Strategic Traffic Management and Operation based on Deep Recommendation Agent

Abstract

Little effort has been made to the strategic traffic operation and management at the individual control-object level. This paper proposes a real-time strategic traffic recommendation framework at the individual intersection-level. An intersection-wise agent, based on a customized deep-and-wide neural architecture, is defined for each signalized intersection. The agent follows an operation on-demand basis, and employs a traffic state indicator to assess the requirement for an operational mode switch. When an intersection triggers its operational mode switch (which essentially means its traffic mode changes), the framework recommends a customized operational scheme. The utility and efficiency of the recommendation framework are demonstrated using real-world traffic data. According to the evaluation result, the framework exhibits consistent performance in different traffic scenarios and has the potential to provide more reasonable operational schemes than a human-based system.