Intelligent traffic adaptive regulation and control system and method based on social perception

Abstract

The invention discloses a smart traffic adaptive regulation and control system and method based on social perception, and belongs to the technical field of traffic Internet of Things. Comprising a multi-modal social perception layer module, an edge intelligent preprocessing node module, a dynamic social flow modeling engine module, a space-time social prediction engine module, an adaptive signal lamp optimization module, an Internet of Vehicles collaborative pushing module and a traffic closed-loop feedback module. A social event factor is dynamically injected through an LSTM-Transform hybrid model, and self-adaptive adjustment of space-time prediction is achieved; the accuracy and robustness of traffic prediction are remarkably improved, and the congestion risk can be accurately recognized in advance in the social event high-incidence period; and prediction misjudgment caused by neglecting social behaviors is avoided, so that the system can dynamically respond to traffic flow changes, the congestion pressure is effectively relieved, and the road traffic efficiency and the overall traffic smoothness are improved.

Description

Technical Field

[0001] This invention belongs to the field of transportation Internet of Things technology, specifically referring to a smart transportation adaptive control system and method based on social perception. Background Technology

[0002] With the acceleration of urbanization, traffic congestion has become one of the major challenges faced by cities around the world; traditional traffic management systems mainly rely on fixed traffic light schedules and limited traffic flow monitoring methods.

[0003] However, existing traffic monitoring systems based on the Internet of Things for intelligent transportation still have certain shortcomings. Existing technologies rely on traditional sensor data and simple statistical models, ignoring the social interaction between pedestrians and vehicles. This results in traffic flow models being unable to quantify social dynamic changes, and the modeling results being out of touch with real traffic scenarios. Static prediction models based on historical traffic flow do not integrate social event data and real-time social interaction information, making it impossible to dynamically identify congestion risks during periods of high incidence of social events. The prediction process ignores the amplifying effect of social behavior on traffic flow, making the prediction results susceptible to social dynamic interference and misjudgments. The system cannot respond to changes in traffic flow in advance, causing congestion mitigation measures to lag behind. Therefore, a smart transportation adaptive control system and method based on social perception is proposed. Summary of the Invention

[0004] The purpose of this invention is to provide a socially-aware intelligent transportation adaptive control system and method to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a smart traffic adaptive control system and method based on social perception, including a multimodal social perception layer module, a social event real-time perception and fusion module, an edge intelligent preprocessing node module, a dynamic social flow modeling engine module, a spatiotemporal social prediction engine module, an adaptive traffic light optimization module, a pedestrian safety and priority passage module, a vehicle network collaborative push module, and a traffic closed-loop feedback module.

[0006] The multimodal social perception layer module is deployed based on multimodal sensors and collects social traffic flow data through multimodal sensor fusion;

[0007] The edge intelligent preprocessing node module performs data preprocessing based on social traffic flow data through edge computing.

[0008] The real-time social event perception and fusion module collects, quantifies, and fuses social event data in real time based on non-sensor data sources through data capture and analysis.

[0009] The dynamic social flow modeling engine module performs social flow modeling based on the preprocessed data using a graph neural network.

[0010] The spatiotemporal social prediction engine module performs spatiotemporal predictions based on social flow modeling and historical social data, using a spatiotemporal prediction model.

[0011] The adaptive traffic light optimization module optimizes traffic lights through adaptive control based on spatiotemporal prediction results.

[0012] The pedestrian safety and priority passage module controls pedestrian safety windows and priority passage based on social flow modeling and prediction results through safety assessment and priority allocation;

[0013] The vehicle-to-everything (V2X) collaborative push module pushes information to the network based on the optimized traffic lights via V2X communication.

[0014] The traffic closed-loop feedback module performs closed-loop feedback based on the vehicle network push results and environmental data through feedback analysis.

[0015] Preferably, the multimodal social perception layer module, based on the traffic scenario, comprehensively selects and deploys a combination of multimodal sensors, including LiDAR, high-definition cameras, miniature weather stations, and voiceprint sensors; based on traffic flow distribution and key monitoring areas, it increases sensor density at key nodes through a gridded and focused deployment strategy, removes interference from special symbols using regular expression rules, integrates a spell check library to correct abnormal data, and aligns and splices features of different modal data to form a unified semantic representation.

[0016] Preferably, the real-time social event perception and fusion module is independent of the multimodal social perception layer. It captures social event information in real time through social media APIs, news data interfaces, and public event calendars. It uses natural language processing technology to analyze the event type, time, location, and scope of impact, quantifying them into structured social event factors, including event impact coefficients, duration factors, and regional coverage, forming structured data that can be input into the spatiotemporal social prediction engine.

[0017] Preferably, the edge intelligent preprocessing node module receives traffic flow data streams, establishes a circular buffer to ensure data continuity, dynamically configures the buffer size according to the computing power and network bandwidth of the edge devices, accurately matches heterogeneous data from LiDAR, high-definition cameras, micro weather stations, and voiceprint sensors in the time dimension through timestamp alignment, smooths the raw data, and effectively removes sensor noise; at the same time, based on a statistical anomaly detection mechanism, it identifies and removes outliers that exceed a reasonable range.

[0018] Preferably, the dynamic social flow modeling engine module receives preprocessed heterogeneous data, including key information such as node location coordinates, behavior type labels, real-time speed, and environmental parameters. Based on the traffic network topology, it initializes the spatiotemporal graph structure by using intersections, pedestrian trajectory points, and vehicle positions as nodes, and the movement paths of pedestrians and vehicles as edges. It then extracts node attribute features from the preprocessed data, including location coordinates, behavior type, real-time speed, density value, and social attributes, forming standardized feature vectors. Based on the relative positions, movement directions, and interaction frequencies between nodes, it calculates the weight features of the edges, achieving the following:

[0019] ,

[0020] In the formula, Representing an edge Interaction weights This represents the Euclidean distance between the pedestrian node j and the vehicle node j. This represents the behavioral factor of pedestrian node i. This refers to weather and environmental factors.

[0021] Preferably, the dynamic social flow modeling engine module loads a pre-trained spatiotemporal graph neural network model. This model optimizes the traffic scenario, runs GNN forward propagation on a graph structure, aggregates neighbor node information through multi-layer message passing, and dynamically updates the feature representation of each node. This is achieved as follows:

[0022] ,

[0023] In the formula, This represents the updated social characteristics of node i at level l+1. Let i represent the set of neighbors of node i. This represents the weight matrix of the science departments in the l-th layer. This represents the characteristics of neighbor node j at layer l. Indicates the social impact coefficient. This represents the activation function.

[0024] Preferably, the spatiotemporal social prediction engine module receives a real-time traffic density matrix and a social interaction heatmap, and associates it with historical social data stored on a cloud platform. It achieves spatiotemporal alignment of multi-source data through timestamp matching and interpolation filling, dynamically identifies key social events from historical social data, extracts event types, durations, and historical impact intensity, and encodes them into quantifiable feature vectors. These vectors are then fused with the real-time social interaction heatmap in a spatial-temporal dimension. Based on the aligned data, a multidimensional time-series input sequence is constructed.

[0025] The constructed time series is input into a pre-trained LSTM-Transformer hybrid prediction model. During the model inference phase, the weights of social event factors are dynamically adjusted based on the output of the real-time social interaction heatmap, as follows:

[0026] ,

[0027] In the formula, Represents the social impact factors at time t. A real-time social interaction heatmap representing time t. This represents the average impact of similar social events in history. Indicates adaptive weights for data quality. Indicates the environmental amplification factor. This represents real-time weather and environmental factors.

[0028] Preferably, the spatiotemporal social prediction engine module runs a model to predict traffic flow and outputs the congestion probability of each road network node, as follows:

[0029] ,

[0030] In the formula, Indicates time The predicted probability of congestion, Represents the spatiotemporal feature weight matrix. Indicates spatiotemporal characteristics, This represents the social impact coefficient; a confidence assessment is performed on the prediction results. If the confidence level of the social event factor is lower than the threshold, the average impact value of similar events in historical data is used for correction.

[0031] Preferably, the adaptive traffic light optimization module acquires a predicted congestion heatmap, social impact factors, and time windows. Based on the social impact factors, it sorts congested areas according to their social risk levels, prioritizing the allocation of traffic resources to high-risk areas and delaying optimization for low-risk areas. It also integrates real-time environmental data, injecting environmental impact coefficients into the social risk ranking, and scans the traffic light phases at adjacent intersections based on the road network topology, eliminating vehicle crossover conflicts through time-series fine-tuning.

[0032] Preferably, the pedestrian safety and priority passage module is either enhanced or independent of the adaptive traffic light optimization module, and utilizes the pedestrian heat map output by dynamic social flow modeling to explicitly incorporate strategies such as pedestrian crossing safety windows, pedestrian red light violation warning linkage, or green channels for special groups into the signal timing.

[0033] Preferably, the vehicle-to-everything (V2X) collaborative push module receives the real-time timing scheme output by the adaptive traffic light optimization module, converts it into a V2X standard message format, embeds social impact prompts and precise timestamps; based on the current traffic flow density and V2X network congestion index, it dynamically adjusts the message transmission frequency and transmission power to optimize transmission efficiency and coverage; through the collaborative construction of a multi-hop propagation network by roadside units and vehicle-mounted equipment, it prioritizes broadcasting messages to high-risk areas; it triggers message pushes at a preset time before traffic light status changes; it dynamically adjusts the push priority for different vehicle types; and it records the success rate, average latency, and vehicle response time of each push in real time to generate structured logs.

[0034] Preferably, the social flow modeling via graph neural networks includes: using an attention mechanism to calculate the interaction weights between pedestrian nodes and vehicle nodes.

[0035] Preferably, the step of pushing vehicle-to-everything (V2X) information via V2X communication includes: broadcasting personalized route guidance messages based on vehicle type and destination information.

[0036] Preferably, a computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the steps of the method as described in claim 10.

[0037] A preferred method for adaptive regulation of intelligent transportation based on social perception includes the following steps:

[0038] S1. Based on the deployment of multimodal sensors, collect social traffic flow data through multimodal sensor fusion;

[0039] S2. Based on social traffic flow data, perform data preprocessing through edge computing;

[0040] S3. Based on the preprocessed data, perform social flow modeling using graph neural networks;

[0041] S4. Based on social flow modeling and historical social data, perform spatiotemporal prediction using a spatiotemporal prediction model;

[0042] S5. Based on the spatiotemporal prediction results, optimize the traffic lights through adaptive control;

[0043] S6. Based on the optimized traffic lights, push vehicle-to-everything (V2X) communication to the network.

[0044] S7. Based on the vehicle network push results and environmental data, a closed-loop feedback is performed through feedback analysis.

[0045] Compared with the prior art, the beneficial effects of the present invention are:

[0046] 1. This invention constructs a social traffic map structure by preprocessing data, uses graph neural networks to quantify the interaction between pedestrians and vehicles, and dynamically updates the node feature representation; it integrates social behavioral factors and environmental dynamics to accurately capture the lagging impact of social interaction on traffic flow, making the social flow model more consistent with real traffic scenarios, providing high-value interaction heatmaps and impact coefficients, inputting key social dimension information into the prediction engine, and effectively improving the system's adaptability to changes in social behavior and modeling accuracy;

[0047] 2. This invention integrates social flow modeling results with historical social data and dynamically injects social event factors through an LSTM-Transformer hybrid model to achieve adaptive adjustment of spatiotemporal prediction; it significantly improves the accuracy and robustness of traffic prediction, especially in accurately identifying congestion risks in advance during periods of high incidence of social events; it avoids prediction misjudgments caused by ignoring social behavior, enabling the system to dynamically respond to changes in traffic flow, effectively alleviate congestion pressure, and improve road traffic efficiency and overall traffic flow.

[0048] 3. This invention dynamically adjusts the timing strategy based on spatiotemporal prediction results, prioritizing high-social-risk areas, integrating environmental data to optimize traffic resource allocation, and eliminating phase conflicts to achieve intelligent dynamic allocation of traffic lights. This optimizes traffic flow efficiency while ensuring pedestrian safety and the passage needs of key areas. By ranking social risk levels and using adaptive environmental weights, it reduces intersection conflicts and traffic delays, improves the response speed and sustainability of the traffic system, and makes signal control more aligned with actual social needs, enhancing overall traffic safety and efficiency. It is expected to reduce the accident rate in high-social-interaction areas and increase the average vehicle speed on main roads through predictive guidance. Attached Figure Description

[0049] Figure 1 This is a schematic diagram of the structure of the intelligent transportation adaptive control system based on social perception according to the present invention;

[0050] Figure 2 The following is the operation flow of the intelligent transportation adaptive control system based on social perception according to the present invention. Figure 1 ;

[0051] Figure 3 The following is the operation flow of the intelligent transportation adaptive control system based on social perception according to the present invention. Figure 2 ;

[0052] Figure 4 The following is the operation flow of the intelligent transportation adaptive control system based on social perception according to the present invention. Figure 3 ;

[0053] Figure 5 This is a flowchart illustrating the operation of the intelligent transportation adaptive control method based on social perception, as described in this invention. Detailed Implementation

[0054] 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.

[0055] Example

[0056] Please see Figures 1-5 As shown, the present invention provides a technical solution: a multimodal social perception layer module, a social event real-time perception and fusion module, an edge intelligent preprocessing node module, a dynamic social flow modeling engine module, a spatiotemporal social prediction engine module, an adaptive traffic light optimization module, a pedestrian safety and priority passage module, a vehicle network collaborative push module, and a traffic closed-loop feedback module.

[0057] The multimodal social perception layer module is deployed based on multimodal sensors and collects social traffic flow data through multimodal sensor fusion;

[0058] The edge intelligent preprocessing node module performs data preprocessing based on social traffic flow data through edge computing.

[0059] The real-time social event perception and fusion module collects, quantifies, and fuses social event data in real time based on non-sensor data sources through data capture and analysis.

[0060] The dynamic social flow modeling engine module performs social flow modeling based on the preprocessed data using a graph neural network.

[0061] The spatiotemporal social prediction engine module performs spatiotemporal predictions based on social flow modeling and historical social data, using a spatiotemporal prediction model.

[0062] The adaptive traffic light optimization module optimizes traffic lights through adaptive control based on spatiotemporal prediction results.

[0063] The pedestrian safety and priority passage module controls pedestrian safety windows and priority passage based on social flow modeling and prediction results through safety assessment and priority allocation;

[0064] The vehicle-to-everything (V2X) collaborative push module pushes information to the network based on the optimized traffic lights via V2X communication.

[0065] The traffic closed-loop feedback module performs closed-loop feedback based on the vehicle network push results and environmental data through feedback analysis.

[0066] In this embodiment, the multimodal social perception layer module comprehensively selects a combination of multimodal sensors, including LiDAR, high-definition cameras, miniature weather stations, and voiceprint sensors, based on the deployment of traffic scenarios. According to the traffic flow distribution and key monitoring areas, the module increases sensor density at key nodes such as intersections, school perimeters, and commercial areas through a gridded and focused deployment strategy.

[0067] By removing interference from special symbols using regular expression rules, integrating a spell check library to correct abnormal data, and aligning and splicing features of different modal data to form a unified semantic representation.

[0068] In this embodiment, the edge intelligent preprocessing node module receives traffic flow data streams, establishes a circular buffer to ensure data continuity, dynamically configures the buffer size according to the computing power and network bandwidth of the edge devices, and accurately matches the heterogeneous data from LiDAR, high-definition cameras, micro weather stations and voiceprint sensors in the time dimension through timestamp alignment.

[0069] The raw data is smoothed to effectively remove sensor noise; at the same time, a statistical anomaly detection mechanism is used to identify and remove outliers that exceed the reasonable range.

[0070] In this embodiment, the dynamic social flow modeling engine module receives preprocessed heterogeneous data, including key information such as node location coordinates, behavior type labels, real-time speed, and environmental parameters. Based on the traffic network topology, it initializes the spatiotemporal graph structure by taking intersections, pedestrian trajectory points, and vehicle positions as nodes and the movement paths of pedestrians and vehicles as edges. It then extracts node attribute features from the preprocessed data, including location coordinates, behavior type, real-time speed, density value, and social attributes, to form a standardized feature vector.

[0071] Specifically, the behavioral factor is a numerical indicator used to quantify the potential impact of traffic participants' micro-behaviors on traffic flow. It transforms raw behavioral signals captured by sensors into traffic-meaning parameters that the system can understand and calculate. The quantification process mainly includes behavior recognition and coefficient assignment.

[0072] Pedestrian behavior quantification based on visual sensors: By deploying high-definition cameras and using computer vision models such as YOLO and OpenPose, pedestrian behavior patterns are detected and identified in real time; different behavior patterns are assigned different numerical coefficients based on their potential interference with traffic flow or risk level.

[0073] Vehicle behavior quantification based on voiceprint sensors: By monitoring sound features in the environment using voiceprint sensors, the anxiety or urgency state of road traffic participants can be indirectly quantified, which can serve as a dynamically adjusted parameter for behavioral factors. Horn density: The frequency and intensity of horn blasts in a specific area per unit time; high horn density usually indicates low traffic flow, high driver anxiety, and a high risk of potential conflict. A weight calculated based on acoustic features can be added to the baseline value of the overall behavioral factor for that area. The identified specific behaviors will be bound to the spatial location information of the behavioral subject. Ultimately, the behavioral factor of pedestrian node i will be the base coefficient corresponding to its identified behavior, and may be the result of weighted fusion with other environmental factors; thus directly affecting the construction of the social flow model.

[0074] Based on the relative positions, movement directions, and interaction frequencies between nodes, the weight characteristics of the edges are calculated, as follows:

[0075] ,

[0076] In the formula, Representing an edge Interaction weights This represents the Euclidean distance between the pedestrian node j and the vehicle node j. This represents the behavioral factor of pedestrian node i. This refers to weather and environmental factors.

[0077] In this embodiment, the dynamic social flow modeling engine module loads a pre-trained spatiotemporal graph neural network model. The spatiotemporal graph neural network model is optimized for traffic scenarios and includes a multi-layer message passing mechanism to capture social interaction patterns.

[0078] Specifically, GNN forward propagation is run on the graph structure, and neighbor node information is aggregated through multi-layer message passing to dynamically update the feature representation of each node, which is implemented as follows:

[0079] ,

[0080] In the formula, This represents the updated social characteristics of node i at level l+1. Let i represent the set of neighbors of node i. This represents the weight matrix of the science departments in the l-th layer. This represents the characteristics of neighbor node j at layer l. Representing the social influence coefficient, this is a node-level measure of social interaction influence output by the dynamic social flow modeling engine module. It quantifies the intensity of social interaction between pedestrians and vehicles, and is derived through dynamic learning from a graph neural network. The calculation is based on the interaction weights between nodes and the aggregation of neighbor features. The activation function is used to generate a real-time interaction heatmap based on the updated node features. The social flow modeling results are organized into a standardized data package, which includes a real-time traffic density matrix, a social interaction heatmap, and interaction influence coefficients.

[0081] In this embodiment, the spatiotemporal social prediction engine module receives a real-time traffic density matrix and a social interaction heatmap, and associates it with historical social data stored on the cloud platform. It achieves spatiotemporal alignment of multi-source data through timestamp matching and interpolation filling, dynamically identifies key social events from historical social data, extracts event types, durations, and historical impact intensity, and encodes them into quantifiable feature vectors, which are then fused with the real-time social interaction heatmap in a spatial-temporal dimension.

[0082] Specifically, based on the aligned data, a multidimensional time-series input sequence is constructed, including:

[0083] Spatial characteristics: Road network node density distribution;

[0084] Temporal characteristics: Historical traffic flow periodicity;

[0085] Social characteristics: social event factors.

[0086] Specifically, the constructed time series is input into a pre-trained LSTM-Transformer hybrid prediction model. During the model inference phase, the weights of social event factors are dynamically adjusted based on the output of the real-time social interaction heatmap, as follows:

[0087] ,

[0088] In the formula, The social impact factor at time t refers to the global social impact metric output by the spatiotemporal social prediction engine module, used to adjust the spatiotemporal prediction model. It is dimensionless and ranges from [0,1]. A real-time social interaction heatmap representing time t. This represents the average impact of similar social events in history. Indicates adaptive weights for data quality. Indicates the environmental amplification factor. This represents real-time weather and environmental factors.

[0089] Through a teacher-student network framework, knowledge from pre-trained large-scale GNN or LSTM-Transformer models is transferred to a more streamlined and computationally less computationally intensive student model. By mimicking the output distribution of the teacher model, the student model significantly reduces the number of parameters and inference latency while maintaining high accuracy. The compressed student model can be deployed on edge intelligent preprocessing node modules to meet real-time requirements.

[0090] Large models are broken down into computational layers or functional modules. The message passing layer of the GNN is deployed on edge nodes for real-time feature extraction, while the complex aggregation and inference layer is deployed in the cloud for collaborative processing. Through the collaboration between the vehicle-to-everything (V2X) collaborative push module and the cloud platform, the computing load is dynamically allocated, avoiding overload of a single edge node.

[0091] In this embodiment, the spatiotemporal social prediction engine module runs a model to predict traffic flow and outputs the congestion probability of each road network node, which is implemented as follows:

[0092] ,

[0093] In the formula, Indicates time The predicted probability of congestion, Represents the spatiotemporal feature weight matrix. Indicates spatiotemporal characteristics, This represents the social impact coefficient; a confidence assessment is performed on the prediction results. If the confidence level of the social event factor is lower than the threshold, the average impact value of similar events in historical data is used for correction.

[0094] In this embodiment, the adaptive traffic light optimization module acquires a predicted congestion heat map, social impact factors, and time windows. Based on the social impact factors, it sorts congested areas according to their social risk levels, prioritizing the allocation of traffic resources to high-risk areas and delaying the optimization of low-risk areas.

[0095] Specifically, by integrating real-time environmental data and incorporating environmental impact coefficients into the social risk ranking, and by scanning the traffic light phases at adjacent intersections based on the road network topology, vehicle crossover conflicts can be eliminated through time-series fine-tuning.

[0096] In this embodiment, the vehicle-to-everything (V2X) collaborative push module receives the real-time timing scheme output by the adaptive traffic light optimization module, converts it into a V2X standard message format, and embeds social impact prompts and precise timestamps.

[0097] Specifically, based on the current traffic density and V2X network congestion index, the message transmission frequency and transmission power are dynamically adjusted to optimize transmission efficiency and coverage. A multi-hop propagation network is constructed through the collaboration of roadside units and vehicle-mounted equipment to prioritize broadcasting messages to high-risk areas. Message push is triggered at a preset time before the traffic light status changes. The push priority is dynamically adjusted for different vehicle types, and the success rate, average latency, and vehicle response time of each push are recorded in real time to generate structured logs.

[0098] In this embodiment, data privacy and security are addressed as follows: the system involves a large amount of personal data such as videos and trajectories; privacy protection technologies employed include federated learning, training models at edge nodes without uploading the original data, data anonymization, and blurring or differential privacy processing of license plates and faces.

[0099] System degradation and robustness: Consider the system degradation scheme when key modules fail, such as the prediction engine. For example, when the confidence of social event factors is low, automatically switch to the classic prediction model based on historical traffic flow to ensure the basic operation of the system.

[0100] In this embodiment, the intelligent transportation adaptive control method based on social perception includes the following steps:

[0101] S1. Based on the deployment of multimodal sensors, collect social traffic flow data through multimodal sensor fusion;

[0102] S2. Based on social traffic flow data, perform data preprocessing through edge computing;

[0103] S3. Based on the preprocessed data, perform social flow modeling using graph neural networks;

[0104] S4. Based on social flow modeling and historical social data, perform spatiotemporal prediction using a spatiotemporal prediction model;

[0105] S5. Based on the spatiotemporal prediction results, optimize the traffic lights through adaptive control;

[0106] S6. Based on the optimized traffic lights, push vehicle-to-everything (V2X) communication to the network.

[0107] S7. Based on the vehicle network push results and environmental data, a closed-loop feedback is performed through feedback analysis.

[0108] In this embodiment, for example, on a weekday evening, as it was getting dark, a primary school was letting out for the day; at this time, it was drizzling, the roads were slippery, and a large-scale event had just ended at a nearby stadium.

[0109] The system first begins with multi-source data acquisition; various sensors deployed in the multimodal social perception layer module around the school, on main roads near the stadium, and at surrounding intersections begin operation; cameras capture large gatherings of parents and students at the school gate, voiceprint sensors detect significantly increased horn noise density due to congestion, and weather sensors record light rain and dim lighting conditions; simultaneously, an independent real-time social event perception and fusion module automatically identifies the periodic event of primary school dismissal and the temporary event of stadium event dispersal through accessed public data sources, and analyzes their time, location, and expected impact range; this massive amount of raw data is sent to the nearest edge intelligent preprocessing node module; this module, on the edge server closest to the intersection, performs real-time cleaning, alignment, and timestamp synchronization of the streaming data from different sensors, filters out noise and outliers, and uploads the processed standardized data stream; the social event module also prepares quantified structured event factors;

[0110] The dynamic social flow modeling engine module receives the fused data stream; based on the real-time road network topology, it constructs a spatiotemporal graph structure, treating pedestrians, vehicles, and intersections as nodes in the graph; through a graph neural network model, it dynamically analyzes the interaction between the flow of students leaving school, the flow of audiences leaving a venue, and the normal traffic flow, quantifies the potential risks caused by rainy days, crowd gatherings, and traffic congestion, and generates a dynamic social interaction heat map.

[0111] Next, the spatiotemporal social prediction engine module compares and analyzes this real-time heatmap with similar scenarios in the historical database, such as school dismissal, rainy days, and event dispersal. Using its LSTM-Transformer hybrid model, it predicts that within the next ten minutes, the probability of severe congestion at the main intersection connecting the school and the stadium is extremely high, and the risk of pedestrians crossing illegally has significantly increased. The prediction results are sent to the adaptive traffic light optimization module and the pedestrian safety and priority passage module. Based on the predicted congestion risk points and levels, the optimization module reallocates green light times, prioritizing the rapid dispersal of traffic flowing from the stadium into the city, while simultaneously fine-tuning the traffic lights at potentially affected adjacent intersections to prevent congestion from spreading. Pedestrian safety... The module pays special attention to the area near school entrances, inserting a protected pedestrian crossing safety window during the signal cycle and activating pedestrian crossing warning devices. The optimized signal timing scheme and risk warnings are converted into standard messages by the vehicle-to-everything (V2X) collaborative push module. Through V2X technology, the system sends suggested routes to vehicles heading towards high-risk areas, reminding drivers to be aware of school zones and slippery road surfaces. For priority vehicles such as buses, green wave traffic guidance information is provided. Finally, the traffic closed-loop feedback module continuously monitors the actual traffic flow changes, vehicle throughput, and V2X message response after the scheme is implemented. This data is collected in real time and used to evaluate the effectiveness of this control measure, serving as a basis for the system to learn and optimize the next decision.

[0112] Working principle: By deploying a combination of multimodal sensors such as LiDAR, high-definition cameras, mini weather stations, and voiceprint sensors, it collects multi-dimensional traffic data such as vehicles, pedestrians, cyclists, weather, and noise in real time; based on the needs of traffic scenarios, it optimizes sensor deployment at key nodes using a gridded and focused reinforcement strategy to achieve time synchronization and feature alignment of different modal data; and cleans the data through regular expression rules and spell check libraries to form a unified semantic representation.

[0113] By receiving traffic flow data output from the multimodal perception layer on edge devices, a dynamic circular buffer is configured to ensure data continuity. Timestamp alignment enables precise matching of heterogeneous data from LiDAR, cameras, and other sources in the time dimension. The raw data is smoothed to remove sensor noise, and a statistical anomaly detection mechanism is used to eliminate unreasonable values. Simultaneously, feature extraction and compression are performed to generate structured preprocessed data packages. Based on the preprocessed data, a spatiotemporal graph of the traffic network is constructed, using intersections, pedestrian trajectory points, and vehicle positions as nodes and movement paths as edges. Attribute features such as node position, behavior type, and speed are extracted to form standardized vectors. By calculating the weights of relative positions, movement directions, and interaction frequencies between nodes, and combining pedestrian behavior factors and environmental factors, the social interaction impact is quantified. An optimized graph neural network is used for multi-layer message passing and feature aggregation, dynamically updating node social features and generating real-time interaction heatmaps and impact coefficients to accurately characterize the dynamic effect of social behavior on traffic flow. The system receives traffic density matrices and social interaction heatmaps output from social flow modeling, correlates them with historical social data to achieve spatiotemporal alignment of multi-source data, dynamically identifies key social events, and encodes them as feature vectors. Spatial features, temporal features, and social features are integrated. The system constructs a multi-dimensional time series based on features and inputs it into an LSTM-Transformer hybrid prediction model. During the inference phase, it dynamically adjusts the weights of social events based on real-time interactive heatmaps to generate a congestion probability prediction for future traffic flow. Based on the congestion heatmap and social impact factors output by the spatiotemporal prediction engine, congested areas are sorted according to social risk levels, prioritizing access to high-risk areas. It dynamically adjusts green light duration and the proportion of green lanes by integrating real-time environmental data, scans adjacent intersection phases based on road network topology, and eliminates vehicle crossover conflicts through time-series fine-tuning to generate an optimized traffic light timing scheme. The optimized scheme is then... The traffic light timing scheme is converted into a V2X standard message format, embedding social impact prompts and precise timestamps; the message transmission frequency and transmission power are dynamically adjusted based on traffic flow density and network congestion status, and a multi-hop propagation network is constructed through roadside units and on-board equipment to prioritize broadcasting information to high-risk areas, triggering push notifications before traffic light status changes, and customizing push priorities for different vehicle types; based on the vehicle network push results and environmental data, the push success rate, traffic flow changes, and system optimization effects are analyzed in real time, structured logs are recorded, and feedback indicators are evaluated; the analysis results are used to trigger system adaptive optimization.

[0114] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their likenesses.

[0115] The present invention and its embodiments have been described above. This description is not restrictive, and the accompanying drawings are only one embodiment of the present invention; the actual structure is not limited thereto. In conclusion, if those skilled in the art are inspired by this description and design similar structures and embodiments without departing from the spirit of the invention, such designs should fall within the protection scope of the present invention.

Claims

1. A socially-aware intelligent transportation adaptive control system, characterized in that: It includes a multimodal social perception layer module, a social event real-time perception and fusion module, an edge intelligent preprocessing node module, a dynamic social flow modeling engine module, a spatiotemporal social prediction engine module, an adaptive traffic light optimization module, a pedestrian safety and priority passage module, a vehicle network collaborative push module, and a traffic closed-loop feedback module. The multimodal social perception layer module is deployed based on multimodal sensors and collects social traffic flow data through multimodal sensor fusion; The edge intelligent preprocessing node module performs data preprocessing based on social traffic flow data through edge computing. The real-time social event perception and fusion module collects, quantifies, and fuses social event data in real time based on non-sensor data sources through data capture and analysis. The dynamic social flow modeling engine module performs social flow modeling based on the preprocessed data using a graph neural network. The spatiotemporal social prediction engine module performs spatiotemporal predictions based on social flow modeling and historical social data, using a spatiotemporal prediction model. The adaptive traffic light optimization module optimizes traffic lights through adaptive control based on spatiotemporal prediction results. The pedestrian safety and priority passage module controls pedestrian safety windows and priority passage based on social flow modeling and prediction results through safety assessment and priority allocation; The vehicle-to-everything (V2X) collaborative push module pushes information to the network based on the optimized traffic lights via V2X communication. The traffic closed-loop feedback module performs closed-loop feedback based on the vehicle network push results and environmental data through feedback analysis.

2. The intelligent transportation adaptive control system based on social perception according to claim 1, characterized in that: The multimodal social perception layer module deploys multimodal sensors according to traffic scenarios; and adopts a gridded and focused deployment strategy based on traffic flow distribution and key monitoring areas.

3. The intelligent transportation adaptive control system based on social perception according to claim 1, characterized in that: The edge intelligent preprocessing node module receives traffic flow data streams, establishes a circular buffer to ensure data continuity, dynamically configures the buffer size according to the computing power and network bandwidth of the edge devices, accurately matches heterogeneous data in the time dimension through timestamp alignment, smooths the original data, and effectively removes sensor noise; at the same time, based on a statistical anomaly detection mechanism, it identifies and removes outliers that exceed a reasonable range.

4. The intelligent transportation adaptive control system based on social perception according to claim 1, characterized in that: The dynamic social flow modeling engine module receives preprocessed heterogeneous data, including node location coordinates, behavior type labels, real-time speed and key environmental parameters. Based on the traffic network topology, it initializes the spatiotemporal graph structure by taking intersections, pedestrian trajectory points and vehicle positions as nodes and the movement paths of pedestrians and vehicles as edges. Node attribute features are extracted from preprocessed data, including location coordinates, behavior type, real-time speed, density value, and social attributes, forming a standardized feature vector. Based on the relative position, movement direction, and interaction frequency between nodes, the weight features of edges are calculated, resulting in: ， In the formula, Representing an edge Interaction weights This represents the Euclidean distance between the pedestrian node j and the vehicle node j. This represents the behavioral factor of pedestrian node i. This refers to weather and environmental factors.

5. The intelligent transportation adaptive control system based on social perception according to claim 4, characterized in that: The dynamic social flow modeling engine module loads a pre-trained spatiotemporal graph neural network model. This model optimizes traffic scenarios, runs GNN forward propagation on a graph structure, aggregates neighbor node information through multi-layer message passing, and dynamically updates the feature representation of each node. This is achieved as follows: ， In the formula, This represents the updated social characteristics of node i at level l+1. Let i represent the set of neighbors of node i. This represents the weight matrix of the science departments in the l-th layer. This represents the characteristics of neighbor node j at layer l. Indicates the social impact coefficient. This represents the activation function.

6. The intelligent transportation adaptive control system based on social perception according to claim 1, characterized in that: The spatiotemporal social prediction engine module receives a real-time traffic density matrix and a social interaction heatmap, and associates it with historical social data stored on a cloud platform. It achieves spatiotemporal alignment of multi-source data through timestamp matching and interpolation, dynamically identifies key social events from the historical social data, extracts event types, durations, and historical impact strengths, and encodes them into quantifiable feature vectors. These vectors are then fused with the real-time social interaction heatmap in both spatial and temporal dimensions. Based on the aligned data, a multi-dimensional time-series input sequence is constructed. This time-series sequence is then input into a pre-trained LSTM-Transformer hybrid prediction model. During the model inference phase, the weights of social event factors are dynamically adjusted based on the output of the real-time social interaction heatmap. ， In the formula, Represents the social impact factors at time t. A real-time social interaction heatmap representing time t. This represents the average impact of similar social events in history. Indicates adaptive weights for data quality. Indicates the environmental amplification factor. This represents real-time weather and environmental factors.

7. The intelligent transportation adaptive control system based on social perception according to claim 6, characterized in that: The spatiotemporal social prediction engine module runs a model to predict traffic flow and outputs the congestion probability of each road network node, as follows: ， In the formula, Indicates time The predicted probability of congestion, Represents the spatiotemporal feature weight matrix. Indicates spatiotemporal characteristics, This represents the social impact coefficient; a confidence assessment is performed on the prediction results. If the confidence level of the social event factor is lower than the threshold, the average impact value of similar events in historical data is used for correction.

8. The intelligent transportation adaptive control system based on social perception according to claim 1, characterized in that: The adaptive traffic light optimization module acquires a predicted congestion heat map, social impact factors, and time windows. Based on the social impact factors, it sorts congested areas according to their social risk levels, prioritizing the allocation of traffic resources to high-risk areas and delaying the optimization of low-risk areas. By integrating real-time environmental data and incorporating environmental impact coefficients into the social risk ranking, and based on the road network topology, scanning the traffic light phases at adjacent intersections, vehicle crossover conflicts are eliminated through time-series fine-tuning.

9. The intelligent transportation adaptive control system based on social perception according to claim 1, characterized in that: The vehicle-to-everything (V2X) collaborative push module receives the real-time timing scheme output by the adaptive traffic light optimization module, converts it into a V2X standard message format, embeds social impact prompts and precise timestamps; based on the current traffic flow density and V2X network congestion index, it dynamically adjusts the message transmission frequency and transmission power to optimize transmission efficiency and coverage; through the collaborative construction of a multi-hop propagation network by roadside units and vehicle-mounted equipment, it prioritizes broadcasting messages to high-risk areas; it triggers message pushes at a preset time before traffic light status changes; it dynamically adjusts the push priority for different vehicle types; and it records the success rate, average latency, and vehicle response time of each push in real time to generate structured logs.

10. The socially-aware-based intelligent transportation adaptive control method according to claim 1, characterized in that, Includes the following steps: S1. Based on the deployment of multimodal sensors, collect social traffic flow data through multimodal sensor fusion; S2. Based on social traffic flow data, perform data preprocessing through edge computing; S3. Based on the preprocessed data, perform social flow modeling using graph neural networks; S4. Based on social flow modeling and historical social data, perform spatiotemporal prediction using a spatiotemporal prediction model; S5. Based on the spatiotemporal prediction results, optimize the traffic lights through adaptive control; S6. Based on the optimized traffic lights, push vehicle-to-everything (V2X) communication to the network. S7. Based on the vehicle network push results and environmental data, a closed-loop feedback is performed through feedback analysis.

Images