Aiming at the shortcomings of traditional clearing methods in multi-agent strategy interaction, channel blockage and disturbance risk response, this paper proposes a method for collaborative trading and supply protection efficiency optimization in cross-regional power mutual assistance market. Based on 6 regional power grid nodes, 15 cross-regional contact channels and 8640 groups of trading time samples, the research uses multi-source data collection, graph structure state representation, multi-agent game and reinforcement learning strategy network to realize mutual transaction clearing, and correcates the guarantee strategy through risk scoring, hierarchical early warning and online calibration mechanism. The experimental results show that the average clearing time of the proposed method is 1.84 s, and the transaction rate reaches 96.8%. In the compound disturbance scenario, the guaranteed supply satisfaction rate remains 90.8%, which is 10.3 percentage points higher than that of rule clearing. The multi-agent bidding strategy gradually converged after 20 rounds, and the price volatility decreased to 4.8%. The research shows that the proposed method can improve the intelligent clearing ability, risk response ability and supply protection resilience of the cross-regional power mutual market, and provide technical support for the collaborative allocation of regional power resources under the new power system.
