Federated Deep Learning-Based Collaborative Optimization and Privacy Preservation in Microgrid Clusters Using GRU Models

Abstract

To address the challenges faced in optimizing the operation of distributed power generation systems, this study proposes an innovative method. The method integrates GRU deep learning to model temporal operational patterns and federated learning to enable privacy-preserving collaborative optimal operation of microgrid clusters. This model employs data augmentation techniques to build local encapsulated models and leverages federated learning to achieve strategy evolution. Case analysis demonstrates that this approach excels in prediction accuracy (test RMSE of 12.5), operational cost control (reducing total cost by 15.1% compared to independent operation), and computational speed, while also offering significant advantages in privacy protection. It provides an effective solution for the optimal operation of microgrid clusters.