Repairing the damaged road network, which mainly focuses on how to reasonably schedule the repair crew to quickly unblock the road network and ensure that rescue teams and emergency resources in the source node can be delivered to different demand nodes in time, is a basic premise for emergency disposal and rescue after the occurrence of extraordinarily serious natural disasters. However, it is difficult for the existing methods to find a feasible scheduling strategy under enormous demand nodes. Therefor the key factors of repairing the damaged road network are first analyzed according to the road network model and the Markov decision-making process, based on which a double-feedback reward function is designed. Then, the deep Q-learning is utilized to solve the optimal scheduling strategy of the repair crew. Finally, comprehensive experimental studies show that for the damaged road network with enormous demand points, the proposed method has high stability and reliability, can achieve a good balance between the repair efficiency and the transportation efficiency, and can make all the demand points achievable with less repair cost, which may provide a useful attempt to repair the damaged road network in complex emergency scenarios of post-disaster.