Ambulance is one of the most crucial medical resources to save patients’lives. Appropriate allocations of limited ambulances to different emergency stations can effectively lower the response time and lift medical service quality. In view of this，we propose a reinforcement learning based scheduling structure to resolve the dynamic ambulance redeployment problems. In order to address the challenges aroused from high-dimensional state spaces，we propose RedCon-DQN by considering multiple scheduling interactive factors，which is based on Deep Q-value Network ( DQN) and can output the optimized redeployment policy given specific environment. In addition，we propose a measurement，emergency-network resilience to evaluate the influences of each individual emergency station on the global optimization objectives. Finally，we construct a environment interactive simulator based on the emergency calls and response data of Nanjing from 2016 to 2017. We validate the advantages of the proposed redeployment policy over the state-of-the-art methods，and further analyze the effectiveness and characteristics in different time periods.