Active integration algorithm of vehicle road cooperative signal based on deep reinforcement learning

Abstract

In the process of integrated processing of vehicle road cooperative signals, due to the low accuracy of the analysis results of road traffic status, the specific effect of using it to manage road traffic is not ideal. For this reason, this paper proposes the research of vehicle road cooperative signal active integration algorithm based on deep reinforcement learning. Taking full account of the shortcomings of the reinforcement learning algorithm in terms of convergence, we introduced deep reinforcement learning, added a new target network to the original reinforcement learning network, combined with the road traffic state transition matrix, and calculated the traffic state of the road environment by updating and iterating. In the active integration stage of vehicle road cooperative signals, based on the boosting algorithm, the differences shown in the analysis results of road traffic status are used to construct corresponding differences between integrators, and the integrated processing of them is realized on the basis of differentiated road traffic status signal weights. In the test results, under the design algorithm, the average travel time of road traffic is only 185s, the average travel speed reaches 5.97m/s, and the average queue length is only 6.52 vehicles, which has good application effect.