Abstract:

In the current cooperative operation of multi-agent integrated energy microgrids, the centralized optimization strategy has been experiencing the contradiction of agent privacy protection and parameter sharing, while in the distributed optimization, the optimization model needs to be simplified and approximated extensively such that the global optimal solution is not guaranteed. With regard to these challenges, this paper proposes a coordinated and optimized operation method for multi-agent integrated energy microgrids based on federated learning to achieve global optimum without compromising the agent privacy. Firstly, the equivalent interactive characteristic packaging model of each integrated energy microgrid is built based on the gated recurrent unit deep learning network and then uploaded to the cloud. Secondly, on condition of no invasion into the internal privacy data of each microgrid, the equivalent model of each individual microgrid is encrypted, and then consolidated in the cloud for federated learning. Thirdly, according to the results of cloud federation learning, the packaging model of interaction characteristics of each integrated energy microgrid at the edge is modified and updated iteratively until the loss function converges. In this way the global collaborative optimization operation of the integrated energy microgrids can be achieved under privacy protection. Finally, the feasibility and effectiveness of the proposed method are verified through case studies simulating a typical integrated energy microgrids. The results show that this method can realize the fast and efficient optimization operation of the integrated energy micro-group and effectively protect the data privacy of all participants.

Key words: integrated energy system, microgrid, federated learning, optimization operation, artificial intelligence