Highway road collaborative control system and method based on deep reinforcement learning

Abstract

The invention discloses a highway road collaborative control system and method based on deep reinforcement learning. The system includes a traffic information interaction module, a traffic control module, a deep learning network training module, and several variable speed limit and ramp control units. The module obtains the traffic status of the road, and then passes it to the traffic control module. The latter continuously optimizes the control strategy through the training module, and uses a deep reinforcement learning algorithm with actor-critic architecture to ensure the stability of the training process. The invention can simultaneously control all the traffic control units in the system without causing problems such as traffic state space explosion, can ensure that vehicles pass through the bottleneck section at a relatively high speed, and will not affect the passage of vehicles on surrounding roads due to problems such as queuing.

Description

technical field [0001] The invention relates to the technical field of traffic control and intelligent transportation, in particular to a system and method for cooperative control of expressway main roads and entrance ramps based on deep reinforcement learning. Background technique [0002] Expressways present frequent, periodic, and long-distance traffic congestion during peak hours, among which, the entrance ramps and adjacent main roads of expressways have become typical bottleneck areas of expressways. Since the early road network planning may be unreasonable, and road reconstruction is difficult, the coordinated management and control of expressway ramps and adjacent main roads is an important way to improve road traffic efficiency and improve driving safety. [0003] The existing cooperative control methods are mainly model predictive control or feedback control methods. In the model predictive control method, the characteristic variables are generally extracted from ...

AI Technical Summary

Problems solved by technology

However, deep reinforcement learning has the following problems when dealing with cooperative control: (1) Synchronous control problem when multi-agents cooperate

For example, the period of the ramp signal light and the period of the variable speed limit control are inconsistent, how to unify the two; (2) the existing reward function is easily affected by random disturbances in the traffic environment; On-ramp queuing issues, which can lead to congestion on surrounding roads

(4) The traditional deep reinforcement learning technology has inherent defects, and it is easy to cause problems such as behavior space state explosion when dealing with multi-agent cooperative control

Images