Emergencies often lead to multi-modal and unbalanced distribution of urban rail transit passenger flows. Aiming at this problem, a emergency control method of multi-modal passenger flow is proposed under the premise of comprehensively considering the operation cost and passenger travel efficiency of urban rail transit(URTS).The main work includes: 1) From the perspective of train scheduling, designing and developing a contingency strategy for multi-modal passenger flow by combining the train operation plan based on full-length and short-turn routing, station passenger flow restrictions, and dynamic train departure intervals; 2) From the perspective of train operation and passenger travel synergy, the emergency control model of passenger flow is established with the objective of minimizing the total train operation time and average passenger waiting time; 3) A multi-agent deep reinforcement learning(MDRL) method with a new action selection, reward and double loop mechanism-ARDQMIX is proposed to realize emergency autonomous perception and control of passenger flow in urban rail transit. The simulation results show that the emergency control method of multi-modal proposed in this study has a good utility for improving passenger travel efficiency and reducing operation cost. Note to Practitioners—Urban rail transit system(URTS) is punctual, fast, safe and large capacity, and has gradually become the preferred means of transportation for passengers. However, due to the rapid growth of travel demand, the URTS in the morning and evening peak hours traffic problems are particularly serious, will produce a multi-modal passenger travel demand, such as can not be quickly alleviated, the traffic problem will be caught in a vicious circle, leading to the urban rail transit system service capacity decline, and even lead to the potential safety problems. Aiming at this problem, a emergency control method of multi-modal passenger flow is proposed under the premise of comprehensively considering the operation cost and passenger travel efficiency of URTS.