Optimization method for signal timing in single-point crossroad based on deep reinforcement learning

Abstract

The invention discloses an optimization method for signal timing in a single-point crossroad based on deep reinforcement learning, aiming at optimizing simulation timing in the single-point crossroadbased on deep reinforcement learning by actual road data, and finally applying a simulation scheme to an actual road. The method overcomes the shortcomings of the existing adaptive signal timing method and provides a novel algorithm that can perform high-dimensional input and fast convergence. The method comprises the following steps: firstly, performing matrix processing on the original data, randomly extracting input data through an empirical playback strategy, selecting an action of reinforcement learning by using an epsilon-greedy strategy, and finally obtaining an optimal solution for signal timing through continuous iterative training.

Description

technical field [0001] The invention discloses a deep reinforcement learning method for controlling and optimizing a single-point signalized intersection, which belongs to the technical field of intelligent transportation. Background technique [0002] With the development of the world economy and the rapid advancement of urbanization, the size and population of cities are also expanding and increasing. Although the expansion of cities has promoted economic development to a certain extent, the emergence of more and more private cars has caused urban traffic congestion. In order to strengthen the management of urban roads and reduce the occurrence of traffic accidents and congestion, it is necessary to set up traffic lights at intersections, thereby increasing road utilization and improving traffic conditions. The current traffic signal control can be roughly divided into three categories: (1) fixed signal timing. This timing strategy sets a fixed green signal ratio and sig...

AI Technical Summary

Problems solved by technology

However, there are still many deficiencies in its control of complex traffic systems, such as poor robustness, not stable enough, etc.

[0004] Reinforcement learning algorithms have been applied in the traffic field relatively early, and can control and optimize simple traffic signal timing. However, for traditional reinforcement learning algorithms, although they can solve simple control problems, due to the limitations of their own algorithms, It cannot solve the problem of high input dimension, and the calculation process is very complicated and cumbersome, so it is not suitable for today's traffic signal control

Images