Fig 1.
Reinforcement learning process based on overflow prediction.
Fig 2.
Relevant definitions in short-distance road section.
Fig 3.
The traffic flow distribution diagram when traffic lights in both entrance and exit of the short-distance road section are red.
Fig 4.
The traffic flow distribution diagram when the traffic light in entrance of the short-distance road section is red while the traffic light in exit of the short-distance road section is green.
Fig 5.
The traffic flow distribution diagram when the traffic light in entrance of the short-distance road section is green while the traffic light in exit of the short-distance road section is red.
Fig 6.
The traffic flow distribution diagram when traffic lights in both entrance and exit of the short-distance road section are green.
Fig 7.
Diagram of the reinforcement learning architecture based on DDPG proposed in this paper.
Fig 8.
Action space schematic based on the distribution of traffic flow in each direction at the intersection.
Fig 9.
Traffic simulation of the intersections.
Table 1.
The traffic flow data of experimental intersections.
Fig 10.
Comparison of the methods proposed in this paper in terms of average queue length metrics at intersections for different time periods.
Fig 11.
Comparison of the methods proposed in this paper in terms of reward value distribution at intersections for different time periods.
Fig 12.
Comparison of the methods proposed in this paper in terms of average stopping delay at intersections for different time periods.
Table 2.
Effectiveness improvement of the P_DDPG method proposed in this paper compared to the traditional CTC method.
Fig 13.
Comparative validation of overflow state feedback indicator in reward functions.
Fig 14.
Comparative validation of TIS indicator in reward functions.