Game-theoretic conflict resolution for tunnel emergencies in mixed connected traffic

The connected and automated vehicles (CAVs) enable beyond-line-of-sight accident detection, supporting optimized lane-changing for upstream vehicles during tunnel emergencies. To mitigate congestion and secondary accidents in tunnels, this study proposes a game-theoretic lane-changing model for mixed traffic that accounts for driver behavior diversity and mandatory lane-changing requirements. Using traffic shock wave theory, the model defines accident impact zones based on warning propagation distance. A cost function combining traffic volume, average speed, and ideal lane-changing time determines target lanes, with critical safe distance as the feasibility criterion. The framework establishes cooperative games among CAVs and incomplete information games between CAVs and connected manual vehicles (CMVs), with payoff functions incorporating success rewards, cooperation incentives, safety benefits, and efficiency gains, solved via Perfect Bayesian Equilibrium and cooperative equilibrium. Validation through a SUMO-Python co-simulation shows that the proposed game-theoretic lane-changing model improves mean speed while reducing mean time loss and mean max jam length across all tested traffic volumes and CAV penetration rates. At 5000 veh/h and 100% CAV penetration, the mean speed of accident-affected lane reaches 19.33 km/h, representing increases of 123.0%, 71.9%, and 3.1% compared with SL2015, LC2013, and SPF, respectively. Inter-lane speed variation remains under 0.05, ensuring flow stability and minimal disruption. Mean time loss drops to 1.46 s, which is 97.8%, 88.0%, and 37.6% lower than the benchmarks, respectively. The model also reduces the mean max jam length by 40% to 80% under CAV penetration rates ranging from 40% to 100%, demonstrating effective utilization of multi-dimensional data collected by CAVs. This model can be directly integrated into tunnel emergency management systems and intelligent transportation infrastructure, providing real-time decision support for dynamic lane-changing coordination in mixed traffic environments.