College entrepreneurship simulation traffic control method and system

Abstract

The invention provides a college entrepreneurship simulation traffic control method and system, and relates to the technical field of college entrepreneurship simulation traffic control. And efficient control of a complex traffic scene is realized through multi-dimensional data fusion and dynamic strategy optimization. The system generates a three-dimensional state tensor based on a traffic perception base, extracts a cross-scene element strategy kernel, and generates a real-time control strategy in combination with a dynamic weight matrix driven by a traffic flow state difference; the double-ring feedback module generates a disturbance gradient through virtual adversarial training of a first ring, and a second ring matches and calibrates a virtual-real strategy based on optimal transmission, so that generalization is improved; in an unforeseen scene, the perception crack repair module deconstructs a strategy unit, and rapid adaptation is realized through traffic flow conservation verification and dynamic topology recombination. In a complex scene, the phase conflict rate of the system is reduced, the traffic efficiency is improved, the strategy iteration period is shortened, and an efficient intelligent control strategy is provided for complex scenes such as college gardens.

Description

Technical Field

[0001] The present invention relates to the technical field of university entrepreneurship simulation traffic control, and in particular to a university entrepreneurship simulation traffic control method and system. Background Art

[0002] In the field of simulated traffic control, which is a field of university entrepreneurship, early research focused on single-dimensional data feedback, such as adjusting signal cycles through traffic flow detection or using historical data to predict short-term traffic flow changes. As technology evolves, some researchers have attempted to introduce machine learning models, but such methods are still limited to static feature fusion and cannot effectively capture the dynamic interactions between traffic participants. While existing virtual-reality training frameworks can achieve basic strategy optimization, they are limited by one-way feedback mechanisms and static scene adaptation rules, leading to frequent phase conflicts and response lags in practical applications. Summary of the Invention

[0003] To achieve the above objectives, the present invention is implemented through the following technical solutions: a university entrepreneurship simulation traffic control system, comprising: The traffic perception base module encodes the dynamic relationships between traffic participants, signal control timing constraints, and environmental mutation factors into a three-dimensional state tensor through multi-dimensional feature field fusion; The policy core dynamic reorganization module performs tensor contraction fusion by extracting the meta-policy core from the cross-scenario policy table and the dynamic weight matrix generated in real time; A dual-loop feedback optimization module: the first loop generates adversarial perturbation training by inverting the strategy trajectory in a virtual environment. The second loop optimally matches the actual drive test data with the virtual strategy table and dynamically adjusts the strength of virtual-real knowledge transfer through strategy entropy. The crack-repair module, upon detecting an unseen scenario pattern, deconstructs the current strategy into basic operational units, reorganizes and filters them through traffic flow conservation verification, and then injects the filtered operational units into the meta-strategy table to form a memory-enhanced strategy growth system. The control execution module converts the operation unit into phase control instructions and feeds back the execution data to the update of the three-dimensional state tensor.

[0004] As the foundation of the entire system, the construction of the traffic perception base begins with the fusion of multi-dimensional feature fields; the dynamic relationships between traffic participants are collected, covering vehicle trajectories, pedestrian movement characteristics and non-motor vehicle behavior data; the interaction matrix between nodes is constructed using spatiotemporal graph convolution with traffic participants as nodes. For example, the relative speed and distance of adjacent vehicles and other information are used as edge weights, and a preliminary representation of the dynamic relationship field is obtained through graph convolution operations; at the same time, the intersection signal control parameters are extracted, including phase timing schemes, cycle timing and transition state conflict points; the timing dependence of phase switching is quantified into a set of sequence data and compressed into a timing control field along the time dimension; for example, the duration and switching order of each phase within a cycle are set to form a vector representation of the timing control field.

[0005] The integration of environmental mutation factors is achieved through multimodal fusion. Real-time weather, emergencies, and abnormal road conditions are quantified into probability distributions. Historical road scenarios are combined to generate environmental disturbance vectors, which are mapped along the third dimension into a mutation factor field. For example, the degree of road slipperiness caused by heavy rain is quantified as a probability value, which serves as a dimension of the mutation factor field. During the fusion process, the law of conservation of traffic flow is used to construct a nonlinear fusion operator. The dynamic relationship field and the mutation factor field are spatially associated after feature alignment. The timing control field is adjusted with phase synchronization constraints and then subjected to a tensor product operation with the dynamic interaction matrix. Finally, a three-dimensional state tensor is generated through physical constraint-driven feature recombination. Its dimensions are the dynamic relationship field, the timing control field, and the mutation factor field. The size of each dimension is determined by the complexity of the actual traffic scenario. For example, the dynamic relationship field is sized to [10×10] to represent the interaction between 10 traffic participants, the timing control field is sized to [1×20] to represent the signal control state at 20 moments in a cycle, and the mutation factor field is sized to [5×1] to represent the probabilities of five environmental mutation scenarios.

[0006] The core of the dynamic reorganization mechanism of the policy core lies in the tensor contraction fusion of the meta-policy core extracted from the cross-scenario policy table and the dynamic weight matrix generated in real time. The meta-policy core is extracted from historical traffic scenario data and covers effective control strategies in different scenarios. For example, representative policy cores are extracted from multi-scenario data such as morning rush hour, evening rush hour and special event scenarios through cluster analysis and other methods.

[0007] The generation of the dynamic weight matrix is ​​constrained by a differential equation driven by the traffic flow state difference. First, features are extracted from the three-dimensional state tensor in real time to construct a state difference feature vector that includes the flow balance deviation, density fluctuation variance, and velocity field divergence. For example, the flow balance deviation is expressed as: , where N is the number of traffic sections, is the flow rate of the i-th road section, is the average flow; the characteristic vector is input into the differential equation constructed by the traffic flow conservation law, where the differential equation is: , where W is the dynamic weight matrix, k is a constant, and δ is the state difference eigenvector; the update amount of the weight basis matrix is ​​calculated in real time through gradient evolution driven by the state difference, and the coefficient matrix of the differential equation is dynamically adjusted by the phase synchronization constraint of the timing control field in the three-dimensional state tensor to ensure that the total complexity of the weight matrix remains constant; the updated weight basis matrix is ​​tensor-contracted with the scenario adaptation coefficient of real-time traffic mode recognition; the scenario adaptation coefficient is jointly calibrated through the perturbation probability distribution of the mutation factor field and the adversarial error generated by inversion of the cross-scenario strategy table, thereby obtaining the final fusion strategy.

[0008] The dual-loop feedback evolution mechanism consists of the first and second loops. The first loop reconstructs the dynamic relationship field of traffic participants and the signal control timing field based on the three-dimensional state tensor in a virtual environment; generates adversarial perturbation sequences including phase switching conflicts, sudden congestion and abnormal behaviors through strategy trajectory inversion; for example, simulates a local congestion scenario caused by a sudden illegal lane change of a vehicle to generate the corresponding perturbation sequence; uses the spatiotemporal convolution kernel to inject multi-scale noise perturbations during the dynamic reorganization of the strategy kernel to form an adversarial training gradient with traffic flow conservation characteristics.

[0009] The second loop synchronously collects the node interaction intensity distribution of the dynamic relationship field and the execution error vector of the signal control timing field in the actual road test, and maps them into the probability distribution of the source domain and the target domain in the optimal transmission; the coupling matrix between the cross-scenario strategy table and the actual road test data is calculated by weighting the policy entropy, where the policy entropy dynamically adjusts the bandwidth threshold of the virtual-real knowledge migration according to the disturbance probability of the mutation factor field; the first and second loops realize bidirectional coupling through the physical constraints of the policy kernel. The first loop embeds the policy disturbance gradient generated by adversarial training into the optimal transmission distance metric space of the second loop, and the second loop back-propagates the phase synchronization constraint conditions in the actual scenario to the virtual strategy trajectory inversion process to realize the calibration of the virtual and real scenarios.

[0010] When an unprecedented scenario pattern is detected, the current control strategy is deconstructed into basic operation units along the physical constraint dimension of the three-dimensional state tensor, including the node interaction parameters of the dynamic relationship field, the phase synchronization function of the timing control field, and the disturbance response threshold of the mutation factor field; the operation units are screened through traffic flow conservation verification, and a conservation verification matrix of flow balance deviation and density fluctuation variance is constructed. The execution error of each operation unit under the disturbance probability of the mutation factor field is calculated, and effective units that meet the traffic flow conservation law and have a phase synchronization error below the set threshold are screened out; for example, for an unprecedented traffic accident scenario, the current strategy is deconstructed to screen out operation units that can still effectively maintain traffic flow conservation in this scenario.

[0011] When the verified units are injected into the meta-strategy table, reorganization optimization is performed based on the dynamic cognitive topology; according to the periodic timing constraints of the timing control field in the three-dimensional state tensor and the environmental disturbance vector of the mutation factor field, the parameter sharing correlation between the operating units is established, and the coupling strength between the operating units is adjusted through the phase synchronization constraint; finally, in the strategy reconstruction process, the topological connection of the dynamic relationship field and the conflict point elimination conditions of the timing control field are enforced, and the scenario adaptability of the reorganized strategy is evaluated using the dynamic weight matrix. The optimal strategy combination that meets the traffic flow conservation boundary is selected and injected into the memory-enhanced strategy growth system to achieve self-balance between parameter sharing and scenario adaptation of the repaired strategy, where the traffic flow conservation boundary is a constraint boundary or reference standard determined based on the traffic flow conservation law.

[0012] The present invention provides a method and system for controlling university entrepreneurship simulation traffic, which has the following beneficial effects: The present invention solves the problem of missing spatiotemporal correlation of heterogeneous data through unified modeling of dynamic relationships among traffic participants, signal timing control, and environmental mutation factors, and improves the global state representation accuracy and physical consistency of control strategies in complex traffic scenarios.

[0013] The present invention realizes the gradient evolution of the strategy under dynamic disturbance and cross-scenario knowledge transfer through the first-loop virtual adversarial training and the second-loop virtual-real optimal transmission matching, thereby optimizing the adaptability and generalization performance of the control strategy in abnormal scenarios such as sudden congestion and phase conflict.

[0014] The present invention realizes the rapid screening of policy units and memory-enhanced policy growth in unprecedented scenarios through the policy deconstruction and dynamic cognitive topological reorganization of the perceived crack repair module, solves the response lag problem caused by reliance on predefined rules, and improves the policy decision-making efficiency and system fault tolerance under sudden disturbances such as complex weather and road anomalies. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is a schematic diagram of the process of the present invention. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present invention will be described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0017] The traffic perception base module generates a three-dimensional state tensor by fusing multi-dimensional feature fields. For example, the system simulates a university entrepreneurial park and deploys a multimodal sensor network at the intersection of the university entrepreneurial park, including high-precision cameras, millimeter-wave radars, and geomagnetic induction coils, to collect dynamic data of traffic participants in real time. For example, the camera captures vehicle trajectories at a rate of 30 frames per second, the millimeter-wave radar measures vehicle speed and acceleration, and the geomagnetic coil detects the position changes of non-motor vehicles. For pedestrian movement characteristics, the system extracts gait frequency and path direction through thermal imaging sensors and combines deep learning to predict their crossing intentions. After preprocessing, the raw data is input into the spatiotemporal graph convolutional network, and an interaction matrix is ​​constructed with traffic participants as nodes. For example, when two electric bicycles pass through the intersection in parallel with a relative speed difference of 3 meters per second and a distance of 1.5 meters, the system encodes the dynamic relationship between the two as the edge weights of the interaction matrix and extracts the spatial correlation features of the dynamic relationship field through graph convolution operations. The signal control parameters collected simultaneously include phase timing schemes, green light duration, and transition conflict points. For example, the original duration of the north-south green light at the intersection is 30 seconds, and the system uses real-time traffic density to predict the duration of the green light. It is dynamically extended to 45 seconds, and the phase switching sequence is compressed into 20 time slices of the time control field along the time dimension; the environmental mutation factors are integrated through multimodal data. For example, in heavy rain, the meteorological sensor detects a road slip coefficient of 0.7. Historical accident data shows that the rear-end collision probability increases by 40% under this condition. The system quantifies it as a probability distribution of the mutation factor field; through the nonlinear fusion operator constructed by the traffic flow conservation law, the dynamic relationship field and the mutation factor field are spatially associated through the feature alignment module, such as the vehicle deceleration behavior and the braking distance on the slippery road surface. Correct parameter binding; after the timing control field is adjusted through phase synchronization constraints, a tensor product operation is performed with the dynamic interaction matrix to ultimately generate a three-dimensional state tensor; the dimension of this tensor is dynamically adjusted according to the complexity of the scenario. For example, the dynamic relationship field is set to 10×10 to represent the interaction intensity of 10 traffic participants, the timing control field is 1×20 to represent the 20 time slice states of a signal cycle, and the mutation factor field is 5×1 to represent the probability of five types of environmental disturbances. Using this method, the system reduces the phase conflict rate from 15% to 9% in heavy rain scenarios and shortens the response time to 0.8 seconds.

[0018] The policy core dynamic reorganization module extracts meta-policy cores from the cross-scenario policy table and performs tensor contraction fusion with the real-time generated dynamic weight matrix. The system extracts multi-scenario policy cores from historical traffic scenarios. For example, the priority release policy for main roads during the morning rush hour includes rules such as a 20% extension of the green light and early activation of the left-turn phase. The pedestrian crossing policy during the evening rush hour includes parameters such as a 10-second increase in the green light cycle and the insertion of a pedestrian-only phase. The dynamic weight matrix is ​​generated based on a differential equation model driven by traffic flow state differences, and the flow balance deviation, density fluctuation variance, and velocity field divergence are calculated in real time. The flow balance deviation is calculated as follows: , where N is the number of traffic sections, is the flow rate of the i-th road section, is the average flow rate; when it is detected that the flow deviation of a lane exceeds the set threshold, the system constructs the state difference feature vector and inputs it into the differential equation: , where W is the dynamic weight matrix, k is the conservation coefficient, and δ is the state difference vector. In heavy rain scenarios, the probability of a slippery road in the mutation factor field is 0.8. The system adjusts the weight matrix through differential equations and performs tensor contraction on the vehicle deceleration parameters and the signal delay strategy. For example, the braking distance of a vehicle on a slippery road needs to be increased by 30%. Based on this, the system dynamically reduces the frequency of green light switching and extends the yellow light duration to 5 seconds to avoid rear-end collisions caused by sudden braking. In sudden traffic accident scenarios, the congestion diffusion speed is fed back to the state difference vector in real time, and the dynamic weight matrix is ​​updated every 10 seconds to optimize the signal timing plan to reduce secondary congestion. Actual tests show that this mechanism reduces the rear-end collision rate in heavy rain scenarios by 35%, and the signal response error is controlled within ±1 second.

[0019] The dual-loop feedback optimization module realizes strategy optimization by matching the first-loop virtual adversarial training with the second-loop actual road test data. The first loop reconstructs the traffic scene in the virtual environment, simulates abnormal events such as illegal lane changes of vehicles and pedestrians running red lights, and generates adversarial perturbation sequences. For example, in the virtual environment, an ambulance is simulated to forcibly change lanes to the left turn lane during the north-south green light, causing the vehicle behind to brake suddenly and trigger local congestion. The system generates adversarial perturbation sequences through strategy trajectory inversion, and uses spatiotemporal convolution kernels to inject multi-scale noise into the strategy kernel. For example, in virtual training, the probability of random vehicle lane changes is increased to 20%, and a perturbation gradient containing phase conflict and congestion diffusion is generated. The second loop simultaneously collects actual road test data, such as the interaction intensity distribution of intersection nodes and signal execution errors. In traffic accident scenarios, the second loop maps the actual phase switching delay data to the source domain distribution of optimal transmission, and the virtual policy table to the target domain distribution, and calculates the coupling matrix through policy entropy weighting. For example, when the actual phase switching delay is 2 seconds, the second loop adjusts the timing control field parameters of the virtual policy through the Sinkhorn algorithm, reducing the phase switching error in virtual training from 3 seconds to 1 second. The second loop achieves bidirectional coupling through physical constraint projection. For example, the adversarial gradient generated by the first loop is embedded in the optimal transmission distance metric of the second loop. The second loop backpropagates the phase synchronization constraints of the actual scenario to the virtual policy core reorganization process. Through this mechanism, the system's average traffic efficiency during peak hours is increased from 800 vehicles per hour to 1,000 vehicles, and phase switching conflicts are reduced by 60%.

[0020] The crack-sensing repair module realizes rapid strategy adaptation in unforeseen scenarios. For example, if a group of electric bicycles suddenly run a red light and cause chaos at the intersection, the system deconstructs the current control strategy into basic operation units along the physical constraint dimension of the three-dimensional state tensor, including the node interaction parameters of the dynamic relationship field, the phase synchronization function of the timing control field and the disturbance response threshold of the mutation factor field. By constructing a conservation verification matrix of flow balance deviation and density fluctuation variance, the effective units for maintaining lane flow balance are screened out. Specifically, the system calculates the execution error of each operation unit under the disturbance probability of the mutation factor field. For example, in the scenario of electric bicycles running a red light, The phase synchronization function must meet the constraint of a 3-second delay in the green light for pedestrians crossing the street, and the node interaction parameters must increase the priority of non-motorized vehicle lanes to 70%. The verified units are reorganized and optimized based on the dynamic cognitive topology. For example, the parameter sharing correlation between operation units is established according to the periodic timing constraints of the timing control field, and the coupling strength is adjusted through phase synchronization constraints. In the scenario where lanes are closed due to road construction, the system selects operation units that dynamically adjust the direction of variable lanes. The reorganized strategy is deployed within 10 seconds, reducing the traffic delay time on the construction section from 15 minutes to 7 minutes, and improving traffic efficiency by 53%.

[0021] The control execution module converts the optimization strategy into executable phase control instructions. For example, during the evening rush hour in university entrepreneurial parks, the system adjusts the signal timing in real time based on the three-dimensional state tensor: the green light on the north-south main road is extended to 50 seconds, the green light on the secondary direction is shortened to 20 seconds, and a 15-second pedestrian-only phase is inserted. The execution data updates the three-dimensional state tensor through the feedback loop. For example, if it detects that the actual number of vehicles passing through the north-south direction is 20% lower than expected, the system automatically reduces the green light duration of the next cycle to 45 seconds and increases the left-turn phase duration by 5 seconds. In scenarios where large-scale events cause an instantaneous surge in pedestrian flow, the system extends the green light for pedestrians to cross the street to 40 seconds through dynamic reorganization strategies, reducing the pedestrian waiting time from 90 seconds to 60 seconds. The feedback mechanism shortens the strategy iteration cycle from 30 minutes in the traditional model to 5 minutes, improving optimization efficiency by 83%.

[0022] By deploying a multimodal sensor network and configuring edge computing nodes; using spatiotemporal graph convolution to construct a dynamic relationship field, and combining historical signal control data to generate a timing control field; extracting meta-strategy cores from historical scenarios and constructing a cross-scenario strategy table; injecting adversarial disturbances into a virtual environment to generate training data, synchronously collecting actual road test data and calibrating the strategy through the optimal transmission algorithm; deploying a perception crack repair algorithm to achieve rapid adaptation to unseen scenarios; and finally continuously optimizing the signal control strategy through a feedback mechanism; through the above technical details and implementation path, the present invention provides a high-precision and highly robust solution for university entrepreneurial parks and similar complex traffic scenarios.

[0023] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A traffic control method for entrepreneurship simulation in colleges and universities, characterized in that: The following steps are involved: Build a traffic perception base, and encode the dynamic relationships between traffic participants, signal control timing constraints, and environmental mutation factors into a three-dimensional state tensor through multi-dimensional feature field fusion; A dynamic reorganization mechanism for policy cores is designed. This involves performing tensor contraction fusion of the meta-policy core extracted from the cross-scenario policy table and the dynamic weight matrix generated in real time. The generation of the dynamic weight matrix is ​​constrained by differential equations driven by traffic flow state differences, and the fusion process maintains the total complexity of the policy. A dual-loop feedback evolution mechanism is established. The first loop generates adversarial perturbation training by inverting the strategy trajectory in the virtual environment. The second loop optimizes the transmission matching between actual drive test data and the virtual strategy table, and dynamically adjusts the strength of virtual-real knowledge transfer through strategy entropy. A traffic-aware crack repair module is deployed. When an unprecedented scenario pattern is detected, the current strategy is deconstructed into basic operation units and reorganized and screened through traffic flow conservation verification. The screened operation units are injected into the meta-strategy table to form a memory-enhanced strategy growth system. The strategy reconstruction process of the memory-enhanced strategy growth system is limited by the physical constraints of the three-dimensional state tensor, and the dynamic cognitive topology is used to achieve self-balancing of the control strategy between parameter sharing and scenario adaptation.

2. A university entrepreneurship simulation traffic control method according to claim 1, characterized in that: The three-dimensional state tensor is based on the spatiotemporal encoding of multidimensional feature fields and the fusion of physical constraints. It collects the dynamic relationships between traffic participants, including vehicle trajectories, pedestrian movement characteristics, and non-motor vehicle behavior data, and constructs an interaction matrix between nodes using spatiotemporal graph convolution. This matrix is ​​encoded into a dynamic relationship field along the first dimension. The intersection signal control parameters, including phase timing schemes, cycle timing, and transition state conflict points, are simultaneously extracted to capture the temporal dependencies of phase switching and compress them into a temporal control field along the second dimension. At the same time, environmental mutation factors are integrated. Real-time weather, emergencies, and road anomalies are quantified into probability distributions through multimodal fusion. Environmental disturbance vectors are generated by combining historical road scenes and mapped into a mutation factor field along the third dimension. A nonlinear fusion operator is then constructed based on the traffic flow conservation law to couple the three sets of heterogeneous fields across dimensions. The dynamic relationship field and the mutation factor field are spatially associated through a feature alignment module. The temporal control field is adjusted with phase synchronization constraints and then tensor-producted with the dynamic interaction matrix. Finally, a three-dimensional state tensor is generated through feature recombination driven by physical constraints.

3. The method for controlling university entrepreneurship simulation traffic according to claim 1, characterized in that: The dynamic weight matrix is ​​implemented through a differential constraint mechanism driven by traffic flow state differences. Features are extracted from the three-dimensional state tensor in real time, and a state difference feature vector containing flow balance deviation, density fluctuation variance and velocity field divergence is constructed and input into the differential equation model constructed by the traffic flow conservation law. The updated amount of the weight basis matrix is ​​calculated in real time through the gradient evolution algorithm driven by the state difference, where the coefficient matrix of the differential equation is dynamically adjusted by the phase synchronization constraint of the timing control field in the three-dimensional state tensor, and the total complexity of the weight matrix is ​​ensured to remain constant through time step control; then the updated weight basis matrix is ​​tensor-contracted with the scenario adaptation coefficient of real-time traffic mode recognition. The scenario adaptation coefficient is jointly calibrated by the perturbation probability distribution of the mutation factor field and the adversarial error generated by inversion of the cross-scenario strategy table.

4. The method for controlling university entrepreneurship simulation traffic according to claim 1, characterized in that: The dual-loop feedback evolution mechanism consists of the first and second loops. The first loop reconstructs the dynamic relationship field of traffic participants and the signal control timing field based on the three-dimensional state tensor in a virtual environment, generates adversarial perturbation sequences including phase switching conflicts, sudden congestion and abnormal behaviors through strategy trajectory inversion, and uses the spatiotemporal convolution kernel to inject multi-scale noise perturbations during the dynamic reorganization of the strategy kernel to form an adversarial training gradient with traffic flow conservation characteristics.

5. The method for controlling university entrepreneurship simulation traffic according to claim 4, characterized in that: The second loop synchronously collects the node interaction intensity distribution of the dynamic relationship field and the execution error vector of the signal control timing field in the actual road test, maps them into the probability distribution of the source domain and the target domain in the optimal transmission, and uses the policy entropy weighted calculation to calculate the coupling matrix between the cross-scenario policy table and the actual road test data, where the policy entropy dynamically adjusts the bandwidth threshold of the virtual-real knowledge migration according to the disturbance probability of the mutation factor field; the second loop realizes bidirectional coupling through the physical constraint projection of the policy kernel. The first loop embeds the policy disturbance gradient generated by adversarial training into the optimal transmission distance metric space of the second loop. The second loop backpropagates the phase synchronization constraint conditions in the actual scenario to the virtual policy trajectory inversion process to form a calibration of the virtual-real scenario.

6. The method for controlling university entrepreneurship simulation traffic according to claim 1, characterized in that: The perception crack repair module includes strategy deconstruction and dynamic cognitive topology. When an unprecedented scene pattern is detected, the current control strategy is deconstructed into basic operation units along the physical constraint dimension of the three-dimensional state tensor, including the node interaction parameters of the dynamic relationship field, the phase synchronization function of the timing control field, and the disturbance response threshold of the mutation factor field; the operation units are screened through traffic flow conservation verification, and the node interaction intensity distribution of the dynamic relationship field in the three-dimensional state tensor and the phase switching timing difference of the timing control field are used to construct a conservation verification matrix of flow balance deviation and density fluctuation variance, calculate the execution error of each operation unit under the disturbance probability of the mutation factor field, and screen out the operation units that meet the traffic flow conservation law and have a phase synchronization error lower than the set value. The effective unit with a fixed threshold is selected; when the verified operation unit is injected into the meta-strategy table, it is reorganized and optimized based on the dynamic cognitive topology. According to the periodic timing constraints of the timing control field in the three-dimensional state tensor and the environmental disturbance vector of the mutation factor field, the parameter sharing correlation between the operation units is established, and the coupling strength between the operation units is adjusted through the phase synchronization constraint; finally, in the process of strategy reconstruction, the topological connection of the dynamic relationship field and the conflict point elimination conditions of the timing control field are forced to be constrained, and the scenario adaptability of the reorganized strategy is evaluated using the dynamic weight matrix. The optimal strategy combination that meets the traffic flow conservation boundary is selected and injected into the memory-enhanced strategy growth system to achieve self-balance between parameter sharing and scenario adaptation of the repaired strategy.

7. A system for a university entrepreneurship simulation traffic control method according to any one of claims 1 to 6, characterized in that: include: The traffic perception base module encodes the dynamic relationships between traffic participants, signal control timing constraints, and environmental mutation factors into a three-dimensional state tensor through multi-dimensional feature field fusion; The policy core dynamic reorganization module performs tensor contraction fusion by extracting the meta-policy core from the cross-scenario policy table and the dynamic weight matrix generated in real time; A dual-loop feedback optimization module: the first loop generates adversarial perturbation training by inverting the strategy trajectory in a virtual environment. The second loop optimally matches the actual drive test data with the virtual strategy table and dynamically adjusts the strength of virtual-real knowledge transfer through strategy entropy. The crack-repair module, upon detecting an unseen scenario pattern, deconstructs the current strategy into basic operational units, reorganizes and filters them through traffic flow conservation verification, and then injects the filtered operational units into the meta-strategy table to form a memory-enhanced strategy growth system. The control execution module converts the operation unit into phase control instructions and feeds back the execution data to the update of the three-dimensional state tensor.

Images