Distributed Graph Publication and Dual-Granularity Privacy Protection Mechanism for Regenerative Braking Control

To address the risk of equipment privacy and operational information leakage caused by centralized publication of vehicle states, energy-storage SOC, and energy-interaction relationships in regenerative braking control, this paper proposes a lightweight distributed graph publication mechanism with dual-granularity privacy protection based on node-attribute LDP (NA-LDP) and Edge-LDP. First, vehicles, energy storage units, substations, and loads are abstracted as nodes in a dynamic directed graph, while energy interactions, braking associations, and communication links are modeled as directed edges. Second, adjacency pruning, index-weight encoding, and control-sensitive feature projection are used to reduce the publication dimension and communication overhead. Furthermore, dynamic privacy budget allocation is combined with SAC/DDPG to realize privacy-aware regenerative braking force allocation and energy-storage power scheduling. Experimental results show that when εtot = 2.0, the proposed method achieves a graph-feature MSE of 0.032, an MAE of 0.109, an energy recovery efficiency of 91.05%, an SOC out-of-bounds penalty of 0.019, and reduces the communication overhead to 24.7%. Under a 10% random edge-disconnection condition, it still maintains an energy recovery efficiency of 88.43%, demonstrating that the proposed method can balance graph data utility, control performance, and robustness.