An adaptive traffic signal control method based on multi-agent reinforcement learning

Abstract

The invention discloses an adaptive traffic signal control method based on multi-agent reinforcement learning, comprising: for each intersection in a traffic road network, establishing an intelligent agent network corresponding to the intersection; acquiring the traffic road network The real-time traffic features in the real-time traffic features; the real-time traffic features are respectively transmitted to the agent network corresponding to the intersection according to the position of the intersection, and the phase of the intersection output by the agent network is obtained, and according to the intersection The phase of the intersection executes the traffic signal control of the intersection; wherein, the agent network is based on the average value of the sum of the number of vehicles waiting on the entry lanes of the intersection corresponding to the agent network and the intersection adjacent to the intersection Determine the phase of this intersection for the reward. The invention improves the coordination ability between independent intelligent bodies, and provides a solution for the traffic signal control in the complex road network environment.

Description

Technical field [0001] The present invention relates to the field of traffic control, and more particularly to an adaptive traffic signal control method based on multi-intelligent body strength. Background technique [0002] Implementing intelligent traffic control is a low cost method for reducing traffic congestion and improving transportation efficiency. Since traffic flow has the characteristics of time variations and randomness, the traffic modeling is still very difficult to model traffic modeling in a complex road network environment in multi-crossroads. [0003] A method of modeling a MarkovDecision Process (MDP) based on multi-intelligent body strength learning, MDP, will be a single intelligent body-critic The algorithm extends into a multi-smart transport environment. [0004] The above method is the scalable independent algorithm, only the possibility of adding the cooperation between the intelligent body from the traffic model, and does not do associated targeted des...

AI Technical Summary

Problems solved by technology

Considering that the multi-agent cooperation problem will encounter difficulties such as instability and incoordination in the independent algorithm, it is difficult to learn in a complex road network environment, such as an environment with high coordination requirements between intersections caused by a large number of vehicles in the road network. good collaboration strategy

Images