Accelerating Realistic Multi-Route Traffic Flow Generation Based on Improved Simulated Annealing Algorithms

Hongrui Chen, Yan Zhang, Xingyuan Dai, Jiaming Sun, Fei Cong, Yanan Lu, Bingji Xu, Yisheng Lv

Abstract: In the traffic simulation system, the data quality and inference speed of traffic flow generation have a direct impact on the correlation between analysis results and actual traffic conditions, as well as the efficiency of utilization. Existing traffic flow generation methods often rely on abundant sensor data and involve time-consuming generation processes. To reduce time costs, in this paper, we propose an efficient Multi-Route Traffic Flow Generator (MTFG). The generation process is divided into two parts: multi-route generation and route quantity search. In the multi-route generation, we employ a Multi-Route Generator (MRG) methodology leveraging topological road network data to generate distinct traffic routes for identical origin-destination pairs. This ensures the attainment of diversified vehicle trajectories even within a sparsely equipped detector network. Subsequently, in the route quantity search, we employ an improved simulated annealing algorithm to systematically explore the traffic route quantity based on the generated multiple routes, thereby significantly amplifying the algorithm's efficacy in route generation. We validate the effectiveness of MTFG using the Malaga open dataset on the SUMO simulation platform. Compared to the strong baseline Flow Generator Algorithm (FGA) with a training time exceeding 2.5h, MTFG achieves the target metric in just 1h under the same conditions, saving over 55% of training time.