Road traffic vehicle dynamic command system based on multi-agent cooperative control

Abstract

The invention discloses a road traffic vehicle dynamic command system based on multi-agent cooperative control, which relates to the field of intelligent traffic systems and comprises a vehicle dynamic ad hoc network layer, a roadside area cooperation layer and a cloud global planning layer. The vehicle dynamic ad hoc network layer is used for carrying out state sharing, formation control and emergency obstacle avoidance among vehicles, and uploading the collected vehicle state data to the roadside area coordination layer; the roadside area collaboration layer is used for carrying out aggregation processing on the vehicle state data, generating a signal lamp dynamic adjustment and vehicle behavior instruction, and uploading the area traffic state data to the cloud global planning layer; and the cloud global planning layer is used for performing macroscopic traffic flow optimization, cross-regional coordination and global priority arbitration, generating a cross-regional optimization instruction and issuing the cross-regional optimization instruction to the roadside region coordination layer and the vehicle dynamic ad hoc network layer. According to the system, through multi-agent three-level cooperative control and dynamic priority arbitration, the road traffic efficiency, the safety and the mixed traffic flow cooperative capability are improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent transportation systems, and more specifically to a road traffic vehicle dynamic command system based on multi-agent cooperative control. Background Technology

[0002] Modern intelligent transportation systems are rapidly developing towards connectivity, collaboration, and intelligence, aiming to address the increasingly complex urban road and highway environments, especially mixed traffic scenarios where human and autonomous vehicles coexist. These systems require deep integration of vehicles, roadside infrastructure, and cloud platforms to achieve comprehensive traffic optimization and safety control.

[0003] However, existing systems often employ relatively independent control levels, lacking cross-level collaborative mechanisms. For example, some solutions focus on local signal optimization through roadside units or rely on vehicle ad hoc networks for platooning, but fail to effectively close the loop between real-time vehicle status, regional collaborative decision-making, and global traffic scheduling. This makes it difficult for the system to cope with large-scale dynamic traffic flow changes, and the information barriers between different levels often limit optimization objectives to local areas, preventing the achievement of true system-level efficiency improvements.

[0004] In terms of mixed traffic flow coordination, existing technologies struggle to unify and coordinate the differentiated decision-making mechanisms of manned and autonomous vehicles. Manned vehicles rely on the driver's subjective judgment, while autonomous vehicles strictly adhere to algorithmic control. This can easily lead to behavioral conflicts in scenarios such as intersection passage, ramp merging, and lane changes. Although some multi-agent cooperative control methods exist, a unified priority arbitration mechanism that integrates vehicle type, task urgency, and credit behavior is still lacking, hindering efficient and fair dynamic allocation of road rights.

[0005] Furthermore, existing systems have significant shortcomings in dynamic response and credit constraints. Traditional signal control models cannot incorporate real-time traffic flow and vehicle priorities, making it difficult to achieve rapid responses in scenarios such as green wave traffic and congestion mitigation. At the same time, the lack of effective credit assessment and joint penalty mechanisms to address the uncertainty of human driving behavior prevents the system from guiding compliant driving behavior and also limits data support for derivative services such as insurance.

[0006] Therefore, how to design a road traffic vehicle dynamic command system based on multi-agent collaborative control that can integrate vehicle, roadside and cloud capabilities to achieve global optimization, hybrid traffic collaboration and credit constraints is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0007] In view of this, the present invention provides a road traffic vehicle dynamic command system based on multi-agent cooperative control, which aims to solve the problems of difficulty in coordination, slow response and lack of global optimization in the existing traffic system under mixed traffic flow environment, realize unified, real-time and optimal cooperative control of manned and unmanned vehicles, and ultimately improve the overall efficiency and safety of the road network.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: This invention provides a road traffic vehicle dynamic command system based on multi-agent cooperative control, including a vehicle dynamic self-organizing network layer, a roadside area cooperation layer, and a cloud-based global planning layer; The vehicle dynamic self-organizing network layer is used for vehicle status sharing, platoon control and emergency obstacle avoidance, and uploads the collected vehicle status data to the roadside area collaboration layer. The roadside area coordination layer is used to aggregate and process vehicle status data, generate dynamic adjustment of traffic lights and vehicle behavior instructions, and upload regional traffic status data to the cloud global planning layer. The cloud-based global planning layer is used for macro-level traffic flow optimization, cross-regional coordination, and global priority arbitration. It generates cross-regional optimization instructions and sends them to the roadside regional collaboration layer and the vehicle dynamic self-organizing network layer.

[0009] Preferably, in the vehicle dynamic ad hoc network layer, state sharing includes: vehicles sharing their own location, speed, vehicle type, task urgency, credit score, lane affiliation, and lateral offset via V2X communication.

[0010] Preferably, in the vehicle dynamic ad hoc network layer, the formation control adopts the Model Predictive Control (MPC) algorithm, whose objective function is... Represented as:

[0011] in, This indicates the longitudinal distance between the vehicle and the vehicle in front at the current moment. Indicates the longitudinal distance between the target vehicles in the formation. This indicates the vehicle's speed at the current moment. Indicates the speed of the formation target. This indicates the vehicle's acceleration at the current moment. This indicates the vehicle's lateral deviation at the current moment. Indicates the horizontal position of the target. Indicates the cost of changing lanes. , , , , This represents the weighting coefficient.

[0012] Preferably, in the vehicle dynamic ad hoc network layer, emergency obstacle avoidance employs a potential field algorithm, whose total potential field function is... Represented as:

[0013]

[0014]

[0015] Vehicles using gradient descent algorithm Real-time solution for collision-free paths; in, This represents the set of dynamic obstacles within the current perception range of the vehicle. This represents the gravitational potential field pointing towards the target path point. Let X represent the repulsive potential field with respect to the i-th obstacle in the obstacle set, and let X represent the current position of the vehicle. This indicates the location of the target path point given by global path planning. This indicates the current position of the i-th obstacle. Represents the gravitational gain coefficient. This represents the repulsive force gain coefficient against the i-th obstacle. This represents the safe distance threshold from the i-th obstacle. This represents the step size coefficient for gradient descent.

[0016] Preferably, in the roadside area coordination layer, the generation of traffic light dynamic adjustment commands adopts an improved Webster model, whose objective function is... Represented as:

[0017] Where C represents the signal period; This represents the green light duration for the i-th phase; This represents the flow rate ratio of the i-th lane; i This represents the dynamic priority of the i-th lane; Represents the multi-lane coordination factor of the i-th lane; , These are the weighting coefficients.

[0018] Preferably, in the roadside area coordination layer, the generation of vehicle behavior instructions includes: Provides lane-level guidance information and green wave speeds displayed in AR-HUD for manned vehicles. It is suggested that the green wave vehicle speed... Represented as:

[0019] in, This indicates the remaining distance between the vehicle and the next green light window. Indicates the lane correction factor. Indicates the remaining time of the green light. Indicates the current time. Indicates the safety factor. This represents the lane change correction factor; In addition, it provides MPC control commands for autonomous vehicles, with control inputs including longitudinal acceleration. [-3.0,2.0] and lateral acceleration .

[0020] Preferably, in the cloud-based global planning layer, the macro-level traffic flow optimization adopts a mixed-integer linear programming (MILP) model, with the objective function being:

[0021] in, Let N represent the delay time of the i-th vehicle, and N represent the total number of vehicles. This represents the priority weight of the i-th vehicle, and M represents the total number of lanes on the road. This represents the real-time traffic density of lane l. This represents the average traffic density across all lanes. This is the lane balance weighting coefficient.

[0022] Preferably, in the cloud-based global planning layer, cross-regional coordination includes regional traffic flow prediction based on an OD matrix prediction model, wherein: In a typical highway scenario, when the regional traffic density... ≥threshold If the red light cycle continues for 5 minutes, it will trigger an extension of 10% to 15% of the upstream red light cycle and guide 20% of vehicles to detour to secondary roads; In highway scenarios, based on regional traffic density The system dynamically controls the number of vehicles merging into the ramp to be ≤30 vehicles / minute, and triggers cluster merging or splitting commands when the distance between adjacent clusters is <300m.

[0023] Preferably, the regional traffic flow prediction based on the OD matrix prediction model uses a lane-level density prediction formula:

[0024] in, This represents the predicted traffic density for lane l at the next time step. This represents the traffic density of lane l at the current moment. The smoothing factor represents the smoothing factor for lane l. Indicates that at time t, from region Drive through lane 1 towards the area The number of vehicles, Indicates that at time t, from region The total number of vehicles exiting through lane l; Indicates time From the lane The proportion of vehicles transferred to lane l.

[0025] Preferably, in the cloud-based global planning layer, the global priority arbitration is based on a blockchain credit mechanism, and the deduction of points for violations is calculated by combining multiple violations:

[0026] in, Indicates the speeding value. This indicates the number of violations such as dangerous lane changes, cutting in front of other lanes, or failing to follow lane-level instructions. These are the corresponding weighting coefficients.

[0027] As can be seen from the above technical solution, compared with the prior art, the technical solution of the present invention has the following beneficial effects: 1. The system achieves comprehensive coordination from micro-level vehicle control to macro-level traffic optimization through a three-level closed-loop architecture of vehicle dynamic self-organizing network, roadside area collaboration and cloud-based global planning, effectively improving road traffic efficiency and traffic flow.

[0028] 2. By integrating MPC formation control, potential field method emergency obstacle avoidance, dynamic signal timing and multimodal human-machine interaction commands, the system significantly enhances the safety, stability and emergency response capability of vehicle driving.

[0029] 3. The system utilizes a dynamic priority arbitration mechanism and a blockchain credit system to achieve unified behavior coordination and standardized management of manned and unmanned vehicles in mixed traffic environments, thereby improving the overall synergy and orderliness of the system. Attached Figure Description

[0030] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0031] Figure 1 A framework diagram of a road traffic vehicle dynamic command system based on multi-agent cooperative control is provided for an embodiment of the present invention; Figure 2This is a schematic diagram illustrating the working process of a road traffic vehicle dynamic command system based on multi-agent cooperative control, provided as an embodiment of the present invention. Detailed Implementation

[0032] 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 some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0033] like Figure 1 As shown, this embodiment provides a road traffic vehicle dynamic command system based on multi-agent cooperative control, including a vehicle dynamic self-organizing network layer, a roadside area cooperation layer, and a cloud-based global planning layer; The vehicle dynamic self-organizing network layer is used for vehicle status sharing, formation control and emergency obstacle avoidance, and uploads the collected vehicle status data to the roadside area collaboration layer. The roadside area coordination layer is used to aggregate and process vehicle status data, generate dynamic adjustment of traffic lights and vehicle behavior instructions, and upload regional traffic status data to the cloud global planning layer. The cloud-based global planning layer is used for macro-level traffic flow optimization, cross-regional coordination, and global priority arbitration. It generates cross-regional optimization instructions and sends them to the roadside regional collaboration layer and the vehicle dynamic self-organizing network layer.

[0034] The system defines a core three-tier architecture. Based on its functional coordination logic, the three layers form a closed-loop data interaction: the vehicle dynamic self-organizing network layer acts as the data acquisition front end, acquiring vehicle status in real time and coordinating initially, providing basic data support for upper-level decision-making; the roadside area coordination layer assumes the role of a local processing hub, quickly responding to traffic demands within the area through edge computing, while transmitting regional traffic status upwards and executing cloud commands downwards; the cloud-based global planning layer coordinates the overall situation from a macro perspective, solving cross-regional traffic coordination problems and avoiding overall efficiency losses caused by local optimization.

[0035] The key advantage of this architecture design lies in balancing real-time performance with global capabilities. It can quickly handle local emergencies through the roadside and vehicle layers, and achieve long-distance, cross-regional traffic optimization (such as global route planning for highways and congestion diversion in urban road networks) through the cloud layer. It can be adapted to various scenarios such as traditional manned driving, manned-unmanned hybrid driving, and fully autonomous driving, significantly improving road traffic efficiency, enhancing driving safety, and realizing the optimization and unified command of global traffic flow.

[0036] The following provides a further detailed description of each structure of the above system; In this embodiment, in the vehicle dynamic self-organizing network layer, vehicles share their own location, speed, vehicle type, task urgency, credit score, lane affiliation, and lateral offset through V2X communication; Here, vehicle type (such as ambulances and new energy vehicles) and task urgency are the core basis for dynamic priority allocation, directly affecting right-of-way allocation and traffic light timing; credit scores are linked to the driver's or vehicle's historical compliance behavior and are an important parameter for constraining violations, which can be stored on the blockchain to ensure immutability; position and speed data are the foundation for platoon control, emergency obstacle avoidance, and green wave speed calculation. For example, based on the position and speed difference of adjacent vehicles, it can be determined whether to adjust the platoon spacing or trigger obstacle avoidance commands; lane affiliation and lateral offset provide basic data for the roadside layer to generate multi-lane commands; in addition, these data are shared in real time through V2X communication, which can solve the perception blind spot problem caused by the reliance of traditional single-vehicle intelligence on its own sensors, and provide a unified data foundation for collaborative decision-making between manned and unmanned vehicles in mixed traffic flow.

[0037] Furthermore, the formation control employs the Model Predictive Control (MPC) algorithm, supporting both single-lane longitudinal formation and multi-lane parallel driving. Its objective function... Represented as:

[0038] in, This indicates the longitudinal distance between the vehicle and the vehicle in front at the current moment. Indicates the longitudinal distance between the target vehicles in the formation; This indicates the vehicle's speed at the current moment. Indicates the speed of the formation target; This indicates the vehicle's acceleration at the current moment; This indicates the vehicle's lateral deviation at the current moment, i.e., its distance from the lane centerline; Indicates the horizontal position of the target; Indicates the cost of changing lanes; Weighting coefficient , , , , The value has been optimized for multiple scenarios to balance security and traffic efficiency: Setting it to the highest level prioritizes maintaining a stable following distance and avoiding the risk of rear-end collisions, making it especially suitable for high-speed scenarios on highways. Secondly, ensure that the speed of vehicles in the convoy is synchronized to reduce the increase in energy consumption caused by frequent acceleration and deceleration; Used to reduce acceleration fluctuations Used to control lateral stability, ensuring that the vehicle remains stable within the lane or during lane changes; This is used to weigh the necessity and risks of lane changing, encouraging lane changing when efficient traffic flow is needed, and suppressing unnecessary lane changing when traffic is heavy.

[0039] In addition, target vehicle distance and formation target speed and the horizontal position of the target It can be dynamically adjusted according to the scenario; for example, it can be reduced during peak hours of urban road congestion. To improve road utilization, higher traffic flow times can be set for highways. To improve traffic efficiency; in multi-lane scenarios, The algorithm can dynamically specify the target lane for vehicles based on traffic conditions. Through rolling optimization, it adjusts the control input in real time according to the deviation of vehicle state, which can effectively deal with dynamic interference during vehicle movement and ensure the control robustness of the formation in single-lane and multi-lane environments.

[0040] Furthermore, the emergency obstacle avoidance employs a potential field algorithm, whose total potential field function is... It is composed of the superposition of gravitational potential field and repulsive potential field, where the repulsive potential field takes into account the influence of obstacles from both the same lane and adjacent lanes, and is expressed as:

[0041]

[0042]

[0043] Vehicles using gradient descent algorithm Real-time solution for collision-free paths; in, This represents the set of dynamic obstacles within the current vehicle's perception range. This set is updated in real time via V2X communication and may include obstacles in front and behind in the same lane, as well as side-by-side or cutting-in obstacles in adjacent lanes. This represents the gravitational potential field pointing towards the target path point; X represents the repulsive potential field targeting the i-th obstacle in the obstacle set; X represents the vehicle's current position. This indicates the location of the target path point given by global path planning; Indicates the current position of the i-th obstacle; Indicates the gravitational gain coefficient; This represents the repulsive force gain coefficient for the i-th obstacle, which can be set differently depending on the lane to which the obstacle belongs and its relative motion state; This represents the safe distance threshold from the i-th obstacle, which can be dynamically adjusted based on relative speed and direction. This represents the step size coefficient of gradient descent; The potential field algorithm simulates physical attraction and repulsion to plan a safe path for vehicles in emergency situations. The gravitational field ensures that the vehicle tends towards the target, while the repulsive field keeps it away from obstacles. The algorithm solves the problem in real time using the gradient descent method, ensuring the speed of obstacle avoidance response and the naturalness of the path.

[0044] In this embodiment, the roadside area coordination layer uses an improved Webster model to generate dynamic traffic light adjustment commands, with the objective function... Represented as:

[0045] Where C represents the signal period; This represents the green light duration for the i-th phase; This represents the flow rate ratio of the i-th lane; i This represents the dynamic priority of the i-th lane; Represents the multi-lane coordination factor of the i-th lane; , These are the weighting coefficients; The dynamic priority Calculation based on aggregated information of vehicles within the lane:

[0046] in , , These represent the average weight of vehicle type, average task urgency, and average credit score within that lane, respectively.

[0047] The multi-lane coordination factor Used to coordinate traffic flow between adjacent lanes:

[0048] in This represents the set of lanes adjacent to the i-th lane. , These represent the real-time traffic density of lanes i and j, respectively. The improved Webster model incorporates dynamic priority and multi-lane coordination factors into the signal timing model, enabling the traffic light control strategy to no longer rely solely on traffic flow, but also respond to the urgency of different vehicles and the traffic flow balance needs between multiple lanes. This improves traffic efficiency while also ensuring fairness, special service guarantees, and overall road network efficiency.

[0049] Furthermore, generating vehicle behavior instructions includes: Provides lane-level guidance information and green wave speeds displayed in AR-HUD for manned vehicles. The suggested calculation formula is:

[0050] in, This indicates the remaining distance between the vehicle and the next green light window. Indicates the remaining time of the green light. Indicates the current time. Indicates the safety factor; This represents the lane correction factor, which is dynamically set according to the vehicle's lane position: 1.0 for the inner lane, 0.95 for the middle lane, and 0.9 for the outer lane. This represents the lane change correction factor, which is 0.85 when the vehicle needs to execute a lane change command and 1.0 when it does not change lanes. Provides MPC control commands for autonomous vehicles, with control inputs including longitudinal acceleration. [-3.0, 2.0] and lateral acceleration Its constraints are: Longitudinal safety distance constraints: , where d represents the distance to the vehicle in front, and v represents the current vehicle speed; Lane boundary constraints: ,in This indicates the vehicle's lateral deviation at the current moment. Indicates the total width of the road. Indicates the width of the vehicle; Lane change safety constraints: When executing a lane change command, the distance between vehicles in adjacent lanes must be maintained. ; This section addresses the differences between manned and autonomous vehicles by designing specific behavioral commands. For manned vehicles, the AR-HUD displays lane-level guidance information and green wave speed (vopt) suggestions, incorporating a lane correction coefficient. and lane change correction factor It can provide differentiated and more accurate traffic suggestions for vehicles in different lanes, and reserve safety redundancy based on theoretically calculated vehicle speed to avoid the risk of running red lights due to driver delays or road conditions; at the same time, the green wave speed combined with the real-time green light remaining time and distance calculation can guide drivers to drive smoothly, reduce sudden acceleration and braking, and reduce energy consumption and accident risk.

[0051] For autonomous vehicles, the lateral acceleration control quantity a1 of the MPC control command enables them to accurately track the target lane or perform cooperative lane changes. Lane boundary constraints ensure that they always drive within the legal range, while lane change safety constraints ensure the safety of cross-lane behavior by communicating with adjacent vehicles. The longitudinal acceleration range and safe distance constraints have been verified for safety, and the lateral acceleration range ensures the smoothness and stability of the lane change process. Both types of commands are generated based on real-time data from the roadside layer, which can effectively coordinate the driving behavior of manned and autonomous vehicles in multi-lane environments and reduce problems such as lane change conflicts and intersection congestion.

[0052] In this embodiment, the macro-traffic flow optimization in the cloud-based global planning layer adopts a mixed-integer linear programming (MILP) model, with the objective function being:

[0053] in, Let N represent the delay time of the i-th vehicle, and N represent the total number of vehicles. Let represent the priority weight of the i-th vehicle, and The sum of the priority weights of all vehicles equals 1; M represents the total number of lanes on the road. This represents the real-time traffic density of lane l. This represents the average traffic density across all lanes. Lane equalization weight coefficient; the priority weight Adjustments based on the lane type the vehicle is in: ,in The lane type correction factor is 1.2 for emergency lanes, 1.0 for main lanes, and 0.8 for auxiliary lanes; In addition, the MILP model here needs to meet the cross-lane traffic transfer constraint: the proportion of cross-lane traffic transfer in any lane within a single cycle shall not exceed 30%; its objective function is to minimize the total delay, and the concept of priority is introduced through weight coefficients. Combined with lane balance optimization, it ensures that the delay of high-priority vehicles is reduced to a greater extent, while improving the overall throughput of the road network. Furthermore, the constraints ensure the feasibility of the optimization scheme and the efficiency of multi-lane coordination.

[0054] Furthermore, cross-regional coordination includes regional traffic flow prediction based on OD matrix prediction models. In the case of ordinary highways, when the regional traffic density... If the red light cycle continues for 5 minutes, it will trigger an extension of 10% to 15% of the upstream red light cycle and guide 20% of vehicles to detour to secondary roads. The guidance strategy prioritizes vehicles in the outer lanes, while vehicles in the inner and middle lanes need to receive lane change instructions at least 2 intersections in advance. In highway scenarios, based on regional traffic density The system dynamically controls the merging rate of ramps to ≤30 vehicles / minute and triggers merging or splitting commands when the distance between adjacent multi-lane platoons is <300m. Merging requires a lateral spacing ≥2.5m, and the speed difference between the target lane platoon and the current lane platoon must be ≤5km / h during cross-lane merging. Merging vehicles are dynamically assigned to merging lanes based on the lane density of the main road, with the inner lane density on the main road ≥0.7. At that time, guide vehicles to merge into the outer lane.

[0055] Furthermore, regional traffic flow prediction is performed based on the OD matrix prediction model, and its prediction formula is lane-level density prediction:

[0056] in, This represents the predicted traffic density for lane l at the next time step. This represents the traffic density of lane l at the current moment. The smoothing factor for lane l; emergency lane and main lane auxiliary lane ; Indicates that at time t, from region Drive through lane 1 towards the area The number of vehicles, Indicates that at time t, from region The total number of vehicles exiting through lane l; Indicates time From the lane The proportion of vehicles transferred to lane l; In this location, under typical highway conditions, the regional flow density ρ ≥ the threshold. The 5-minute trigger condition ensures the accuracy of congestion response and avoids misoperation caused by short-term traffic flow fluctuations. The strategy of extending the upstream red light cycle by 10% to 15% and guiding 20% ​​of vehicles to detour alleviates congestion through traffic restriction and diversion. Extending the red light reduces the flow of vehicles entering the congested area, while detour guidance disperses traffic pressure. The combination of the two can quickly reduce the traffic density in the area. In particular, lane-level detour guidance ensures that vehicles can safely and orderly change from the inner lane to the outer detour lane, avoiding secondary congestion caused by improper guidance strategies.

[0057] In highway scenarios, controlling the number of vehicles merging into the ramp to ≤30 vehicles / minute can avoid mainline congestion caused by a large number of vehicles merging in at once, ensuring the efficiency of the highway mainline traffic. The triggering of merge / split commands when the distance between adjacent clusters is <300m is based on the needs of highway platooning. Through dynamic adjustment of the cluster, the reasonable platoon size is ensured, which not only improves traffic efficiency, but also facilitates platoon splitting and obstacle avoidance in emergency situations. The lateral spacing and speed difference constraints of cluster operations are the key to ensuring the safety of multi-lane platooning coordination.

[0058] Furthermore, the global priority arbitration is based on a blockchain credit mechanism, and the penalty points for violations are calculated by combining multiple violations:

[0059] in, Indicates the speeding value. This indicates the number of violations such as dangerous lane changes, cutting in front of other lanes, or failing to follow lane-level instructions. These are the corresponding weighting coefficients; The credit mechanism establishes a multi-lane collaborative compliance score. If there are no cross-lane violations and lane-level instructions are followed for 30 consecutive days, the credit score will be increased by 0.1 points, with a maximum accumulation of 1.0 points. The premium adjustment formula is revised as follows:

[0060] in, This indicates the current credit score. This indicates the total points deducted for multi-lane related violations over the past 6 months; The introduction of blockchain technology makes the generation and deduction of vehicle credit scores traceable and tamper-proof, avoiding human intervention. At the same time, multiple entities such as traffic management departments, insurance companies, and car manufacturers participate as nodes to achieve cross-platform sharing of credit data. For example, insurance companies can adjust premiums based on vehicle credit scores and multi-lane violation records, forming a linkage mechanism between violation points and premium increases. This further strengthens the constraints on drivers' dangerous lane changes, cutting in, and other behaviors that affect the efficiency of multi-lane coordination, providing a good traffic environment for system collaborative control.

[0061] Furthermore, the system supports dynamic scenario adaptation, including ramp merging, exiting, and emergency scenarios; among these, the ramp merging success rate... Represented as:

[0062] Diversion effect evaluation Represented as:

[0063] in, Indicates the degree of compliance with instructions. Indicates the window period utilization rate. This indicates the traffic density in the area before the diversion. This indicates the traffic density in the area after traffic diversion.

[0064] By quantifying the success rate of ramp merging, the system not only focuses on vehicle compliance with instructions but also introduces an assessment of the efficiency of coordination timing, aiming to guide vehicles not only to obey commands but also to cooperate efficiently. Simultaneously, the diversion effect evaluation provides the system with an objective and quantitative performance feedback mechanism. By comparing changes in regional traffic density before and after the implementation of measures, the effectiveness of the diversion strategy can be quickly determined, providing data-driven decision-making basis for subsequent global route planning and resource scheduling. This ensures that the system can continuously learn and optimize, improving the robustness and responsiveness of the overall road network.

[0065] like Figure 2 As shown below, the specific working process of the road traffic vehicle dynamic command system based on multi-agent cooperative control in this implementation will be described in detail: 1) Vehicle status perception and sharing; After the system is started, all types of vehicles on the road (manned and unmanned) collect their own status data in real time through their onboard sensors, including but not limited to precise location, instantaneous speed, vehicle type, urgency of the current task, and credit score calculated based on historical behavior. These data are shared in the ad hoc network via V2X communication modules (such as those based on IEEE 802.11p or C-V2X protocols) in the form of broadcast or multicast, forming a dynamic and transparent local information environment, providing a comprehensive and real-time data foundation for subsequent collaborative control.

[0066] 2) Regional data aggregation and local optimization decision-making; The roadside-deployed intelligent agents (integrating sensors such as millimeter-wave radar, lidar, and cameras, as well as edge computing units) continuously receive status data uploaded by all vehicles within their coverage area; the edge computing units quickly fuse, clean, and aggregate this massive and heterogeneous data to generate a real-time traffic situation profile of the current area. Based on this, this layer executes the core local optimization algorithm: ① Dynamic traffic light adjustment: using an improved Webster model incorporating dynamic priorities, the optimal green light time for each phase in the current cycle is calculated to minimize the total intersection delay; ② Vehicle behavior instruction generation: calculating and issuing green wave speed suggestions for manned vehicles via V2X and AR-HUD display, and generating and issuing precise MPC control instructions for autonomous vehicles; the processed regional state summary is synchronously uploaded to the cloud global planning layer.

[0067] 3) Macroeconomic situation analysis and overall optimization decision-making; The cloud center receives regional traffic status data uploaded from different roadside coordination layers and constructs a real-time digital twin model of the global road network; Based on this macro-level situation, the cloud platform executes two core functions: ① Macro-level traffic flow optimization: using optimization models such as MILP to minimize global total delay, it performs cross-regional route planning, traffic allocation, and congestion prediction; ② Global priority arbitration and cross-regional coordination: arbitrates conflicting right-of-way requests based on vehicle type, task urgency, and credit score; simultaneously, it predicts traffic flow direction based on the OD matrix and dynamically generates control instructions, such as issuing instructions in advance to restrict upstream ramp merging traffic or induce vehicles to detour when it detects that a section of highway is about to become congested.

[0068] 4) Optimize command issuance and vehicle execution; The global optimization instructions generated in the cloud are sent to the corresponding roadside area coordination layer. The roadside layer adjusts its local control strategy according to the cloud instructions and sends the executable instructions to the relevant vehicles through V2X communication. In addition, when roadside equipment fails, regional communication is interrupted, or global emergency dispatch is required, the cloud planning layer can rely on the wide area communication network to directly send optimization instructions to the vehicle dynamic self-organizing network layer as a necessary supplement and backup channel. After receiving instructions, the vehicle dynamic self-organizing network layer executes them according to its own driving mode: the AR-HUD interface of manned vehicles updates guidance information, and the driver operates accordingly; while unmanned vehicles directly input control commands into their autonomous driving algorithm to achieve precise execution.

[0069] 5) Closed-loop feedback and continuous optimization; After a vehicle executes a command, its new state (such as its new position and speed) is collected by sensors and uploaded via V2X, initiating a new data cycle. The roadside layer and the cloud layer compare the expected results with the actual results, and adaptively adjust and learn their respective models and parameters to form a continuously optimized closed-loop control system, ensuring that the system can effectively respond to the dynamic changes in traffic flow.

[0070] This embodiment of the road traffic vehicle dynamic command system based on multi-agent collaborative control organically integrates micro-level vehicle perception and control, meso-level roadside area collaboration, and macro-level global optimization scheduling into a whole through a closed-loop data flow and decision-making mechanism of a three-level collaborative architecture of vehicle-roadside-cloud. The system not only has the ability to adapt to different scenarios such as urban roads and highways, but also effectively solves the problem of coordination between manned and unmanned vehicles in mixed traffic flow through dynamic priority allocation, credit mechanism and multimodal human-machine interaction, showing significant effects in improving traffic efficiency, ensuring driving safety and reducing energy consumption.

[0071] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple; relevant parts can be referred to the method section.

[0072] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A road traffic vehicle dynamic command system based on multi-agent cooperative control, characterized in that, The vehicle dynamic ad hoc network layer, the roadside area coordination layer, and the cloud global planning layer are included. The vehicle dynamic ad hoc network layer is configured to share the state among vehicles, control the formation, and avoid obstacles in an emergency, and upload the collected vehicle state data to the roadside area coordination layer. The roadside area coordination layer is configured to aggregate the vehicle state data, generate a signal light dynamic adjustment and a vehicle behavior instruction, and upload the regional traffic state data to the cloud global planning layer. The cloud global planning layer is configured to optimize the macroscopic traffic flow, coordinate across regions, and perform global priority arbitration, generate a cross-regional optimization instruction, and distribute the instruction to the roadside area coordination layer and the vehicle dynamic ad hoc network layer. 2.The road traffic vehicle dynamic command system based on multi-agent collaborative control according to claim 1, wherein, In the vehicle dynamic ad hoc network layer, the state sharing includes sharing the position, speed, vehicle type, task urgency, credit score, lane ownership, and lateral offset of the vehicle through V2X communication.

3. The road traffic vehicle dynamic command system based on multi-agent collaborative control according to claim 1, characterized in that, In the vehicle dynamic ad hoc network layer, the platoon control adopts a model predictive control (MPC) algorithm, and a target function of the MPC algorithm is is represented as: wherein, denotes the longitudinal distance between the vehicle and the preceding vehicle at the current time, denotes the platoon target longitudinal vehicle distance, denotes the speed of the vehicle at the current time, denotes the platoon target speed, denotes the acceleration of the vehicle at the current time, denotes the lateral offset of the vehicle at the current time, denotes the target lateral position, denotes the lane change cost, , , , , denotes the weight coefficient.

4. The road traffic vehicle dynamic command system based on multi-agent collaborative control according to claim 1, characterized in that, In the vehicle dynamic ad hoc network layer, the emergency obstacle avoidance adopts a potential field algorithm, and a total potential field function thereof is is expressed as: Vehicle by gradient descent algorithm Real-time solving of collision-free paths; wherein, represents a set of dynamic obstacles within the current vehicle perception range, represents an attractive potential field pointing to the target path point, represents a repulsive potential field for the i-th obstacle in the obstacle set, X represents the current vehicle position, represents the target path point position given by the global path planning, represents the current position of the i-th obstacle, represents the attractive gain coefficient, represents the repulsive gain coefficient for the i-th obstacle, represents the safety distance threshold to the i-th obstacle, represents the step coefficient for gradient descent.

5. The road traffic vehicle dynamic command system based on multi-agent collaborative control according to claim 1, characterized in that, In the road side area coordination layer, the signal light dynamic adjustment instruction is generated by using an improved Webster model, and a target function of the improved Webster model is represented as: wherein C denotes a signal cycle; denotes the green time of the i-th phase; denotes the flow ratio of the i-th lane; i denotes the dynamic priority of the i-th lane; denotes the multi-lane coordination factor of the i-th lane; , is a weight coefficient.

6. The road traffic vehicle dynamic command system based on multi-agent collaborative control according to claim 1, characterized in that, In the roadside area coordination layer, generating the vehicle behavior instruction includes: Lane-level guidance information and green wave vehicle speed for providing AR-HUD display for a human-driven vehicle It is suggested that the green wave vehicle speed is represented as: wherein, represents a remaining distance of the vehicle to the next green light window, represents a lane correction coefficient, represents a green light remaining time, represents a current time, represents a safety coefficient, represents a lane change correction coefficient; and providing MPC control instructions for the autonomous vehicle, the control inputs including longitudinal acceleration [-3.0, 2.0] and lateral acceleration .

7. The road traffic vehicle dynamic command system based on multi-agent collaborative control according to claim 1, characterized in that, In the cloud global planning layer, the macroscopic traffic flow optimization adopts a mixed integer linear programming (MILP) model, and the objective function is: wherein, denotes the delay time of the ith vehicle, N denotes the total number of vehicles, denotes the priority weight of the ith vehicle, M denotes the total number of lanes of the road, denotes the real-time traffic density of the lth lane, denotes the average traffic density of all lanes, is the lane equalization weight coefficient. 8.The road traffic vehicle dynamic command system based on multi-agent collaborative control according to claim 1, wherein, In the cloud global planning layer, the cross-regional coordination includes predicting the regional traffic flow based on an OD matrix prediction model, wherein: In ordinary highway scenarios, when the regional traffic density ≥ threshold for 5 minutes, then trigger the upstream red light cycle to extend by 10%~15%, and guide 20% vehicles to detour to secondary roads; In the highway scenario, based on regional traffic density Dynamic control ramp on-vehicle ≤ 30 vehicles / minute, and trigger cluster merging or splitting instructions when the adjacent cluster distance < 300 m.

9. The road traffic vehicle dynamic command system based on multi-agent collaborative control according to claim 8, characterized in that, The prediction formula for predicting the regional traffic flow based on the OD matrix prediction model is the lane-level density prediction: in, This represents the predicted traffic density for lane l at the next time step. This represents the traffic density of lane l at the current moment. The smoothing factor represents the smoothing factor for lane l. Indicates that at time t, from region Drive through lane 1 towards the area The number of vehicles, Indicates that at time t, from region The total number of vehicles exiting through lane l; Indicates time From the lane The proportion of vehicles transferred to lane l.

10. The road traffic vehicle dynamic command system based on multi-agent collaborative control according to claim 1, characterized in that, In the cloud global planning layer, the global priority arbitration is based on a blockchain credit mechanism, and the score deduction for violating behavior is calculated by stacking multiple violating behaviors: wherein, represents an overspeed value, represents the number of violations of dangerous lane changing, crossing lane merging, or driving out of lane level instructions, is a corresponding weight coefficient.

Images