CN-121982788-A - Intelligent toll station lane scheduling system based on deep reinforcement learning

CN121982788ACN 121982788 ACN121982788 ACN 121982788ACN-121982788-A

Abstract

The invention discloses a deep reinforcement learning-based intelligent toll station lane scheduling system, which is used for realizing remarkable improvement of lane utilization rate and optimization of traffic efficiency by collecting traffic flow data in real time and dynamically scheduling by using a deep reinforcement learning algorithm. And the lane scheduling strategy is adaptively adjusted according to the real-time data, so that the limitation of the traditional fixed mode or manual experience is avoided, and the scientific rationality of lane scheduling is ensured. The method and the device not only improve the traffic efficiency, but also reduce the congestion degree and improve the user experience, thereby realizing the comprehensive improvement of the operation efficiency of the toll station. Decision making is carried out based on real-time traffic flow data, interference of human factors is avoided, and reliability and accuracy of a scheduling result are ensured.

Inventors

ZHAO XIAORONG
LI SHIJIE
BAI YUQIONG
WANG QIGUANG
CHEN RONGXIN
XUE CHUNMING
WEI SEN
ZHAO YOUZHANG
LIU HUAYI
LI ZHIHUI
BO XIULI
DONG RUIJIE
YAN HAOYANG

Assignees

山西省智慧交通研究院有限公司
山西省智慧交通实验室有限公司

Dates

Publication Date: 20260505
Application Date: 20251229

Claims (10)

1. The intelligent toll station lane scheduling system based on deep reinforcement learning is characterized by comprising a traffic flow data acquisition module, a reinforcement learning decision module, a lane control module and a traffic information release module; The traffic flow data acquisition module comprises a sensor of a preset type; the reinforcement learning decision module comprises a state encoder and an action generator, and is used for outputting a dynamic lane scheduling strategy; The lane control module is used for dynamically switching the lane direction of the toll station according to the dynamic lane scheduling strategy of the reinforcement learning decision module; the traffic information issuing module is used for issuing traffic information, and the traffic information is used for guiding vehicles to select corresponding lanes according to dynamic switching of the lane directions of the toll stations.
2. The intelligent toll booth lane scheduling system based on deep reinforcement learning of claim 1, wherein the preset type of sensor comprises at least one of a geomagnetic sensor, a high definition license plate recognition camera, a radar integrated machine, an ETC/RFID reader, and an audio sensor.
3. The intelligent toll station lane scheduling system based on deep reinforcement learning according to claim 2, wherein the geomagnetic sensor is arranged at an entrance, a middle part and an exit of each lane and is used for collecting vehicle presence signals, vehicle passing time and time occupancy, calculating vehicle speed and vehicle body length according to double coil intervals, and deducing the queuing length of the lanes in real time through continuous triggering states.
4. The intelligent toll booth lane scheduling system based on deep reinforcement learning of claim 3 wherein the high definition license plate recognition camera is disposed on a toll booth ceiling or portal frame for capturing vehicle characteristic information including at least one of license plate number, pass time stamp, vehicle type, payment means, vehicle color.
5. The intelligent toll booth lane scheduling system based on deep reinforcement learning of claim 4, wherein the radar integrated machine is disposed at 500 meters in front of a toll booth and at a toll booth entrance area for monitoring full-face traffic flow conditions including at least one of traffic volume, average speed, lane occupancy, vehicle queue formation and dissipation process, abnormal parking, congestion.
6. The deep reinforcement learning based intelligent toll booth lane scheduling system of claim 5, wherein the ETC/RFID reader is disposed above each lane for obtaining transaction data comprising at least one of ETC transaction time, OBU information.
7. The intelligent toll booth lane scheduling system based on deep reinforcement learning of claim 6, wherein the audio sensor is disposed at a toll booth or toll booth ceiling for capturing abnormal acoustic events including at least one of sudden braking, collision.
8. The deep reinforcement learning based intelligent toll station lane scheduling system of claim 7, wherein the reinforcement learning decision module is configured to learn and optimize the dynamic lane scheduling strategy by training through a reinforcement learning algorithm based on the vehicle data.
9. The intelligent toll station lane scheduling system based on deep reinforcement learning of claim 8, wherein the lane control module is configured to control lanes of a toll station according to the dynamic lane scheduling policy to alleviate vehicle congestion or improve vehicle traffic efficiency.
10. The intelligent toll station lane scheduling system based on deep reinforcement learning of claim 9, wherein the traffic flow data acquisition module is configured to acquire vehicle data in real time, the vehicle data including at least one of lane direction and queuing length, exit vehicle arrival rate, entrance vehicle arrival rate; the traffic flow data acquisition module is used for transmitting the vehicle data to the reinforcement learning decision module in real time by adopting a wireless transmission technology; the mode of the traffic information issuing module issuing traffic information comprises at least one of information boards and broadcasting.

Description

Intelligent toll station lane scheduling system based on deep reinforcement learning Technical Field The invention relates to the technical field of intelligent traffic, in particular to an intelligent toll station lane scheduling system based on deep reinforcement learning. Background With the rapid increase of the traffic flow of the expressway, the problem of the traffic jam of the toll station lane is increasingly prominent. The traditional toll station lane scheduling mostly adopts a fixed mode (such as separation of ETC special lanes and manual lanes, fixed lane direction and the like), or relies on manual experience or preset rules, and is difficult to dynamically adjust according to actual traffic conditions, so that the utilization rate of lanes is low, the traffic efficiency is low, the user experience is poor, and traffic jam and potential safety hazards are caused. The intelligent and dynamic lane scheduling system is particularly important for improving the lane utilization rate, the traffic efficiency and the user experience of the toll station. Disclosure of Invention In order to solve the limitations and defects existing in the prior art, the invention provides an intelligent toll station lane scheduling system based on deep reinforcement learning, which comprises a traffic flow data acquisition module, a reinforcement learning decision module, a lane control module and a traffic information release module; The traffic flow data acquisition module comprises a sensor of a preset type; the reinforcement learning decision module comprises a state encoder and an action generator, and is used for outputting a dynamic lane scheduling strategy; The lane control module is used for dynamically switching the lane direction of the toll station according to the dynamic lane scheduling strategy of the reinforcement learning decision module; the traffic information issuing module is used for issuing traffic information, and the traffic information is used for guiding vehicles to select corresponding lanes according to dynamic switching of the lane directions of the toll stations. Optionally, the preset type of sensor includes at least one of a geomagnetic sensor, a high-definition license plate recognition camera, a thunder vision all-in-one machine, an ETC/RFID reader-writer and an audio sensor. Optionally, the geomagnetic sensor is arranged at an entrance, a middle part and an exit of each lane and is used for collecting vehicle presence signals, vehicle passing time and time occupancy, calculating vehicle speed and vehicle body length according to double-coil intervals and deducing the queuing length of the lane in real time through continuous triggering states. Optionally, the high definition license plate recognition camera is arranged on a toll booth ceiling or a portal frame and is used for capturing vehicle characteristic information, wherein the vehicle characteristic information comprises at least one of license plate numbers, passing time stamps, vehicle types, payment modes and vehicle colors. Optionally, the radar integrated machine is arranged in front of the toll station by 500 meters and in an entrance area of the toll station, and is used for monitoring the traffic flow condition of the full section, wherein the traffic flow condition of the full section comprises at least one of flow, average speed, lane occupation rate, vehicle queue formation and dissipation process, abnormal parking and congestion. Optionally, the ETC/RFID reader is disposed above each lane, and is configured to obtain transaction data, where the transaction data includes at least one of ETC transaction time and OBU information. Optionally, the audio sensor is disposed at a toll booth or a toll booth ceiling for capturing abnormal acoustic events including at least one of sudden braking, collision. Optionally, the reinforcement learning decision module is configured to train through a reinforcement learning algorithm according to the vehicle data, and learn and optimize the dynamic lane scheduling policy. Optionally, the lane control module is configured to control lanes of the toll station according to the dynamic lane scheduling policy, so as to alleviate vehicle congestion or improve vehicle passing efficiency. Optionally, the traffic flow data acquisition module is configured to acquire vehicle data in real time, where the vehicle data includes at least one of a lane direction and a queuing length, an arrival rate of an exit vehicle, and an arrival rate of an entrance vehicle; the traffic flow data acquisition module is used for transmitting the vehicle data to the reinforcement learning decision module in real time by adopting a wireless transmission technology; the mode of the traffic information issuing module issuing traffic information comprises at least one of information boards and broadcasting. The invention has the following beneficial effects: The invention provides a deep reinforcement learning-based in