Traffic signal lamp control method based on time distribution and reinforcement learning

Abstract

The invention discloses a traffic signal lamp control method based on time distribution and reinforcement learning. The traffic signal lamp control method comprises the following steps: (1) configuring a simulated intersection environment and traffic flow data to a traffic simulator, and building an intelligent agent network; (2) an intelligent agent network generating the action of the next signal period according to the road condition state and transmitting the action to a traffic simulator to simulate one signal period; (3) storing the experience of the last signal period into a replay buffer; (4) sampling an experience from the replay buffer to train an intelligent agent network, judging whether the number of simulated steps reaches a preset value, and if not, returning to the step (2), otherwise, executing the next step; and (5) resetting the traffic simulator, testing the intelligent agent network, and performing traffic signal lamp control application after the test is finished.By adopting the traffic signal lamp control method of the invention, the traffic efficiency can be obviously improved, and the method can be more easily applied to actual roads.

Description

technical field [0001] The invention belongs to the field of traffic signal lamp control, in particular to a traffic signal lamp control method based on time allocation and reinforcement learning. Background technique [0002] For a long time, traffic congestion has not only plagued the daily travel of people all over the world, but also caused serious economic losses. Studies have pointed to inefficient traffic light control signals as one of the most prominent causes of frequent congestion. Therefore, how to optimize the traffic light control mechanism and improve the overall traffic efficiency has attracted extensive attention from academia and governments. [0003] Traditional traffic signal control strategies mostly rely on static schedules designed by traffic engineers, or dynamically adjust traffic signals based on real-time traffic information and artificially specified rules. However, due to the complexity and variability of real traffic scenarios, these methods s...

AI Technical Summary

Problems solved by technology

[0004] Most of the phase-switching models check the traffic flow distribution on the lane connecting the intersection every short period of time (usually about 5 seconds), and then the model generates an operation signal whether to switch to the next phase and Immediate execution, so most of these methods cannot directly implement the very important countdown function on the real road

Images