The invention belongs to the field of intelligent traffic, and provides a self-adaption traffic signal control system and method based on deep reinforcement learning. According to the self-adaption traffic signal control system and method based on deep reinforcement learning, real-time interaction of the intersection environment and a controller is achieved by using an interaction module, namely the traffic state of an intersection is collected in real time by a state sensing module, and an optimal decision scheme of the present traffic state is given through a control decision module; and meanwhile, a control core (Q value network ) in the controller can be continuously updated by adopting a framework of reinforcement learning through an update module, and thus the optimal effect of a future control scheme is improved. According to the self-adaption traffic signal control system and method based on deep reinforcement learning, various influencing factors can be synthetically collectedin both dimensions of time and space; a recurrent neural network is used for improving the extraction capability and the generalization capability of characteristics of a high-dimensional input matrix; and the requirements of complexity, instantaneity, dynamics, randomness, adaption and the like in self-adaption traffic signal control can be met, the traffic control efficiency in the intersectionis improved, and travel delaying is reduced.