An emergency dispatch strategy for faulted transportation electrification coupled networks based on safety-guided graph reinforcement learning
基于安全引导图强化学习的故障交通电气化耦合网络应急调度策略

摘要

针对交通电气化耦合网络在突发故障下的恢复难题，本文提出一种基于安全引导图强化学习的故障耦合网络应急调度策略。首先，构建故障耦合网络的双层恢复架构，上层实施配电网开关重构与交通潮汐车道调控，确立交通流量与电网功率边界。下层依托上层优化的通行条件，引导电动汽车驶向目标充电站，利用移动储能支撑电网负荷缺口。其次，通过图注意网络刻画电力-交通-充电站的状态传播与交通流演变过程，精准挖掘耦合网络的拓扑内嵌故障特征信息。然后，采用改进Dueling双重Q网络算法，求解出配电网开关重构序列、潮汐车道调控指令与充电站引导决策的最优调度策略。构建考虑拉格朗日安全约束的奖励机制，引导智能体主动规避不安全动作。并建立基于混合整数线性规划专家知识的引导学习机制，提升策略初始探索效率。最后，仿真验证表明，该策略提升了故障恢复效率，降低了配电网网损与交通拥堵程度，同时满足多维物理安全约束，有效增强了耦合网络的故障应急韧性。

Abstract

To address the recovery challenges of transportation electrification coupled networks under sudden faults, this paper proposes an emergency dispatch strategy for faulted coupled networks based on safety-guided graph reinforcement learning. Firstly, a double layer recovery framework for faulted coupled networks is constructed: the upper layer implements distribution network switch reconfiguration and traffic tidal lane regulation to define the boundaries of traffic flow and power grid power; the lower layer leverages the optimized traffic conditions from the upper layer to guide electric vehicles to target charging stations and utilizes mobile energy storage to offset the power grid load gaps. Secondly, the Graph Attention Network(GAT) is adopted to characterize the state propagation and traffic flow evolution processes of the power-traffic-charging station, accurately mining the topology-embedded fault feature information of the coupled network. Then, a modified Dueling Double Q-Network(MD3QN) algorithm is utilized to solve the optimal dispatch strategy including distribution network switch reconfiguration sequences, tidal lane regulation commands, and charging station guidance decisions. A reward mechanism considering Lagrange safety constraints is established to guide the agent to actively avoid unsafe actions, and a guided learning mechanism based on mixed-integer linear programming(MILP) expert knowledge is developed to improve the initial exploration efficiency of the strategy. Finally, simulation results show that the proposed strategy effectively improves the fault recovery efficiency, reduces the power loss of distribution network and traffic congestion, meets the multi-dimensional physical safety constraints, and significantly improves the fault emergency resilience of the coupled network.