive explanations for interpretability at household and cluster levels.
• Implemented a stability controller to monitor behavioral drift and trigger adaptive re-clustering.
• EXFEDL showed improved forecasting accuracy over centralized and standard federated learning methods.
• The framework provided consistent explanations for the model outputs.
• Demonstrated adaptability to evolving load patterns, maintaining reliability over time.

Resumen