Abstract: Current research on microgrid（MG） energy management based on federated deep reinforcement learning has not considered the problem of multiple types of energy conversion and power trading among MGs, while frequent interaction of model parameters leads to large communication latency. Based on this, an MC including multiple types of energies such as wind, solar, electricity, and gas is studied. An energy management model that allows for inter-MG electricity trading and intra-MG energy conversion is constructed. A federated dueling deep Q-network（Dueling DQN） learning algorithm based on the sine cosine algorithm（SCA） is proposed, and a multi-microgrid energy management and optimization strategy considering energy conversion and trading is designed based on the proposed algorithm. The simulation results show that the proposed energy management strategy can achieve higher rewards to maximize microgrid economic benefits while protecting data privacy and reducing communication latency.