A deep reinforcement learning traffic control model for Pedestrian and vehicle evacuation in the parking lot

Abstract

As a necessary component of connecting the building interior and urban road network for evacuation, the evacuation of parking lots significantly impacts overall evacuation efficiency. However, existing emergency evacuation studies have ignored the control of pedestrian-vehicle mixed flow in parking lot environments, leading to underestimated evacuation time estimates. Therefore, this paper proposes a pedestrian-vehicle mixed-flow model to simulate the parking lot evacuation process at the microscopic level. Moreover, a deep reinforcement learning (DRL)-based evacuation control model is developed to control pedestrian evacuation speed. The numerical study shows that this control model can effectively reduce evacuation clearance time by 7.75 % when faced with enormous evacuation demands.