Combined variable speed limit and lane change guidance for secondary crash prevention using distributed deep reinforcement learning

Abstract

Abstract The primary objective of this paper is to develop a combined variable speed limit (VSL) and lane change guidance (LCG) controller to prevent secondary crashes (SCs) and improve traffic efficiency on freeways. VSL controllers deliver speed limit instructions and LCG controllers deliver lane-changing instructions. A distributed deep reinforcement learning (RL)–based combined controller was proposed. The performance of the combined controller was evaluated in terms of safety and efficiency. Simulation experiments indicated that due to the complementation of VSL and LCG, the developed combined controller achieved higher performance in general than any single subcontroller. VSL control in a combined controller contributed prior effects on SC prevention and efficiency improvement, while LCG control improved the drawback of VSL by reducing the number of tough lane changes and avoiding extra SC risks caused by speed limit in relatively uncongested conditions. Moreover, the results of attention area investigation and sensitivity analysis revealed that the developed controller was able to accurately capture the spatial and temporal impact areas caused by prior crashes and generate proper interventions of traffic flow proactively.