Coordinated optimization of timetabling, skip-stopping, and passenger flow control with passenger loss

Peak-hour congestion constrains operational efficiency and travel experience, motivating coordinated timetabling and flow control to balance train occupancy and passenger demand. However, stricter flow control can alleviate platform crowding but prolong outside-station waiting, potentially causing passenger loss that remains underexplored. This study jointly optimizes timetabling, skip-stopping, and flow control to minimize waiting time and probabilistically modeled passenger loss. The problem is formulated as an event-driven Markov decision process with hybrid discrete-continuous actions. An adaptive parameterized deep Q-network (APDQN) is developed with a risk-aware action selection, where infeasible actions are excluded and operational risk guides exploration and exploitation. Chengdu metro case studies show coordinated control better balances waiting time and passenger loss than flow control alone or combined train-side control. GA remains competitive in small and medium cases, whereas APDQN achieves better solution quality in large case. Comparisons with DQN, ADQN, and PDQN demonstrate benefits of hybrid actions and risk-aware selection.