As natural carriers of resources like heating

ventilation

air conditioning systems

and electric vehicles

buildings can provide significant flexibility for the distribution network operation. For the coordinated operation of the distribution network and buildings

although multi-agent deep reinforcement learning can address the parameter dependence of model-based methods

it faces the privacy leakage risk during centralized training. Thus

this paper proposes a federated split reinforcement learning method that avoids the interaction of privacy data among multiple agents. This method is implemented in a hierarchical solution framework that integrates optimization and reinforcement learning for coordinated scheduling on the premise of operational security and data privacy. Firstly

a model for the coordinated operation of the distribution network and buildings is established. Then

this paper constructs a hierarchical solution framework combining reinforcement learning and optimization to learn the decision policies of buildings while guaranteeing the safe and economic operation of the distribution network. Next

a federated reinforcement learning algorithm incorporating split learning is proposed

where split learning is used in deploying and optimizing the global value function in blocks via gradients

enabling privacy-preserving and model-free coordinated operation of buildings. Finally

the effectiveness of the proposed method is validated using an IEEE 33-bus test system with 20 connected buildings.