Traffic signal control method and system based on reinforcement learning and graph attention network

Abstract

The invention provides a traffic signal control method and system based on reinforcement learning and a graph attention network, and the method comprises an initialization step: defining each variablein a traffic signal control problem, and initializing a traffic signal algorithm model; an observation information vectorization step: performing dimensionality reduction on the observation information vector through a multilayer perceptron to obtain dimensionality-reduced data hi; a graph attention mechanism construction step: constructing an attention mechanism hmi suitable for the traffic signal algorithm model by using the data hi after dimension reduction and starting from the attention mechanism; a loss function construction step: calculating a loss function according to the attention mechanism hmi; a behavior updating step: performing iterative computation on the loss function according to a reward function in the constructed traffic signal algorithm model to obtain a final trafficsignal algorithm model; and a prediction result calculation step: calculating the control strategy pi of the traffic signal according to the constructed final traffic signal algorithm model to realize signal control. The traffic signal control method and system are suitable for large-scale complex traffic road conditions.

Description

technical field [0001] The invention relates to the fields of computer software and traffic, in particular to a traffic signal control method and system based on reinforcement learning and graph attention network. Background technique [0002] One of the most common questions in people's lives is "how to coordinate the control of signal lights at different intersections?" The coordinated control of signal lights at intersections is crucial to the efficiency of urban traffic networks, because signal lights at different intersections affect each other. Especially when the intersections are not far apart. The better the coordination effect between signal lights at intersections, the more conducive to the improvement of the traffic efficiency of the entire traffic network. [0003] In the field of transportation, a more classic approach is to solve the problem of adjusting signal lights at intersections under certain assumptions. However, this approach often does not perform w...

AI Technical Summary

Problems solved by technology

[0005] However, the current reinforcement learning methods applied in the field of traffic signal control are still unable to achieve efficient sharing and collaborative control of signals between intersections.

Images