Towards Scalable Dynamic Traffic Assignment With Streaming Agents: A Decentralized Control Approach Using Genetic Programming

Abstract

Traffic assignment is of great importance in real life from foot traffic assignment of a building to vehicle traffic assignment of a city. With the rapid increase of the number of agents and the size of the traffic network, the problem becomes more and more challenging nowadays. To solve large-scale efficient dynamic traffic assignment, this article proposes a decentralized control approach to achieving this. First, we transform the traffic assignment problem into a routing rule generation problem by designing a corresponding simulation-based optimization framework that can evaluate the performance of routing rules. Then, we propose a genetic programming hyper-heuristic algorithm to generate the optimal routing rule. During execution, agents can collect environmental information and plan their paths when moving according to the generated rule. In this way, the proposed method can handle the dynamically changing traffic flow efficiently instead of pre-planing the whole path for each agent. The proposed method is verified on both synthetic networks and real-world networks in terms of sensitivity, generality, and scalability. The experimental results demonstrate that our method is effective in urban-scale traffic networks and outperforms the compared algorithms.