A dynamic topology-aware multi-agent cooperative scheduling method and device for an intelligent transportation system

Abstract

This invention relates to the field of traffic control system technology, and discloses a dynamic topology perception multi-agent cooperative scheduling method and device for intelligent transportation systems. The method includes: collecting the coordinates of intersection control nodes and their equipment installation angles to generate a phase conflict matrix; constructing a traffic network topology map and maintaining online status masking vectors for neighboring nodes, masking their characteristic parameters to update the scheduling status when a neighboring node becomes disconnected; determining a global congestion adaptive scaling factor based on the average queuing pressure, and calculating the priority score of the travel direction by combining queuing length, cumulative waiting time, and trunk line coordination gain. This invention utilizes masking vectors to maintain the continuity of scheduling logic under topology changes, guides distributed decision-making towards the optimal state of system traffic allocation through adaptive adjustment, collaboratively enhances the efficiency of trunk line traffic flow, and ensures the safety of phase scheduling.

Description

Technical Field

[0001] This invention relates to a dynamic topology sensing multi-agent cooperative scheduling method and device for intelligent transportation systems, belonging to the field of traffic control technology. Background Technology

[0002] Urban traffic signal management uses a central control platform or edge control terminal to determine the phase scheduling scheme based on a pre-set traffic network topology. This network topology involves the spatial location of intersections, the distribution of approach lanes, and phase conflict relationships, which are usually entered manually during the equipment installation phase.

[0003] In distributed sensing networks, roadside agents face dynamic conditions such as equipment maintenance, communication downtime, or the addition of new nodes. Because the control logic of existing systems is based on static configuration, real-time changes in physical space cannot be automatically translated to the scheduling logic layer. The lag in configuration data maintenance causes traffic signals to continue using the old conflict logic after topology changes, resulting in a deviation between phase commands and actual traffic paths, increasing the risk of intersection operation. Increasing the polling frequency of the central platform and edge nodes increases the communication load, and the continuity of local scheduling cannot be guaranteed in the environment of network link fluctuations. When dealing with the correlation between nodes, due to the lack of effective conversion of physical parameters such as sensor installation angles, the system is unable to automatically generate a phase conflict matrix that conforms to physical laws in a dynamic topology environment, resulting in a mismatch between the collaborative scheduling logic and the actual traffic flow situation.

[0004] Specifically, existing technologies suffer from the following shortcomings: 1. The manual maintenance mechanism of topology information leads to a time lag between control logic and physical reality; 2. Changes in neighbor nodes under dynamic operating conditions cause input loss or computational crashes in the cooperative scheduling model; 3. The lack of extraction of physical installation parameters makes it difficult to automatically define the phase conflict safety red line in an uncalibrated environment. Therefore, how to extract the physical installation posture of sensing devices and automatically convert it to generate phase conflict logic, and how to maintain the cooperative balance among multiple agents in a dynamic topology environment, are the technical problems to be solved by this invention. Summary of the Invention

[0005] To address the problems mentioned in the background art, the technical solution of the present invention is as follows: A dynamic topology sensing multi-agent cooperative scheduling method for intelligent transportation systems, comprising the following steps:

[0006] Step 101: Obtain the latitude and longitude coordinates and equipment installation angles uploaded by the control nodes at each intersection;

[0007] Step 102: Based on the equipment installation angle, the equipment orientation is converted into the road entrance direction by physical orientation comparison. The equipment installation orientation angle is vector-aligned with the road orientation angle in the high-precision map to simulate the virtual movement trajectory of different traffic directions. The vector cross product operation is performed to determine the set of intersection points of different trajectory lines in two-dimensional space. Based on this, a phase conflict matrix is ​​generated as the security constraint boundary of the logic layer.

[0008] Step 103: Using the control nodes at each intersection as graph nodes and the road connection relationships as directed edges, construct a traffic network topology graph and maintain online state masking vectors for neighboring nodes.

[0009] Step 104: Monitor heartbeat signals. When the disconnection time of a neighboring node exceeds a preset threshold, set the corresponding bit in the online state masking vector to zero. In the normalization layer of the graph attention network aggregation calculation, the online state masking vector is converted into a bias term and added to the attention score to force the attention weight of the disconnected node to approach zero, so as to update the scheduling state under the dynamic topology.

[0010] Step 105: Determine the average queuing pressure of the traffic network and transform the physical pressure characteristics of the entire network into control variables. Based on this, calculate the global congestion adaptive scaling factor according to the average queuing pressure.

[0011] Step 106: Combine the historical traffic flow trend prediction of the intersection to correct the signal timing parameters, obtain queue length data, cumulative waiting time weight, trunk line coordination gain and red light timeout penalty, use the global congestion adaptive scaling factor to weight the queue length data, and calculate the real-time priority score of each traffic direction.

[0012] Step 107: Based on the real-time priority score, select the passage direction with the highest score and its non-conflicting accompanying direction, and retrieve the corresponding signal control command according to the phase conflict matrix. Then, execute the blocking logic by identifying whether there is a passage conflict in the generated signal scheme.

[0013] Preferably, step 102 includes: step 1021, calculating the spatial connection structure between intersections based on latitude and longitude coordinates; step 1022, determining the vector orientation of the sensor coverage area based on the device installation angle; step 1023, comparing the vector orientation with a preset geomagnetic azimuth reference, and converting the sensor viewpoint into the road entrance direction; step 1024, analyzing the overlapping areas of traffic trajectories in the same intersection's road entrance directions, and marking the corresponding phase group as a conflict state in the phase conflict matrix when there are spatial conflict points in the traffic trajectories.

[0014] Preferably, in step 104, when the graph attention network aggregates the features of neighboring nodes, it multiplies the online state masking vector with the corresponding attention weights element by element, so that the feature update weight of the disconnected node on the feature of the current intersection control node is 0.

[0015] Preferably, in step 106, the real-time priority score for each traffic direction is determined. Follow these rules: ,in, This represents the real-time queue length for the corresponding direction of travel. This is a global congestion adaptive scaling factor. As a weight for cumulative waiting time, For trunk line coordination gain, This is a red light timeout penalty triggered by the duration of the red light.

[0016] Preferably, before step 106, the method further includes: step 1061, collecting historical traffic data at the intersection and generating a prediction result containing traffic trend prediction values ​​using a time series prediction model; step 1062, correcting the signal timing parameters based on the traffic trend prediction values, wherein the time span of the traffic trend prediction values ​​is 15 minutes.

[0017] Preferably, in step 103, the online status masking vector is dynamically updated based on the real-time monitored heartbeat signal. When the heartbeat signal interruption duration exceeds a preset threshold, topological isolation of the corresponding neighboring nodes in the traffic network topology map is automatically triggered.

[0018] Preferably, in step 106, the global congestion adaptive scaling factor is positively correlated with the average queuing pressure, establishing a negative feedback adjustment mechanism between local pressure and global equilibrium. When the average queuing pressure is greater than the preset saturation threshold, the green light gain of the overloaded inlet lane is suppressed by increasing the global congestion adaptive scaling factor, inducing distributed decision-making to suppress local greed and solving the synchronous oscillation problem in large-scale distributed control.

[0019] Preferably, in step 106, the trunk line coordination gain is determined based on the position weight of the current intersection in the preset green wave band. By increasing the priority scores of the upstream overflow direction and the downstream absorption direction, the continuity of trunk line traffic flow is maintained.

[0020] Preferably, step 105 includes: step 1051, calculating the instantaneous queue length of each controlled intersection in the traffic network; step 1052, calculating the average queue saturation of the entire network, and converting the average queue saturation of the entire network into a dimensionless parameter characterizing the global congestion level, as a control variable fed back to the control nodes of each intersection; in step 107, by comparing the control logic of each traffic direction through the phase conflict matrix, when a traffic conflict is identified in the generated signal timing scheme, the green light path driving signal of the conflict direction is cut off through the blocking logic.

[0021] A dynamic topology-sensing multi-agent cooperative scheduling device for intelligent transportation systems includes: a memory for storing computer program instructions; and a processor for executing the computer program instructions to implement the steps of a dynamic topology-sensing multi-agent cooperative scheduling method for intelligent transportation systems.

[0022] Compared with the prior art, the beneficial effects of the present invention are:

[0023] 1. By utilizing geographic coordinates collected by positioning sensors and device installation angles obtained by attitude sensors, this invention establishes a transformation mechanism for automatically mapping physical entity spatial attitude to logical layer safety constraints. This mechanism directly converts physical space installation parameters into safety red lines for the logical control layer, solving the problem of time lag between control logic and physical reality caused by manual pre-configuration, as pointed out in the background art. This self-generated phase conflict matrix mode eliminates the potential for logical conflicts caused by data entry deviations during manual pre-configuration, achieving real-time definition of phase safety boundaries under uncalibrated, dynamic access conditions, and ensuring the logical security of signal control strategies during system initialization and device reconnection phases.

[0024] 2. A dynamic graph topology maintenance mechanism based on online state mask vectors is established. When nodes go offline, fail, or new devices are added in the traffic network, the mask vector is multiplied element-wise with the attention weights to achieve deterministic logical isolation during the inference phase of the graph attention network. This hard masking method solves the problem of input loss or computational crashes in the cooperative scheduling model caused by changes in neighboring nodes under dynamic conditions. This approach ensures that the cooperative scheduling model can maintain the continuous transmission of feature information between adjacent nodes even when nodes are frequently added or removed dynamically. It avoids the decision oscillations caused by slow convergence of traditional attention mechanisms when features change abruptly, thus improving the system's robustness in complex network environments.

[0025] 3. A global congestion adaptive scaling factor is introduced into the signal phase priority calculation framework. Combined with 15-minute traffic trend prediction to correct signal timing parameters, this mechanism suppresses potential local scheduling oscillations that may occur during topology switching by the distributed agent from both spatiotemporal dimensions. When the entire network is in a high-saturation state, the scheduling priority of local traffic directions is dynamically adjusted based on the average queuing pressure of the entire network. When the entire network is in a high-saturation state, the change in the value of this factor induces each intersection to increase the phase rotation frequency and compress redundant green light duration, thereby effectively destroying the greedy scheduling tendency of local intersections, reducing the system-level spillover risk caused by queuing spread in one direction, and promoting the evolution of distributed decision-making towards the system-optimal traffic allocation. Attached Figure Description

[0026] Figure 1This is a flowchart of a dynamic topology sensing multi-agent cooperative scheduling method for intelligent transportation systems according to the present invention.

[0027] Figure 2 This is a schematic diagram of a dynamic topology perception multi-agent collaborative scheduling logic architecture for intelligent transportation systems according to the present invention.

[0028] The objectives, features, and advantages of this invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0029] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0030] A dynamic topology-sensing multi-agent cooperative scheduling method for intelligent transportation systems includes the following steps:

[0031] Step 101: Obtain the latitude and longitude coordinates and equipment installation angles uploaded by the control nodes at each intersection;

[0032] Step 102: Based on the equipment installation angle, the equipment orientation is converted into the road entrance direction by physical orientation comparison. The equipment installation orientation angle is vector-aligned with the road orientation angle in the high-precision map to simulate the virtual movement trajectory of different traffic directions. The vector cross product operation is performed to determine the set of intersection points of different trajectory lines in two-dimensional space. Based on this, a phase conflict matrix is ​​generated as the security constraint boundary of the logic layer.

[0033] Step 103: Using the control nodes at each intersection as graph nodes and the road connection relationships as directed edges, construct a traffic network topology graph and maintain online state masking vectors for neighboring nodes.

[0034] Step 104: Monitor heartbeat signals. When the disconnection time of a neighboring node exceeds a preset threshold, set the corresponding bit in the online state masking vector to zero. In the normalization layer of the graph attention network aggregation calculation, the online state masking vector is converted into a bias term and added to the attention score to force the attention weight of the disconnected node to approach zero, so as to update the scheduling state under the dynamic topology.

[0035] Step 105: Determine the average queuing pressure of the traffic network and transform the physical pressure characteristics of the entire network into control variables. Based on this, calculate the global congestion adaptive scaling factor according to the average queuing pressure.

[0036] Step 106: Combine the historical traffic flow trend prediction of the intersection to correct the signal timing parameters, obtain queue length data, cumulative waiting time weight, trunk line coordination gain and red light timeout penalty, use the global congestion adaptive scaling factor to weight the queue length data, and calculate the real-time priority score of each traffic direction.

[0037] Step 107: Based on the real-time priority score, select the passage direction with the highest score and its non-conflicting accompanying direction, and retrieve the corresponding signal control command according to the phase conflict matrix. Then, execute the blocking logic by identifying whether there is a passage conflict in the generated signal scheme.

[0038] Preferably, step 102 includes: step 1021, calculating the spatial connection structure between intersections based on latitude and longitude coordinates; step 1022, determining the vector orientation of the sensor coverage area based on the device installation angle; step 1023, comparing the vector orientation with a preset geomagnetic azimuth reference, and converting the sensor viewpoint into the road entrance direction; step 1024, analyzing the overlapping areas of traffic trajectories in the same intersection's road entrance directions, and marking the corresponding phase group as a conflict state in the phase conflict matrix when there are spatial conflict points in the traffic trajectories.

[0039] Preferably, in step 104, when the graph attention network aggregates the features of neighboring nodes, it multiplies the online state masking vector with the corresponding attention weights element by element, so that the feature update weight of the disconnected node on the feature of the current intersection control node is 0.

[0040] Preferably, in step 106, the real-time priority score for each traffic direction is determined. Follow these rules: ,in, This represents the real-time queue length for the corresponding direction of travel. This is a global congestion adaptive scaling factor. As a weight for cumulative waiting time, For trunk line coordination gain, This is a red light timeout penalty triggered by the duration of the red light.

[0041] Preferably, before step 106, the method further includes: step 1061, collecting historical traffic data at the intersection and generating a prediction result containing traffic trend prediction values ​​using a time series prediction model; step 1062, correcting the signal timing parameters based on the traffic trend prediction values, wherein the time span of the traffic trend prediction values ​​is 15 minutes.

[0042] Preferably, in step 103, the online status masking vector is dynamically updated based on the real-time monitored heartbeat signal. When the heartbeat signal interruption duration exceeds a preset threshold, topological isolation of the corresponding neighboring nodes in the traffic network topology map is automatically triggered.

[0043] Preferably, in step 106, the global congestion adaptive scaling factor is positively correlated with the average queuing pressure, establishing a negative feedback adjustment mechanism between local pressure and global equilibrium. When the average queuing pressure is greater than the preset saturation threshold, the green light gain of the overloaded inlet lane is suppressed by increasing the global congestion adaptive scaling factor, inducing distributed decision-making to suppress local greed and solving the synchronous oscillation problem in large-scale distributed control.

[0044] Preferably, in step 106, the trunk line coordination gain is determined based on the position weight of the current intersection in the preset green wave band. By increasing the priority scores of the upstream overflow direction and the downstream absorption direction, the continuity of trunk line traffic flow is maintained.

[0045] Preferably, step 105 includes: step 1051, calculating the instantaneous queue length of each controlled intersection in the traffic network; step 1052, calculating the average queue saturation of the entire network, and converting the average queue saturation of the entire network into a dimensionless parameter characterizing the global congestion level, as a control variable fed back to the control nodes of each intersection; in step 107, by comparing the control logic of each traffic direction through the phase conflict matrix, when a traffic conflict is identified in the generated signal timing scheme, the green light path driving signal of the conflict direction is cut off through the blocking logic.

[0046] A dynamic topology-sensing multi-agent cooperative scheduling device for intelligent transportation systems includes: a memory for storing computer program instructions; and a processor for executing the computer program instructions to implement the steps of a dynamic topology-sensing multi-agent cooperative scheduling method for intelligent transportation systems.

[0047] Example 1: This example combines Figures 1 to 2 This document describes a dynamic topology-sensing multi-agent cooperative scheduling method and device for intelligent transportation systems, such as... Figure 1As shown, the process first proceeds to step 101 to obtain node coordinates and angles, specifically acquiring the latitude and longitude coordinates and equipment installation angles uploaded by the control nodes at each intersection. Then, it moves to step 102 to determine intersection relationships and generate a conflict matrix. This step determines the entrance direction based on the equipment installation angle, compares the traffic trajectory, and generates a phase conflict matrix by determining the geometric intersection relationship. During this process, a phase conflict matrix is ​​generated as a fixed output of spatial intersection parameters. Next, step 103 is executed to construct a traffic network topology graph, using control nodes as graph nodes and road connections as directed edges to build the network. Online state masking vectors are maintained for neighboring nodes, and online state masking vectors are output as identifiers of the connectivity status of neighboring nodes. Finally, step 104 is executed to update the scheduling status under the dynamic topology, specifically by monitoring heartbeat signals; if a connection fails to connect for a timeout, the corresponding logical bit is set to zero, and aggregation calculations are performed. The system first masks the characteristic parameters of the neighboring node, then proceeds to step 105 to determine the global congestion adaptive scaling factor, determine the average queuing pressure of the traffic network, and calculate the global congestion adaptive scaling factor based on the average queuing pressure. This yields the global congestion adaptive scaling factor used to transform the average queuing pressure of the network. Simultaneously, the system receives scheduling calculation feature inputs including queue length data, cumulative waiting time weight, trunk line coordination gain, and red light timeout penalty. Combining these with the aforementioned scaling factor, the system proceeds to step 106 to calculate the real-time priority score for each traffic direction. Here, multi-dimensional data features are integrated, and the queue length data is weighted using the global congestion adaptive scaling factor to calculate the score. Finally, the system reaches step 107 to retrieve the corresponding signal control command, selects the highest score and non-conflicting accompanying direction based on the real-time priority score, and retrieves the signal command according to the phase conflict matrix.

[0048] like Figure 2 As shown in the diagram, all branches in the structure uniformly point to the core objective of dynamic topology perception and multi-agent collaborative scheduling. In the safety red line dimension of the logic control layer, the system sequentially executes the control logic of collecting the coordinates of the intersection control nodes, converting the equipment installation angle into the entrance direction, and determining the geometric intersection relationship to generate the matrix. In the dimension of the dynamic graph topology maintenance mechanism, the system achieves state updates by automatically triggering the topology isolation of neighboring nodes, maintaining the online state masking vector, and executing the masking of graph attention network feature parameters. In the dimension of adaptive feedback of global congestion, the system accurately determines the global congestion adaptive scaling factor by statistically analyzing the average queuing pressure of the entire network. In the dimension of priority evaluation of the traffic direction, the system comprehensively performs weighted processing of queuing length data, combines trunk line coordination gain and waiting weight, superimposes red light timeout penalty term calculation, and finally cuts off the conflict driving signal through interlocking logic to ensure the safety of system operation.

[0049] Example 2: During the intersection initialization phase, the electronic compass module integrated into the intersection control node is used to collect the horizontal azimuth angle relative to the magnetic north pole in real time. This invention automatically generates a phase conflict matrix. Its core lies in establishing a nonlinear mapping from the magnetic north physical coordinate system to the road topology logical coordinate system. Specifically, the system first uses the electronic compass to collect the installation azimuth angle θinst of the device and aligns it with the preset road azimuth angle θroad in the high-precision map. It automatically determines the heading of the controlled sensor viewpoint of the node. Subsequently, the algorithm simulates the virtual motion trajectory of different traffic directions according to the geometric topology of the intersection entrance lane and performs vector cross product operation to determine the set of intersection points of different trajectory lines in two-dimensional space. If the envelope lines of two trajectories intersect in space, they are determined to have spatial conflict points. This perception-modeling design eliminates the logical redundancy or security loopholes caused by manual input of phase tables in traditional traffic control systems. Especially at non-standard shaped intersections, the system can automatically lock phase combinations with potential conflicts based on the physical angles perceived by the sensors, directly transforming the interference constraints of physical space into the safety red line of the logical layer, which has significant substantive features.

[0050] During the real-time scheduling and operation phase of the system, the system statistically analyzes the instantaneous queue length at each controlled intersection within the traffic network. By calculating the average ratio of the network-wide average queue length to parking space capacity, the average queue saturation of the entire network is obtained. avg The system uses a linear normalized mapping formula: λ SO =1+α•(Saturation _avg / Saturation limit This is transformed into a dimensionless global congestion adaptive scaling factor λ. SO The sensitivity coefficient α is used to adjust the global weights, and the saturation threshold is... limit The value is set to 0.9. When calculating the real-time priority score S for each traffic direction, the system performs normalization processing on the real-time queue length P, the cumulative waiting time weight W, the trunk line coordination gain G, and the red light timeout penalty T. For example, the queue length P is scaled according to the ratio of the number of vehicles to the capacity of the entrance lane, so that all input variables are converted into dimensionless values ​​of the same magnitude for weighted calculation. When the queuing pressure of the entire network increases, the scaling factor λ is increased. SO When the signal is increased, the system accelerates the relative change rate of each phase score by increasing the queuing weight, prompting the current phase to reach the switching threshold point in advance and shortening the green light dwell time of a single phase. This macroscopically compresses the total cycle time of the signal machine. When a traffic conflict is detected in the signal timing scheme, the system uses interlocking logic to cut off the green light path driving signal in the conflict direction, ensuring the safety of coordinated scheduling in a dynamic topology environment.

[0051] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0052] 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. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A dynamic topology-sensing multi-agent cooperative scheduling method for intelligent transportation systems, characterized in that, Includes the following steps: Step 101: Obtain the latitude and longitude coordinates and equipment installation angles uploaded by the control nodes at each intersection; Step 102: Based on the equipment installation angle, the equipment orientation is converted into the road entrance direction by physical orientation comparison. The equipment installation orientation angle is vector-aligned with the road orientation angle in the high-precision map to simulate the virtual movement trajectory of different traffic directions. The vector cross product operation is performed to determine the set of intersection points of different trajectory lines in two-dimensional space. Based on this, a phase conflict matrix is ​​generated as the security constraint boundary of the logic layer. Step 103: Using the control nodes at each intersection as graph nodes and the road connection relationships as directed edges, construct a traffic network topology graph and maintain online state masking vectors for neighboring nodes. Step 104: Monitor heartbeat signals. When the disconnection time of a neighboring node exceeds a preset threshold, set the corresponding bit in the online state masking vector to zero. In the normalization layer of the graph attention network aggregation calculation, the online state masking vector is converted into a bias term and added to the attention score to force the attention weight of the disconnected node to approach zero, so as to update the scheduling state under the dynamic topology. Step 105: Determine the average queuing pressure of the traffic network and transform the physical pressure characteristics of the entire network into control variables. Based on this, calculate the global congestion adaptive scaling factor according to the average queuing pressure. Step 106: Combine the historical traffic flow trend prediction of the intersection to correct the signal timing parameters, obtain queue length data, cumulative waiting time weight, trunk line coordination gain and red light timeout penalty, use the global congestion adaptive scaling factor to weight the queue length data, and calculate the real-time priority score of each traffic direction. Step 107: Based on the real-time priority score, select the passage direction with the highest score and its non-conflicting accompanying direction, and retrieve the corresponding signal control command according to the phase conflict matrix. Then, execute the blocking logic by identifying whether there is a passage conflict in the generated signal scheme.

2. The dynamic topology sensing multi-agent cooperative scheduling method for intelligent transportation systems according to claim 1, characterized in that, Step 102 includes: Step 1021, calculating the spatial connection structure between intersections based on latitude and longitude coordinates; Step 1022, determining the vector orientation of the sensor coverage area based on the equipment installation angle; Step 1023, comparing the vector orientation with the preset geomagnetic azimuth reference, and converting the sensor viewpoint into the road entrance direction; Step 1024, analyzing the overlapping areas of traffic trajectories in the same intersection's road entrance directions, and marking the corresponding phase group as a conflict state in the phase conflict matrix when there are spatial conflict points in the traffic trajectories.

3. The dynamic topology sensing multi-agent cooperative scheduling method for intelligent transportation systems according to claim 1, characterized in that, In step 104, when aggregating the features of neighboring nodes, the graph attention network multiplies the online state masking vector with the corresponding attention weights element by element, so that the feature update weight of the lost node for the current intersection control node is 0.

4. The dynamic topology sensing multi-agent cooperative scheduling method for intelligent transportation systems according to claim 1, characterized in that, In step 106, the real-time priority score for each traffic direction is determined. Follow these rules: ,in, This represents the real-time queue length for the corresponding direction of travel. This is a global congestion adaptive scaling factor. As a weight for cumulative waiting time, For trunk line coordination gain, This is a red light timeout penalty triggered by the duration of the red light.

5. The dynamic topology sensing multi-agent cooperative scheduling method for intelligent transportation systems according to claim 1, characterized in that, Before step 106, the following steps are also included: Step 1061, collecting historical traffic data at the intersection and using a time series prediction model to generate a prediction result containing traffic trend prediction values; Step 1062, correcting the signal timing parameters based on the traffic trend prediction values, with the time span of the traffic trend prediction values ​​being 15 minutes.

6. The dynamic topology sensing multi-agent cooperative scheduling method for intelligent transportation systems according to claim 1, characterized in that, In step 103, the online status masking vector is dynamically updated based on the real-time monitored heartbeat signal. When the heartbeat signal interruption duration exceeds a preset threshold, topological isolation of the corresponding neighboring nodes in the traffic network topology map is automatically triggered.

7. The dynamic topology sensing multi-agent cooperative scheduling method for intelligent transportation systems according to claim 1, characterized in that, In step 106, the global congestion adaptive scaling factor is positively correlated with the average queuing pressure, establishing a negative feedback adjustment mechanism between local pressure and global equilibrium. When the average queuing pressure is greater than the preset saturation threshold, the green light gain of the overloaded inlet lane is suppressed by increasing the global congestion adaptive scaling factor, inducing distributed decision-making to suppress local greed and solving the synchronous oscillation problem in large-scale distributed control.

8. The dynamic topology sensing multi-agent cooperative scheduling method for intelligent transportation systems according to claim 1, characterized in that, In step 106, the trunk line coordination gain is determined based on the position weight of the current intersection in the preset green wave band. By increasing the priority scores of the upstream overflow direction and the downstream absorption direction, the continuity of trunk line traffic flow is maintained.

9. A dynamic topology sensing multi-agent cooperative scheduling method for intelligent transportation systems according to claim 1, characterized in that, Step 105 includes: Step 1051, calculating the instantaneous queue length of each controlled intersection in the traffic network; Step 1052, calculating the average queue saturation of the entire network, and converting the average queue saturation of the entire network into a dimensionless parameter characterizing the global congestion level, which is used as a control variable fed back to the control nodes of each intersection; In Step 107, by comparing the control logic of each traffic direction through the phase conflict matrix, when a traffic conflict is detected in the generated signal timing scheme, the green light path driving signal of the conflicting direction is cut off through the blocking logic.

10. A dynamic topology sensing multi-agent cooperative scheduling device for intelligent transportation systems, characterized in that, include: Memory is used to store computer program instructions; A processor is used to execute computer program instructions to implement the steps of a dynamic topology-sensing multi-agent cooperative scheduling method for intelligent transportation systems as described in claim 1.

Images