Traffic pressure dispersion method comprehensively applying intelligent technology and dynamic strategy
By constructing a multi-dimensional traffic perception network and multi-agent reinforcement learning decision-making, the spatiotemporal optimization of traffic lights and lane resources is achieved, solving the problems of data distortion, insufficient prediction, and spatiotemporal fragmentation in existing traffic control systems, and improving traffic management effectiveness and system response capabilities.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- DEZHOU UNIV
- Filing Date
- 2026-02-24
- Publication Date
- 2026-05-15
AI Technical Summary
Existing traffic control systems suffer from data distortion at the perception level, lack of prediction and coordination at the decision-making level, and spatiotemporal fragmentation at the execution level, resulting in poor traffic management effectiveness, especially under complex traffic flow and inclement weather conditions.
A multi-dimensional heterogeneous traffic perception network is constructed, and multi-source data trust fusion and multi-agent reinforcement learning decision-making are adopted to achieve spatiotemporal synchronous optimization of traffic lights and lane resources. Combined with traffic pressure index based on psychological factors, dynamic traffic management strategies are generated, and control commands are executed through the NTCIP protocol.
It improved the accuracy of traffic parameters and the real-time response capability of the system, reduced delays and accident risks in severe weather, enhanced intersection capacity, and reduced deployment costs.
Smart Images
Figure CN122050136A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent transportation systems and urban traffic congestion management technology, and in particular to a traffic pressure relief method that comprehensively utilizes intelligent technology and dynamic strategies. Background Technology
[0002] With the acceleration of urbanization, the number of motor vehicles has continued to surge, leading to an increasingly acute contradiction between urban road traffic supply and demand. Traditional traffic control systems have mainly gone through three stages of development: timing control, sensor control, and adaptive control.
[0003] Existing technologies have the following limitations. First, at the perception level, traditional sensor control can only acquire information on the presence or absence of vehicles at a cross-section or simple traffic flow statistics. Although adaptive systems such as SCATS and SCOOT have achieved dynamic timing to some extent, they mainly rely on relatively sparse detector data. When facing complex mixed traffic flows or inclement weather, a single sensor is highly susceptible to the effects of ambient light and occupancy, leading to data distortion and thus misleading control strategies. In addition, existing congestion assessment indicators are mostly based on simple physical parameters (such as average speed and occupancy rate), ignoring the significant impact of driver psychological factors (such as anxiety about long red lights and cautious driving behavior in inclement weather) on the micro-behavior of traffic flow.
[0004] At the decision-making level, current traffic signal control methods mostly employ rule-based or shallow model optimization approaches. With the rise of deep reinforcement learning, many studies have attempted to optimize signal timing using DQN or PPO algorithms. However, most existing solutions suffer from problems such as a simplistic state space design, a lack of prediction of future traffic flow and coordination with the surrounding road network, and reward functions that typically only pursue maximizing traffic volume, leading to extreme delays for vehicles in certain directions, without considering the safety hazards caused by frequent signal switching.
[0005] At the implementation level, existing traffic management strategies suffer from a severe disconnect between time and space. Signal control (time allocation) and lane management (spatial allocation) are often two independent systems. For example, the switching of tidal flow lanes is usually based on a fixed schedule or manual dispatch, and cannot be dynamically responded to based on real-time minute-level traffic fluctuations. This static spatial configuration cannot adapt to the highly random and sudden nature of modern urban traffic flow.
[0006] Therefore, how to construct a comprehensive traffic management method that can accurately perceive traffic conditions around the clock and use artificial intelligence algorithms to achieve synchronous dynamic optimization of traffic light lanes in time and space is a difficult problem that urgently needs to be overcome in the field of intelligent transportation. Summary of the Invention
[0007] To overcome the shortcomings of existing technologies, this invention provides a traffic congestion mitigation method that comprehensively utilizes intelligent technology and dynamic strategies, characterized by comprising the following steps: S1. Construct a multi-dimensional heterogeneous traffic perception network and perform data cleaning; collect holographic traffic flow parameters by deploying millimeter-wave radar, high-point surveillance video, geomagnetic sensors, floating car data and meteorological environmental data through API interfaces on the roadside; use an anomaly detection algorithm based on isolated forest to remove sensor noise, and use Kalman filtering to perform spatiotemporal interpolation to complete the missing data. S2, based on the improved Dempster-Shafer evidence theory, establishes a multi-source data trust fusion; establishes an identification framework including smooth traffic, slow traffic, light congestion, severe congestion and deadlock; for the conflict of different sensors' state judgments of the same traffic section, introduces a time decay factor based on historical accuracy and sensor dynamic trust weights, calculates the corrected basic probability allocation, and outputs a high-confidence road network state vector through Dempster synthesis rules. S3. Construct a multidimensional dynamic traffic stress index that includes psychological resistance factors; combine micro-vehicle operation parameters and macro-environmental parameters, and introduce a nonlinear mapping function of driver psychological anxiety under different red light waiting times to calculate a comprehensive stress index that reflects the real traffic load. S4. Establish a hierarchical and collaborative multi-agent reinforcement learning decision-making architecture; define an upper-level regional coordination agent to generate sub-regional boundary traffic control strategies, and define a lower-level intersection control agent to generate specific signal phase timing and lane function dynamic allocation instructions. S5 generates a spatiotemporal dual-dimensional collaborative traffic management strategy; the lower-level intersection control agent outputs traffic light cycle, green light ratio adjustment instructions, and switching instructions for tidal lanes, left turns by borrowing lanes, and variable directional lanes based on the real-time state space. S6, based on the standardized command issuance and closed-loop feedback of the NTCIP protocol, maps the generated control strategy into object identifier operation commands conforming to the NTCIP 1202 v03 standard, issues them to the traffic signal controller and roadside execution unit via the SNMP protocol, and uses the traffic flow response at the next moment to calculate the reward function value and update the reinforcement learning network parameters online.
[0008] Furthermore, to better implement the present invention, the specific algorithm flow for multi-source data trust fusion in step S2 includes: defining an identification framework. ,in Represents the i-th level of traffic congestion; for the sensor Evidence observed at time t Construct the basic probability assignment function ; Calculate any two sensors and Conflict coefficient between : ; like If the conflict exceeds a preset threshold, a trust level correction mechanism is activated; historical trust levels of the sensor are incorporated. Its update formula is:
[0009] in For memory decay factor, For sensors The matching degree between the previous judgment result and the fusion result; the corrected basic probability allocation is as follows: Final fusion result Calculated using orthogonal and rule-based methods.
[0010] Furthermore, to better realize the present invention, in step S3, the calculation model of the multidimensional dynamic traffic pressure index includes a physical congestion component. And psychological congestion The physical congestion component Calculated based on the weighted ratio of road segment saturation to travel time:
[0011] The psychological congestion component A driver anxiety growth function in the form of Logit is introduced:
[0012] in, This represents the average number of times a vehicle stops on this road segment. The number of parking stops is the anxiety threshold. The normalized severe weather index, The weather sensitivity coefficient is used; the final multidimensional dynamic traffic pressure index is... ,in The weighting coefficients are dynamic and vary with... Increases and decreases.
[0013] Furthermore, to better realize the present invention, in S4, the lower-level intersection control agent adopts the Soft Actor-Critic algorithm based on maximum entropy; state space This includes: current phase duration, vehicle queue length vectors for each approach lane, average vehicle speed for each lane, remaining space in the downstream road segment, and current lane function configuration status; the action space is a hybrid action space, including: discrete actions. This determines whether to switch phases; continuous action. This determines the duration of the green light in the next phase; discrete actions. This determines the activation status of the special lane; the reward function R is defined as:
[0014] in The term is the square of the queue length. For average vehicle delay, Penalties for frequently switching signals or lane functions. This represents the estimated carbon emissions.
[0015] Furthermore, to better realize the present invention, in step S5, the dynamic allocation logic of the tidal lane in the spatiotemporal dual-dimensional collaborative diversion strategy includes a hysteresis threshold control and a safe clearing protection mechanism; a positive flow rate is set. and reverse traffic The conditions for opening a tidal flow lane are:
[0016] The conditions for closing the tidal flow lane are:
[0017] in A hysteresis interval is formed; during the state transition, strict timing protection is implemented. Stage T1: The lane entrance displays a red cross, while the exit remains green, for a duration of... Phase T2: Confirm that there are no vehicles remaining in the lane; Phase T3: Switch the colors of variable message signs and smart road studs on the ground to allow reverse traffic.
[0018] Furthermore, to better implement the present invention, the NTCIP protocol mapping in S6 specifically includes: Using the SNMP (Simple Network Management Protocol) Set operation, write the following object value to the signal controller: phaseForceOff, OID: 1.3.6.1.4.1.1206.4.2.3.4, is used to forcibly terminate the current phase; phaseHold, OID: 1.3.6.1.4.1.1206.4.2.3.1, is used to maintain a green light for a specific phase; maxGreen, OID:1.3.6.1.4.1.1206.4.2.3.2.2, is used to dynamically modify the maximum green light time parameter; specialFunctionOutput, OID:1.3.6.1.4.1.1206.4.2.3.8, is used to control the tidal lane screen or variable guidance sign driver connected to the signal.
[0019] Furthermore, in order to better realize the present invention, the diversion method also includes an operation method with an extreme event takeover module. When a traffic accident or extreme weather is detected, the system automatically interrupts the output of the reinforcement learning model and switches to a rule-based expert system. The expert system executes the contingency plan strategy, including clearing all red lines, mandatory green wave diversion in specific directions, and issuing diversion guidance information at surrounding intersections.
[0020] The beneficial effects of this invention are: The accuracy of the fused traffic parameters is significantly improved compared to single-sensor systems under adverse weather conditions. Through synchronized adjustment of spatiotemporal resources, intersection capacity is greatly enhanced in scenarios with significant tidal flow. The control strategy based on a multi-dimensional dynamic traffic pressure index reduces extreme waiting times for drivers in adverse environments, lowering the risk of secondary accidents caused by road rage. The NTCIP-based design significantly reduces system deployment and modification costs. Attached Figure Description
[0021] Figure 1 This is a flowchart illustrating the overall process of the method of the present invention. Detailed Implementation
[0022] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.
[0023] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention.
[0024] like Figure 1 As shown, this embodiment uses a typical urban main road intersection (including tidal flow lanes) as a scenario to illustrate the system's construction and operation mechanism.
[0025] I. System Architecture and Data Awareness This invention employs a three-layer architecture: edge, cloud, and terminal. In the perception layer, a radar-visual integrated unit is installed at each approach lane to provide long-range target tracking; simultaneously, it connects to roadside units to acquire V2X data and to micro weather stations to obtain environmental data. In the computing layer, an edge computing gateway is deployed to run data fusion algorithms and reinforcement learning inference models, and communicates with traffic signals via an NTCIP proxy service.
[0026] II. Data Fusion Based on DS Evidence Theory To address the data conflict issue of sensors under adverse weather conditions (e.g., radar detects a moving target but video fails to detect it due to rain interference), the system first constructs a Basic Probability Assignment (BPA) for each sensor. Next, it calculates the conflict coefficient. If the conflict coefficient exceeds a threshold, the evidence sources are weighted and corrected based on the sensor's historical confidence level (dynamically updated by historical accuracy). Finally, the corrected BPA is used for Dempster synthesis to output high-confidence traffic conditions, effectively eliminating false alarms.
[0027] III. Calculation of Multidimensional Dynamic Traffic Pressure Index The system calculates a stress index in real time, encompassing both physical and psychological dimensions. The physical component is calculated based on the ratio of traffic flow to capacity; the psychological component is calculated based on a driver anxiety function, which increases non-linearly with the number of red light waits and is adjusted for severe weather (the more severe the weather, the lower the anxiety threshold). The final comprehensive index serves as the basis for triggering intensive traffic control strategies.
[0028] IV. Spatiotemporal Coordination Reinforcement Learning Control The intersection control agent operates based on the Soft Actor-Critic algorithm. When it detects excessively long queues in a certain direction and an excessive overall pressure index, the agent generates a hybrid action: in the time dimension, it dynamically extends the green light time for that direction; in the spatial dimension, if the oncoming lane is empty, it issues a command to activate the tidal flow lane or the left-turn function. The reward function is designed to guide the agent to reduce delays, lower carbon emissions, and avoid the safety hazards caused by frequent switching.
[0029] V. Safety Switching Logic for Tidal Flow Lanes When the agent decides to activate the tidal flow lane, the system implements strict timing protection: Warning period: Change notifications are displayed via VMS.
[0030] Clearing period: The upstream entrance signal light for this lane is closed to traffic, the lane displays a red cross, and the system waits for the calculated theoretical clearing time.
[0031] Confirmation period: The sensing device is invoked to confirm that there are no vehicles remaining in the lane.
[0032] Initial phase: The color of the smart road studs on the ground is flipped, a green arrow is displayed at the entrance, and the traffic lights are activated in accordance with the new timing scheme.
[0033] VI. NTCIP-based command issuance The edge gateway translates AI decisions into standard SNMP commands. For example, it can maintain a green light by setting a phaseHold object, force a red light to end by setting a phaseForceOff object, and drive a reversible lane sign by setting a specialFunctionOutput object. The system also monitors the status objects fed back by the traffic lights to ensure that the commands are executed correctly.
[0034] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Any other modifications or equivalent substitutions made by those skilled in the art to the technical solutions of the present invention, as long as they do not depart from the spirit and scope of the technical solutions of the present invention, should be covered within the scope of the claims of the present invention.
Claims
1. A traffic congestion mitigation method that comprehensively utilizes intelligent technology and dynamic strategies, characterized in that, Includes the following steps: S1. Construct a multi-dimensional heterogeneous traffic perception network and perform data cleaning; collect holographic traffic flow parameters by deploying millimeter-wave radar, high-point surveillance video, geomagnetic sensors, floating car data and meteorological environmental data through API interfaces on the roadside; use an anomaly detection algorithm based on isolated forest to remove sensor noise, and use Kalman filtering to perform spatiotemporal interpolation to complete the missing data. S2, based on the improved Dempster-Shafer evidence theory, establishes a multi-source data trust fusion; establishes an identification framework including smooth traffic, slow traffic, light congestion, severe congestion and deadlock; for the conflict of different sensors' state judgments of the same traffic section, introduces a time decay factor based on historical accuracy and sensor dynamic trust weights, calculates the corrected basic probability allocation, and outputs a high-confidence road network state vector through Dempster synthesis rules. S3. Construct a multidimensional dynamic traffic stress index that includes psychological resistance factors; combine micro-vehicle operation parameters and macro-environmental parameters, and introduce a nonlinear mapping function of driver psychological anxiety under different red light waiting times to calculate a comprehensive stress index that reflects the real traffic load. S4, establish a hierarchical and collaborative multi-agent reinforcement learning decision-making architecture; Define an upper-level regional coordination agent to generate sub-region boundary traffic control strategies, and define a lower-level intersection control agent to generate specific signal phase timing and lane function dynamic allocation instructions. S5 generates a spatiotemporal dual-dimensional collaborative guidance strategy; The lower-level intersection control agent outputs traffic light cycle, green light ratio adjustment instructions, and switching instructions for tidal lanes, left turns by using other lanes, and variable directional lanes based on the real-time state space. S6, based on the standardized command issuance and closed-loop feedback of the NTCIP protocol, maps the generated control strategy into object identifier operation commands conforming to the NTCIP 1202 v03 standard, issues them to the traffic signal controller and roadside execution unit via the SNMP protocol, and uses the traffic flow response at the next moment to calculate the reward function value and update the reinforcement learning network parameters online.
2. The traffic congestion mitigation method that comprehensively utilizes intelligent technology and dynamic strategies according to claim 1, characterized in that: In step S2, the specific algorithm flow for multi-source data trust fusion includes: defining the identification framework. ,in Represents the i-th level of traffic congestion; for the sensor Evidence observed at time t Construct the basic probability assignment function ; Calculate any two sensors and Conflict coefficient between : ； like If the conflict exceeds a preset threshold, a trust level correction mechanism is activated; historical trust levels of the sensor are incorporated. Its update formula is: ; in For memory decay factor, For sensors The matching degree between the previous judgment result and the fusion result; the corrected basic probability allocation is as follows: The final fusion result is calculated using orthogonal sums and rules.
3. The traffic congestion mitigation method that comprehensively utilizes intelligent technology and dynamic strategies according to claim 1, characterized in that: In S3, the calculation model for the multidimensional dynamic traffic pressure index includes a physical congestion component. And psychological congestion The physical congestion component Calculated based on the weighted ratio of road segment saturation to travel time: ； The psychological congestion component A driver anxiety growth function in the form of Logit is introduced: ； in, This represents the average number of times a vehicle stops on this road segment. The number of parking stops is the anxiety threshold. The normalized severe weather index, The weather sensitivity coefficient is used; the final multidimensional dynamic traffic pressure index is... ,in The weighting coefficients are dynamic and vary with... Increases and decreases.
4. The traffic congestion mitigation method that comprehensively utilizes intelligent technology and dynamic strategies according to claim 1, characterized in that: In S4, the lower-level intersection control agent adopts the Soft Actor-Critic algorithm based on maximum entropy; state space This includes: current phase duration, vehicle queue length vectors for each approach lane, average vehicle speed for each lane, remaining space in the downstream road segment, and current lane function configuration status; action space. The mixed action space includes: discrete actions This determines whether to switch phases; continuous action. This determines the duration of the green light in the next phase; discrete actions. This determines the activation status of the special lane; the reward function R is defined as: ； in The term is the square of the queue length. For average vehicle delay, Penalties for frequently switching signals or lane functions. This represents the estimated carbon emissions.
5. The traffic congestion mitigation method that comprehensively utilizes intelligent technology and dynamic strategies according to claim 1, characterized in that: In S5, the dynamic allocation logic of the tidal lane in the spatiotemporal dual-dimensional collaborative diversion strategy includes hysteresis threshold control and a safe clearing protection mechanism; a positive flow rate is set. and reverse traffic The conditions for opening a tidal flow lane are: ； The conditions for closing the tidal flow lane are: ； in A hysteresis interval is formed; during the state transition, strict timing protection is implemented. Stage T1: The lane entrance displays a red cross, while the exit remains green, for a duration of... Phase T2: Confirm that there are no vehicles remaining in the lane; Phase T3: Switch the colors of variable message signs and smart road studs on the ground to allow reverse traffic.
6. The traffic congestion mitigation method that comprehensively utilizes intelligent technology and dynamic strategies according to claim 1, characterized in that: The NTCIP protocol mapping in S6 specifically includes: Using the SNMP (Simple Network Management Protocol) Set operation, write the following object value to the signal controller: phaseForceOff, OID: 1.3.6.1.4.1.1206.4.2.3.4, is used to forcibly terminate the current phase; phaseHold, OID: 1.3.6.1.4.1.1206.4.2.3.1, is used to maintain a green light for a specific phase; maxGreen, OID:1.3.6.1.4.1.1206.4.2.3.2.2, is used to dynamically modify the maximum green light time parameter; specialFunctionOutput, OID:1.3.6.1.4.1.1206.4.2.3.8, is used to control the tidal lane screen or variable guidance sign driver connected to the signal.
7. The traffic congestion mitigation method that comprehensively utilizes intelligent technology and dynamic strategies according to claim 1, characterized in that: The traffic management method also includes an operation method with an extreme event takeover module. When a traffic accident or extreme weather is detected, the system automatically interrupts the output of the reinforcement learning model and switches to a rule-based expert system. The expert system executes the contingency plan strategy, including clearing all red lines, mandatory green wave evacuation in specific directions, and issuing diversion guidance information at surrounding intersections.
8. A traffic congestion mitigation system that implements the traffic congestion mitigation method that comprehensively utilizes intelligent technology and dynamic strategies as described in any one of claims 1 to 7, characterized in that, The traffic congestion mitigation system includes edge sensing nodes, a regional control server, roadside execution units, and a data communication network. The edge sensing nodes integrate millimeter-wave radar and AI cameras, and have built-in high-performance GPU computing units to run data fusion algorithms. The regional control server is deployed in the cloud and runs an upper-level regional coordination agent to perform macro-strategy calculations. The roadside execution units include signal controllers, variable lane controllers, and V2X communication modules that conform to the NTCIP standard. The data communication network adopts a hybrid networking architecture based on fiber optic backhaul and 5G wireless backup.
9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the traffic congestion relief method that integrates intelligent technology and dynamic strategies as described in any one of claims 1 to 7.
10. The traffic congestion mitigation method that comprehensively utilizes intelligent technology and dynamic strategies according to claim 4, characterized in that: Carbon emissions in the reward function The estimation is based on the VT-Micro model, according to the vehicle's instantaneous speed. and acceleration Real-time calculation: ； These are the emission coefficients calibrated for different vehicle models.