An adaptive traffic signal control system and method based on deep reinforcement learning

Abstract

The invention belongs to the field of intelligent transportation, and proposes an adaptive traffic signal control system and method based on deep reinforcement learning. The present invention utilizes the interaction module to realize the real-time interaction between the intersection environment and the controller, that is, the state perception module collects the traffic state of the intersection in real time and provides an optimal decision-making scheme under the current traffic state through the control decision-making module; at the same time, the present invention can be updated by The module adopts the framework of reinforcement learning to continuously update the control core (Q value network) inside the controller to further improve the optimization effect of future control schemes. The present invention can comprehensively collect various influencing factors in two dimensions of time and space; use the recurrent neural network to improve the feature extraction ability and generalization ability for high-dimensional input matrix; can realize the complexity, Real-time, dynamic, random, adaptable and other requirements to improve the efficiency of traffic control at intersections and reduce travel delays.

Description

technical field [0001] The invention relates to the field of intelligent transportation, in particular to an adaptive traffic signal control system and method based on deep reinforcement learning. Background technique [0002] With the deepening of China's urbanization process, the urban population and vehicles continue to increase, so urban traffic management needs to propose an adaptive urban traffic signal control method that can meet the dynamic needs. The salient features of the urban traffic system are: the dynamic fluctuation of traffic demand, the instability of time and space, the diversity of influencing factors, and the complexity of control strategies. [0003] The adaptive control methods in the prior art mostly adopt control system methods such as fuzzy control, neural network, and genetic algorithm. The adaptive control scheme in the prior art has the following characteristics: due to calculation conditions and modeling reasons, most adaptive control The sche...

AI Technical Summary

Problems solved by technology

However, the adaptive signal control scheme represented by Q-learning usually uses artificial feature variables as the traffic state, which simplifies the complexity of expressing the traffic state and cannot fully consider various potential information of the traffic state; secondly, the core of Q-learning is the state -The mapping relationship of the behavior value table leads to Q-learning leading to a large state space, low learning efficiency, and low efficiency of control strategy

Images