Continuous intersection signal cooperative control method based on deep reinforcement learning

Abstract

The invention provides a continuous intersection signal cooperative control method based on deep reinforcement learning, and the method employs a DQN strategy of upper and lower Agent networks to process the signal timing of a continuous intersection, so as to reduce the complexity of state obtaining and feedback evaluation, and solve a problem of signal optimization of the continuous intersection. In order to ensure the stability of a training target and prevent the training of the training target from falling into oscillation divergence in a feedback cycle of a target value and a predicted value, a Duffing Double optimization method is adopted to perform DQN optimization training, and compared with a conventional DQN control model, the method can switch intersection phases in real time according to different road environments and traffic states, improves the cooperative ability between intersections, improves the stability of the system, smooth driving at the intersection is guaranteed, the traffic capacity of the intersection is improved, and a new solution and a theoretical basis are provided for relieving traffic jam, improving travel efficiency and reducing safety accidents.

Description

technical field [0001] The invention belongs to the technical field of vehicle-road coordination / arterial coordinated control, and specifically relates to a traffic signal control model based on vehicle-road coordination and deep reinforcement learning, which is applicable to any adjacent upstream and downstream intersections on the arterial coordinated control road section. Background technique [0002] For the urban road network, the intersection is the node to realize the traffic flow conversion of each road section, and it is also the main bottleneck restricting the traffic capacity of the road network. The traffic signal control system assigns the corresponding right of way to the conflicting traffic flows, realizes the separation of the conflicting traffic flows, and improves the traffic efficiency and safety of the intersection. Scholars at home and abroad have carried out a lot of research and application on traffic signal control, and have proposed a variety of sign...

AI Technical Summary

Problems solved by technology

[0012] 2. The existing green wave optimization algorithm often requires a large amount of calculation and is not flexible. The current research is mostly aimed at the single-phase coordination path, and the integrated optimization model for phase difference and vehicle speed. The traditional model is not enough Supports the case of consecutive intersections where signal coordination benefits are low

[0013] 3. The main difficulty in the coordinated design of continuous intersections is that the density of transverse intersections is high and the intervals between intersections are not uniform. The conventional green belt control method is difficult to realize. Although all kinds of sophisticated optimization algorithms can theoretically reduce the traffic capacity of the road network, the practicability and operability of the algorithm may not be guaranteed. If the neural network technology can be combined with the optimization scheme in the vehicle-road collaborative environment control, and let the signal phases of adjacent intersections be coordinated and controlled, then the signal control problem of continuous intersections can be solved

Images