Electronic information traffic flow automatic control system based on wireless sensor network

Abstract

The present invention relates to the technical field of traffic flow control, and discloses an automatic control system for electronic information traffic flow based on a wireless sensor network. The multi-source perception module collects multi-dimensional traffic data of the road network through the wireless sensor network, and generates traffic situation characteristics through a specific data fusion method. The spatiotemporal feature fusion module uses the spatiotemporal attention mechanism to associate cross-modal features, and the collaborative decision-making module inputs the fused features into a pre-trained distributed decision model to generate a control instruction set. The dynamic game optimization module constructs a multi-objective game optimization model and uses a dynamic game strategy decomposition algorithm to optimize traffic signal parameters. The hierarchical execution module executes control instructions in a distributed manner through a three-level control architecture of a central decision-making layer, a regional coordination layer, and an intersection execution layer. The system can comprehensively collect and integrate traffic data, realize scientific decision-making and precise control, effectively improve road traffic efficiency, balance road network load, and alleviate traffic congestion.

Description

Technical Field

[0001] The present invention relates to the technical field of traffic flow control, in particular to an electronic information traffic flow automatic control system based on a wireless sensor network. Background Art

[0002] With the acceleration of urbanization and the rapid growth of car ownership, urban traffic congestion is becoming increasingly serious. Traditional traffic flow control methods, such as timed signal control and manual intervention, are no longer able to meet the complex needs of modern transportation and have exposed many drawbacks.

[0003] Previous methods for collecting traffic data were significantly inadequate. Early methods relied on a single type of sensor, such as geomagnetic sensors, which could only capture simple traffic flow data and failed to fully reflect traffic conditions. For example, it was impossible to determine a vehicle's specific location, speed, and other information, making it difficult to detect sudden traffic incidents such as accidents and road construction in real time. This meant that traffic management departments lacked sufficient and accurate data support when formulating control strategies, and control measures were often unintended. Although video monitoring equipment was later introduced, the various types of sensors were independent of each other, and the data was not effectively integrated, resulting in "data silos." Data from different sources differed in time and space, making it difficult to integrate and utilize them, and thus failing to provide complete and accurate traffic situation awareness for traffic control.

[0004] From a traffic flow control perspective, traditional timed signal control methods switch lights according to a preset fixed time period, disregarding real-time changes in actual traffic flow. During peak hours, traffic at some intersections surges, but signal light durations are not adjusted in a timely manner, resulting in long queues and extremely low road efficiency. Meanwhile, during low traffic periods, intersections are sparsely populated, yet signal lights continue to switch according to fixed durations, resulting in a waste of road resources. While manual intervention can, to a certain extent, adapt to on-site conditions, it relies on manpower, has a slow response time, and is susceptible to human influence, making it impossible to achieve large-scale, efficient traffic control.

[0005] Furthermore, traditional traffic control systems lack a holistic understanding of the overall traffic situation and the ability to coordinate traffic flow. Controls at various intersections and road sections operate independently, failing to fully consider the coherence and mutual influence of traffic flows within the region. When congestion occurs at a particular intersection, there's no timely coordination with surrounding intersections, leading to a widening congestion and impacting traffic flow across the entire region. For example, in some urban commercial districts, where multiple intersections connect and traffic flows are complex and volatile, traditional control methods are unable to effectively coordinate traffic lights at these intersections, exacerbating traffic congestion.

[0006] Furthermore, traditional systems fail to fully consider traffic environmental factors. Meteorological conditions, such as heavy rain and fog, can significantly impact traffic flow and vehicle speeds, but traditional traffic control systems rarely incorporate these factors into their decision-making. In inclement weather, visibility decreases, vehicle speeds slow, and traffic flow distribution shifts. Traditional control methods fail to adapt in a timely manner, further exacerbating traffic congestion and safety risks. Summary of the Invention

[0007] The purpose of the present invention is to provide an electronic information traffic flow automatic control system based on a wireless sensor network to solve the problems raised in the above background technology.

[0008] To achieve the above-mentioned object, the present invention provides the following technical solution: an automatic control system for electronic information traffic flow based on a wireless sensor network, the system comprising:

[0009] Multi-source sensing module: used to collect multi-dimensional traffic data of the road network through a wireless sensor network, which includes a vehicle-mounted positioning terminal, a geomagnetic flow sensor, a video monitoring unit and a weather monitoring station;

[0010] Spatiotemporal feature fusion module: Based on the spatiotemporal attention mechanism, cross-modal feature correlation is performed on multi-dimensional traffic data to generate traffic situation features that integrate spatiotemporal characteristics;

[0011] Collaborative decision-making module: inputs the traffic situation characteristics into a pre-trained distributed decision-making model to generate a set of regional traffic flow control instructions;

[0012] Dynamic game optimization module: Constructs a multi-objective game optimization model based on the control instruction set. The multi-objective game optimization model takes maximizing traffic efficiency and balancing road network load as optimization goals, and uses a dynamic game strategy decomposition algorithm to collaboratively optimize traffic signal parameters;

[0013] Hierarchical execution module: Based on the multi-objective game optimization model, the optimal control strategy is output, and the distributed execution of control instructions is realized through a three-level control architecture. The three-level control architecture includes a central decision-making layer, a regional coordination layer and an intersection execution layer. The central decision-making layer generates a global phase timing plan based on the road network topology, the regional coordination layer uses a sliding window prediction algorithm to dynamically adapt the traffic parameters of the sub-region, and the intersection execution layer tracks the signal cycle and the green-to-signal ratio based on the fuzzy model prediction control algorithm.

[0014] Preferably, the data fusion method of the multi-source perception module includes:

[0015] Perform spatiotemporal calibration on vehicle-mounted positioning terminal data and geomagnetic flow sensor data to construct a real-time traffic distribution heat map of the road network; perform multi-scale wavelet decomposition on video monitoring unit data and meteorological monitoring station data to extract traffic event feature vectors;

[0016] Construct a dual-branch feature fusion network, where the first branch uses a three-dimensional convolution kernel to extract the spatial correlation features of the traffic distribution heat map, and the second branch uses a gated temporal network to extract the dynamic evolution law of the traffic event feature vector;

[0017] The spatial correlation features and dynamic evolution features are fused through the cross-channel attention mechanism to generate a spatiotemporal joint matrix. The spatiotemporal joint matrix is modeled with multi-granularity features based on residual dilated convolution, and traffic situation characteristics including vehicle density, event impact range and meteorological interference intensity are output.

[0018] Preferably, the distributed decision-making model adopts a spatiotemporal graph convolutional network architecture to generate multi-level control instructions based on a dynamic node aggregation mechanism; the spatiotemporal graph convolutional network architecture includes:

[0019] Construct a road network-vehicle interaction graph, where nodes include intersection nodes, road section nodes, vehicle nodes, and weather nodes. Node attributes include queue length, average vehicle speed, and visibility parameters.

[0020] A bimodal graph attention mechanism is used. In the first stage, the spatial graph convolution layer is used to calculate the association weights between intersection nodes and adjacent road sections. In the second stage, the temporal graph convolution layer is used to screen the importance of historical traffic states.

[0021] Based on the multi-head spatiotemporal attention module, node features are iteratively optimized. Each attention head integrates static road network properties and dynamic traffic parameters. The network training stability is improved through skip connections and layer normalization mechanisms. The final output is a control instruction set that includes emergency vehicle priority passage and regional congestion evacuation.

[0022] Preferably, the dynamic game strategy decomposition algorithm integrating Nash equilibrium solution and constraint relaxation mechanism includes:

[0023] The traffic signal optimization problem is modeled as a multi-objective mixed integer dynamic game problem, where the decision variables include discrete signal phase switching variables and continuous green light duration adjustment variables.

[0024] Initialize the relaxed game subproblem and calculate the approximate set of equilibrium solutions. Use a dynamic benefit adjustment mechanism to update the utility functions of the game participants according to real-time traffic demand.

[0025] In the decomposition stage, the global game problem is divided into multiple sub-game problems based on the Pareto frontier analysis technology; in the coordination stage, virtual game players are introduced to balance the conflicts of interest between the sub-games;

[0026] A parallel game solver is used to alternately optimize the subproblems, updating the local equilibrium solution in each iteration and synchronizing the global game strategy through coordination variables.

[0027] Preferably, the sliding window prediction algorithm dynamically adapts the sub-region traffic parameters including:

[0028] A time-varying traffic flow prediction model is constructed, discretizing the vehicle kinematic equations into state transition equations that include car-following model errors, lane-changing decision delays, and signal control influences.

[0029] A dynamic window optimization objective function is designed, which includes queue length suppression term, traffic delay balance term and emergency vehicle priority guarantee term.

[0030] Preferably, the intersection execution layer implements tracking of signal cycle and green-to-signal ratio based on fuzzy model predictive control algorithm, including:

[0031] A three-layer fuzzy prediction and reasoning system is designed to map phase deviation and traffic fluctuation into fuzzy rule confidence. An adaptive domain adjustment mechanism is constructed to dynamically adjust the coverage interval of the fuzzy set according to the traffic state change rate.

[0032] Preferably, the three-dimensional convolution kernel adopts a multi-scale dilated convolution structure to improve feature extraction efficiency, including:

[0033] The real-time traffic distribution heat map of the road network is divided into multi-level grid units, each of which stores the traffic speed variance and density gradient statistics;

[0034] In the feature extraction stage, the dilated convolution kernel is used to expand the spatial perception range, and in the feature reconstruction stage, the transposed convolution layer is used to restore local detail features;

[0035] A spatial attention gating mechanism is introduced to perform dynamic weighted fusion of feature maps.

[0036] Preferably, the spatial graph convolution layer adopts a dynamic adjacency coding mechanism, including:

[0037] Define the dynamic association vector between the intersection node and the adjacent road segments, including vehicle turning probability, lane connection relationship and historical traffic efficiency;

[0038] The dynamic correlation vector is converted into the bias weight parameters of the graph convolution kernel through a nonlinear mapping layer;

[0039] The bias weight parameters are integrated into the graph convolution operation process to generate a node feature representation with spatial adaptability.

[0040] Preferably, the dynamic profit adjustment mechanism is implemented based on an online reinforcement learning strategy, including:

[0041] Collect the strategy selection records and equilibrium solution distribution in the historical game process as training data sets;

[0042] Construct a deep Q-learning network to fit the changing patterns of the payoff functions of the game participants;

[0043] Update network parameters online through asynchronous policy gradient descent method, and adjust the utility function weights of each participant in real time;

[0044] When an unexpected traffic event is detected, an emergency reconstruction operation of the profit function is triggered.

[0045] Preferably, the error modeling of the state transfer equation adopts interval envelope analysis technology, including: modeling the vehicle motion uncertainty as interval parameters, with the upper and lower bounds determined by the standard deviation of the sensor measurement error; performing orthogonal decomposition on the interval parameters to separate the system deterministic components and the environmental random interference components.

[0046] Compared with the prior art, the present invention has the following beneficial effects:

[0047] At the data collection and fusion level, the multi-source perception module leverages wireless sensor networks to integrate on-board positioning terminals, geomagnetic flow sensors, video monitoring units, and meteorological monitoring stations to collect multi-dimensional traffic data. Through a unique data fusion method, different types of data are spatiotemporally aligned and feature extracted to construct a more comprehensive and accurate picture of traffic trends. For example, after spatiotemporally aligning data from on-board positioning terminals and geomagnetic flow sensors, a real-time traffic distribution heat map is constructed for the road network, clearly depicting real-time traffic dynamics. Multi-scale wavelet decomposition is performed on video monitoring unit and meteorological station data to extract traffic event feature vectors, effectively mining potential traffic anomaly information. Compared to traditional single-data collection methods, this significantly improves the richness and accuracy of data, providing a solid data foundation for subsequent regulatory decisions.

[0048] The spatiotemporal feature fusion module performs cross-modal feature correlation on multi-dimensional traffic data based on the spatiotemporal attention mechanism. This innovative fusion method can accurately capture the intrinsic connection between traffic data in the temporal and spatial dimensions, and generate traffic situation characteristics that integrate spatiotemporal characteristics. Compared with traditional data processing methods, it is no longer limited to the analysis of single-dimensional data, but starts from a comprehensive perspective of time and space to more comprehensively and deeply understand the changing patterns of traffic situation. For example, when analyzing traffic congestion during peak hours in the morning and evening, it can not only take into account the spatial distribution of traffic flow at the current intersection, but also combine the temporal change trends in historical data to provide a more forward-looking and targeted basis for traffic control.

[0049] The collaborative decision-making module inputs the integrated traffic situation features into a pre-trained distributed decision-making model to generate a set of regional traffic flow control instructions. This distributed decision-making model utilizes a spatiotemporal graph convolutional network architecture. By constructing a road network-vehicle interaction graph and employing a bimodal graph attention mechanism, it fully considers the interrelationships between various factors in the road network, such as the connections between intersection nodes, road section nodes, vehicle nodes, and meteorological nodes. This makes the generated control instructions more scientific and rational, capable of comprehensively balancing traffic demands across different regions and scenarios, achieving effective control of regional traffic flows and improving overall traffic operation efficiency.

[0050] The dynamic game optimization module constructs a multi-objective game optimization model based on the control instruction set. With the goals of maximizing traffic efficiency and balancing road network load, it uses a dynamic game strategy decomposition algorithm to collaboratively optimize traffic signal parameters. This optimization approach breaks the limitations of traditional control, which focuses solely on a single objective. By integrating Nash equilibrium solving with a constraint relaxation mechanism, it can dynamically adjust traffic signal phase switching variables and green light duration adjustment variables to better adapt to complex and changing traffic flow. For example, when traffic flow varies significantly across different time periods and sections, it can reasonably allocate green light duration, ensuring traffic efficiency on main roads while avoiding excessive congestion on certain branches, thereby achieving load balancing across the entire road network.

[0051] The hierarchical execution module outputs the optimal control strategy based on a multi-objective game optimization model, and realizes the distributed execution of control instructions through a three-level control architecture. The central decision-making layer generates a global phase timing plan based on the road network topology, controlling the overall traffic rhythm from a macro perspective; the regional coordination layer uses a sliding window prediction algorithm to dynamically adapt the traffic parameters of the sub-region, and can flexibly adjust the control strategy within the region according to the real-time traffic conditions; the intersection execution layer uses a fuzzy model predictive control algorithm to track the signal cycle and green signal ratio, and performs fine-grained control based on the specific traffic conditions of each intersection. This hierarchical execution architecture has a clear division of labor and coordinated cooperation, which greatly improves the accuracy and timeliness of the execution of control instructions, ensuring that traffic control strategies can be efficiently implemented in actual scenarios.

[0052] Furthermore, key algorithms across all modules in the system have been optimized and improved. For example, the three-dimensional convolution kernel utilizes a multi-scale dilated convolution structure to improve feature extraction efficiency; the spatial graph convolution layer employs a dynamic adjacency encoding mechanism to enhance spatial adaptability; the dynamic benefit adjustment mechanism leverages an online reinforcement learning strategy to achieve more flexible utility function adjustments; and the error modeling of the state transition equation utilizes interval envelope analysis techniques to improve prediction accuracy. These algorithmic optimizations further enhance the system's performance and adaptability, enabling more stable and efficient operation in complex traffic scenarios and unexpected traffic incidents, providing strong technical support for alleviating urban traffic congestion, improving road efficiency, and ensuring traffic safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 This is a working principle diagram of the electronic information traffic flow automatic control system based on wireless sensor network according to the present invention;

[0054] Figure 2 This is the workflow diagram for data fusion of multi-source perception modules;

[0055] Figure 3 This is the workflow diagram of the distributed decision-making model;

[0056] Figure 4 Workflow diagram of the dynamic adjacency encoding mechanism for the spatial graph convolutional layer. DETAILED DESCRIPTION

[0057] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0058] See also Figures 1-4 The present invention provides an electronic information traffic flow automatic control system based on a wireless sensor network, and its specific implementation method will be described in detail below.

[0059] Multi-source sensing module: This module utilizes a wireless sensor network to collect multi-dimensional traffic data from the road network. Vehicle-mounted positioning terminals acquire real-time information such as vehicle location and speed. Geomagnetic flow sensors accurately measure traffic flow on a road section by sensing changes in the magnetic field generated by passing vehicles. Video monitoring units capture video through cameras, visually monitoring traffic scenes and identifying traffic events such as accidents and traffic violations. Meteorological monitoring stations collect weather conditions, visibility, and other meteorological data. This multi-dimensional traffic data provides comprehensive foundational information for subsequent analysis and control.

[0060] The spatiotemporal feature fusion module uses a spatiotemporal attention mechanism to perform cross-modal feature correlation on the multi-dimensional traffic data collected by the multi-source perception module. Different types of traffic data have different characteristics and importance in time and space. The spatiotemporal attention mechanism automatically learns the correlations between these features, highlighting key information and suppressing redundant information. This generates traffic situation features that integrate spatiotemporal characteristics, providing more valuable information for subsequent decision-making.

[0061] Collaborative Decision-Making Module: This module feeds the traffic situation features generated by the spatiotemporal feature fusion module into a pre-trained distributed decision-making model. This model, trained with extensive historical traffic data, possesses the ability to analyze and make decisions regarding complex traffic conditions. By inputting current traffic situation features, the model generates a set of control instructions for regional traffic flow, encompassing traffic control strategies for individual intersections and road sections.

[0062] Dynamic Game Optimization Module: This module constructs a multi-objective game optimization model based on the control instruction set generated by the collaborative decision-making module. This model optimizes traffic signal parameters using a dynamic game strategy decomposition algorithm, maximizing traffic efficiency and balancing road network load. In real-world traffic scenarios, traffic demands at different intersections and road sections interact with each other. This dynamic game strategy decomposition algorithm can achieve the optimal configuration of traffic signal parameters while considering the interests and constraints of all parties involved, thereby improving the operational efficiency of the entire road network.

[0063] Hierarchical Execution Module: Based on the optimal control strategy output by the dynamic game optimization module, distributed execution of control instructions is achieved through a three-level control architecture. The central decision-making layer generates a global phase timing plan based on the road network topology and comprehensive consideration of global traffic conditions, regulating traffic flow from a macro perspective. The regional coordination layer uses a sliding window prediction algorithm to dynamically adapt sub-regional traffic parameters and adjust local control strategies based on real-time traffic changes. The intersection execution layer uses a fuzzy model predictive control algorithm to track signal cycles and green-to-signal ratios. Based on the actual traffic flow and phase deviation at the intersection, it accurately controls the switching of traffic lights to ensure efficient operation of intersection traffic.

[0064] Other technical features of the present invention are described in detail below by way of embodiments. Example 1

[0065] In this embodiment, the data fusion method of the multi-source perception module is described in detail. First, the data of the vehicle-mounted positioning terminal and the data of the geomagnetic flow sensor are time-space calibrated. The vehicle position and time information obtained by the vehicle-mounted positioning terminal differ from the traffic flow data monitored by the geomagnetic flow sensor in time and space. Through time-space calibration, the data of the two are unified into the same time-space coordinate system. For example, based on the vehicle's driving trajectory and timestamp, combined with the location information of the geomagnetic flow sensor, the exact time and flow data of each vehicle passing through each road section are determined, and then a real-time traffic distribution heat map of the road network is constructed. The heat map uses intuitive color changes to show the traffic density in different areas of the road network. Red indicates high traffic density, and green indicates low traffic density.

[0066] Multiscale wavelet decomposition is used to extract traffic event feature vectors from video surveillance unit data and meteorological monitoring station data. Video footage captured by the video surveillance unit contains rich traffic scene information, while data from the meteorological monitoring station reflects the impact of weather on traffic. Multiscale wavelet decomposition can analyze data at different temporal and spatial scales, extracting the characteristics of traffic events such as accidents and congestion, as well as the impact of meteorological factors such as heavy rain and fog on traffic.

[0067] A two-branch feature fusion network is constructed. The first branch uses a three-dimensional convolution kernel to extract the spatial correlation features of the traffic distribution heat map. The three-dimensional convolution kernel performs a convolution operation on the heat map in three spatial dimensions, which can capture the spatial relationship between traffic flows in different areas, such as the mutual influence of traffic flows on adjacent road sections. The second branch uses a gated temporal network to extract the dynamic evolution of traffic event feature vectors. The gated temporal network can learn how traffic event characteristics change over time based on time series data, such as the changing impact of traffic accidents from their occurrence to their resolution.

[0068] A cross-channel attention mechanism fuses spatial correlation features with dynamic evolution features to generate a joint spatiotemporal matrix. This cross-channel attention mechanism automatically assigns weights to features across different channels, highlighting important information. Multi-granularity feature modeling is performed on the joint spatiotemporal matrix based on residual dilated convolution, outputting traffic situation features including traffic density, event impact range, and meteorological interference intensity. By introducing a dilated convolution kernel, residual dilated convolution expands the receptive field, extracting richer multi-granularity features and making the output traffic situation features more comprehensive and accurate.

[0069] In this process, if related calculations are involved, taking multi-scale wavelet decomposition as an example, assuming that the signal , its wavelet decomposition formula is:

[0070] ;

[0071] in, It represents the wavelet transform coefficient, t is the time variable, a is the scale parameter, which determines the degree of expansion and contraction of the wavelet function and controls the frequency range of the decomposition; b is the translation parameter, which determines the position of the wavelet function on the time axis; Represents the conjugate function of the wavelet function. By selecting appropriate wavelet basis functions and different scale and translation parameters, the video monitoring unit data and meteorological monitoring station data can be effectively decomposed at multiple scales to extract valuable feature vectors. Example 2

[0072] This embodiment focuses on the distributed decision-making model. This distributed decision-making model uses a spatiotemporal graph convolutional network architecture. First, a road network-vehicle interaction graph is constructed. The nodes in the graph include intersection nodes, road section nodes, vehicle nodes, and weather nodes. Each node has specific attributes, such as the queue length at the intersection node, which reflects the number of vehicles waiting to pass through the intersection; the average speed at the road section node, which reflects the driving speed of vehicles on the road section; the vehicle node can contain information such as vehicle type and driving direction; and the visibility parameter of the weather node is used to measure the impact of weather conditions on traffic.

[0073] A bimodal graph attention mechanism is used. In the first stage, the association weights between intersection nodes and adjacent road sections are calculated through the spatial graph convolution layer. The spatial graph convolution layer uses a dynamic adjacency encoding mechanism to define the dynamic association vector between intersection nodes and adjacent road sections. This vector contains information such as vehicle turning probability, lane connection relationship, and historical traffic efficiency. The dynamic association vector is converted into the bias weight parameter of the graph convolution kernel through a nonlinear mapping layer, and the bias weight parameter is then integrated into the graph convolution operation process to generate a spatially adaptable node feature representation. For example, if the probability of a vehicle turning left at an intersection is high, and the turning lane has a close connection with the adjacent road section and has a good historical traffic efficiency, then when calculating the association weight, the adjacent road section will be given a higher weight, highlighting its important impact on intersection traffic.

[0074] In the second stage, the temporal graph convolution layer filters historical traffic status for importance. This layer analyzes historical traffic data based on time series and selects historical status information that is crucial for current decision-making. For example, during the morning rush hour, traffic congestion during the same time period over the past week can be analyzed to provide a reference for current traffic control decisions.

[0075] Node features are iteratively optimized based on a multi-head spatiotemporal attention module, with each attention head integrating static road network attributes with dynamic traffic parameters. Network training stability is enhanced through skip connections and layer normalization. Skip connections prevent the vanishing gradient problem, enabling the network to better learn deep-level features; layer normalization normalizes the inputs of each layer, accelerating network convergence. Ultimately, the model output includes a set of control instructions for emergency vehicle priority and regional congestion evacuation. For example, priority routes are planned for emergency vehicles like fire trucks and ambulances, and evacuation strategies are developed for congested areas. Example 3

[0076] This embodiment details a dynamic game strategy decomposition algorithm that integrates Nash equilibrium solving and constraint relaxation mechanisms. First, the traffic signal optimization problem is modeled as a multi-objective mixed-integer dynamic game problem. The decision variables include a discrete signal phase switching variable and a continuous green light duration adjustment variable. The signal phase switching variable determines the timing of signal light switching between different phases, such as switching from a straight-ahead phase to a left-turn phase. The green light duration adjustment variable is used to adjust the green light duration for each phase to accommodate varying traffic flow demands.

[0077] Initialize the relaxed game subproblems and calculate an approximate set of equilibrium solutions. In real-world traffic scenarios, various complex constraints exist, such as maximum and minimum green light time limits at intersections and right-of-way priorities for vehicles traveling in different directions. Through the constraint relaxation mechanism, these constraints are appropriately relaxed, transforming the complex global game problem into multiple relatively simple relaxed game subproblems. A dynamic payoff adjustment mechanism is employed to update the utility functions of the game participants based on real-time traffic demand. For example, when traffic on a particular road suddenly increases, the utility function weight for vehicles traveling in that direction is correspondingly increased to encourage more green light time to be allocated to that direction.

[0078] In the decomposition phase, the global game problem is divided into multiple sub-games using Pareto front analysis. Pareto front analysis can identify the optimal trade-offs between multiple optimization objectives, breaking the global problem down into multiple sub-problems, each corresponding to a local optimization objective. In the coordination phase, a virtual player is introduced to balance the conflicts of interest between the sub-problems. The virtual player can be viewed as a coordinator, adjusting the decision-making strategies of each sub-problem to achieve an optimal balance of traffic flow across the entire road network.

[0079] A parallel game solver is used to alternately optimize the subproblems, updating the local equilibrium solution with each iteration and synchronizing the global game strategy through coordination variables. The parallel game solver can process multiple subproblems simultaneously, improving solution efficiency. During each iteration, the global game strategy is adjusted through coordination variables based on the local equilibrium solutions of each subproblem, gradually approaching the optimal traffic signal control solution.

[0080] Taking the calculation of Nash equilibrium solution as an example, assuming that there are n participants in the game, the strategy set of each participant i is , the utility function is ,in . Nash equilibrium solution , satisfying any ,have , for all That is, each participant chooses its own optimal strategy while the strategies of the other participants are fixed, thus achieving a stable equilibrium state. In the traffic signal optimization problem of the present invention, by continuously adjusting the decision variables of each intersection and road section, the optimal solution that satisfies the Nash equilibrium conditions is found, thus achieving the optimal control of traffic signals. Example 4

[0081] This embodiment details the process of dynamically adapting the sub-region traffic parameters using the sliding window prediction algorithm. First, a time-varying traffic flow prediction model is constructed, and the vehicle kinematic equation is discretized into a state transition equation. The vehicle kinematic equation describes the motion of a vehicle on the road. Considering various factors in actual traffic, such as the error of the following model, the lane change decision delay, and the influence of signal control, the equation is discretized. Assume that the state of the vehicle at time k is , the state transfer equation can be expressed as:

[0082] ;

[0083] in, is the state transfer matrix, which describes the change of vehicle state over time. Its elements are related to factors such as the vehicle's dynamic characteristics and road conditions; is the input matrix, which is used to control the input Mapped to the state space, It can represent the vehicle's acceleration, steering angle and other control quantities; The noise is process noise, which includes uncertain factors such as following model error, lane change decision delay and signal control influence. The interval envelope analysis technique is used to model the error. The vehicle motion uncertainty is modeled as interval parameters. The upper and lower bounds are determined by the standard deviation of sensor measurement error. The interval parameters are orthogonally decomposed to separate the system deterministic components and the environmental random interference components.

[0084] Design a dynamic window optimization objective function, which includes queue length suppression, traffic delay balancing, and emergency vehicle priority guarantee. The queue length suppression term is used to reduce the queue length of vehicles at the intersection and improve the traffic capacity of the intersection; the traffic delay balancing term ensures that the traffic delays of vehicles in all directions are as balanced as possible to avoid excessive waiting time for vehicles in one direction; and the emergency vehicle priority guarantee provides priority passage for emergency vehicles such as fire trucks and ambulances. For example, the queue length suppression term can be expressed as ;in is the set of all lanes at the intersection, is the number of vehicles queuing in lane i. By minimizing this term, the queue length can be effectively reduced. By adjusting the weights of these three terms, different optimization objectives can be weighed according to actual traffic demand, achieving dynamic adaptation of sub-area traffic parameters. Example 5

[0085] This embodiment focuses on the intersection execution layer to achieve tracking of signal cycle and green-to-signal ratio based on fuzzy model predictive control algorithm. A three-layer fuzzy predictive reasoning system is designed to map phase deviation and traffic fluctuation into fuzzy rule confidence. Phase deviation refers to the difference between the actual phase and the ideal phase at the current intersection, and traffic fluctuation reflects the change in vehicle flow. The fuzzy predictive reasoning system includes three processes: fuzzification, fuzzy reasoning and defuzzification. In the fuzzification process, the precise values of phase deviation and traffic fluctuation are converted into membership in fuzzy sets. For example, the phase deviation is divided into fuzzy sets such as "negative large", "negative small", "zero", "positive small", and "positive large", and its membership in each fuzzy set is determined according to the actual phase deviation value.

[0086] During the fuzzy inference process, pre-defined fuzzy rules, combined with the fuzzy membership of phase deviation and traffic fluctuation, are used to infer adjustment strategies for the signal period and green-to-signal ratio. For example, if the phase deviation is positive and the traffic fluctuation is large, the fuzzy rules may indicate increasing the signal period and green-to-signal ratio in the corresponding direction. The defuzzification process converts the results of the fuzzy inference into precise control variables, which are used to adjust the signal period and green-to-signal ratio of the traffic light.

[0087] An adaptive domain adjustment mechanism is constructed to dynamically adjust the coverage of fuzzy sets based on the rate of change of traffic conditions. When traffic conditions fluctuate dramatically, the coverage of the fuzzy sets is appropriately expanded to improve system adaptability. When traffic conditions are relatively stable, the coverage of the fuzzy sets is reduced to enhance control precision. For example, during peak traffic hours, when traffic volume fluctuates significantly, expanding the coverage of the fuzzy sets allows the fuzzy model to better handle various complex traffic conditions. During off-peak traffic hours, when traffic volume is relatively stable, reducing the coverage of the fuzzy sets allows for more precise signal control. Example 6

[0088] This embodiment details how a three-dimensional convolution kernel uses a multi-scale dilated convolution structure to improve feature extraction efficiency. First, the real-time traffic distribution heat map of the road network is divided into multi-level grid cells, and each cell stores the traffic speed variance and density gradient statistics. The traffic speed variance reflects the degree of discreteness of vehicle speed, and the density gradient statistics reflect the spatial variation of traffic density. By storing these statistics, the traffic characteristics of each grid cell can be more comprehensively described.

[0089] During the feature extraction phase, dilated convolution kernels are used to expand the spatial perception range. By setting the dilation rate during the convolution operation, the dilated convolution kernel can cover a larger spatial area without increasing the number of parameters. For example, while a regular convolution kernel can only perceive a few adjacent pixels, a dilated convolution kernel can perceive pixels in more distant areas by setting an appropriate dilation rate, thereby capturing a wider range of spatially correlated features. During the feature reconstruction phase, a transposed convolution layer is used to restore local detail features. The transposed convolution layer can be viewed as the inverse operation of the convolution layer. It can restore the feature map that has been reduced in size after the convolution operation to its original size while retaining important detail information.

[0090] A spatial attention gating mechanism is introduced to dynamically weight the feature maps. This mechanism automatically assigns weights based on the importance of different regions in the feature map. Regions with significant traffic flow fluctuations and a greater impact on the overall traffic situation are given higher weights to highlight their characteristic information. Relatively unimportant regions are given lower weights to suppress redundant information. This dynamic weighted fusion approach enables more effective extraction of traffic situation features, improving both efficiency and accuracy.

[0091] Through the detailed explanation of the above embodiments, the electronic information traffic flow automatic control system based on wireless sensor network of the present invention has been fully explained in various technical details. From multi-source perception, feature fusion, decision optimization to instruction execution, each link works closely together to achieve efficient automatic control of traffic flow, which has significant practicality and innovation.

[0092] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0093] 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. An automatic control system for electronic information traffic flow based on wireless sensor network, characterized in that: include: Multi-source sensing module: used to collect multi-dimensional traffic data of the road network through a wireless sensor network, which includes a vehicle-mounted positioning terminal, a geomagnetic flow sensor, a video monitoring unit and a weather monitoring station; Spatiotemporal feature fusion module: Based on the spatiotemporal attention mechanism, cross-modal feature correlation is performed on multi-dimensional traffic data to generate traffic situation features that integrate spatiotemporal characteristics; Collaborative decision-making module: inputs the traffic situation characteristics into a pre-trained distributed decision-making model to generate a set of regional traffic flow control instructions; Dynamic game optimization module: Constructs a multi-objective game optimization model based on the control instruction set. The multi-objective game optimization model takes maximizing traffic efficiency and balancing road network load as optimization goals, and uses a dynamic game strategy decomposition algorithm to collaboratively optimize traffic signal parameters; Hierarchical execution module: Based on the multi-objective game optimization model, it outputs the optimal control strategy and implements distributed execution of control instructions through a three-level control architecture. The three-level control architecture includes a central decision-making layer, a regional coordination layer, and an intersection execution layer. The central decision-making layer generates a global phase timing plan based on the road network topology. The regional coordination layer uses a sliding window prediction algorithm to dynamically adapt sub-regional traffic parameters. The intersection execution layer uses a fuzzy model predictive control algorithm to track signal cycles and green-to-signal ratios. The data fusion method of the multi-source perception module includes: Perform spatiotemporal calibration on vehicle-mounted positioning terminal data and geomagnetic flow sensor data to construct a real-time traffic distribution heat map of the road network; perform multi-scale wavelet decomposition on video monitoring unit data and meteorological monitoring station data to extract traffic event feature vectors; Construct a dual-branch feature fusion network, where the first branch uses a three-dimensional convolution kernel to extract the spatial correlation features of the traffic distribution heat map, and the second branch uses a gated temporal network to extract the dynamic evolution law of the traffic event feature vector; The spatial correlation features and dynamic evolution features are integrated through a cross-channel attention mechanism to generate a spatiotemporal joint matrix. The spatiotemporal joint matrix is modeled with multi-granularity features based on residual dilated convolution, and traffic situation features including vehicle density, event impact range, and meteorological interference intensity are output. The dynamic game strategy decomposition algorithm integrates Nash equilibrium solution and constraint relaxation mechanism, including: The traffic signal optimization problem is modeled as a multi-objective mixed integer dynamic game problem, where the decision variables include discrete signal phase switching variables and continuous green light duration adjustment variables. Initialize the relaxed game subproblem and calculate the approximate set of equilibrium solutions. Use a dynamic benefit adjustment mechanism to update the utility functions of the game participants according to real-time traffic demand. In the decomposition stage, the global game problem is divided into multiple sub-game problems based on the Pareto frontier analysis technology; in the coordination stage, virtual game players are introduced to balance the conflicts of interest between the sub-games; A parallel game solver is used to alternately optimize the subproblems, updating the local equilibrium solution in each iteration and synchronizing the global game strategy through coordination variables.

2. The electronic information traffic flow automatic control system according to claim 1 is characterized in that: The distributed decision-making model adopts a spatiotemporal graph convolutional network architecture to generate multi-level control instructions based on a dynamic node aggregation mechanism; the spatiotemporal graph convolutional network architecture includes: Construct a road network-vehicle interaction graph, where nodes include intersection nodes, road section nodes, vehicle nodes, and weather nodes. Node attributes include queue length, average vehicle speed, and visibility parameters. A bimodal graph attention mechanism is used. In the first stage, the spatial graph convolution layer is used to calculate the association weights between intersection nodes and adjacent road sections. In the second stage, the temporal graph convolution layer is used to screen the importance of historical traffic states. Based on the multi-head spatiotemporal attention module, node features are iteratively optimized. Each attention head integrates static road network properties and dynamic traffic parameters. The network training stability is improved through skip connections and layer normalization mechanisms. The final output is a control instruction set that includes emergency vehicle priority passage and regional congestion evacuation.

3. The electronic information traffic flow automatic control system according to claim 1 is characterized in that: The sliding window prediction algorithm dynamically adapts the sub-region traffic parameters including: A time-varying traffic flow prediction model is constructed, discretizing the vehicle kinematic equations into state transition equations that include car-following model errors, lane-changing decision delays, and signal control influences. A dynamic window optimization objective function is designed, which includes queue length suppression term, traffic delay balance term and emergency vehicle priority guarantee term.

4. The electronic information traffic flow automatic control system according to claim 1 is characterized in that: The intersection execution layer implements tracking of signal cycle and green-to-signal ratio based on fuzzy model predictive control algorithm, including: A three-layer fuzzy prediction and reasoning system is designed to map phase deviation and traffic fluctuation into fuzzy rule confidence. An adaptive domain adjustment mechanism is constructed to dynamically adjust the coverage interval of the fuzzy set according to the traffic state change rate.

5. The electronic information traffic flow automatic control system according to claim 1 is characterized in that: The three-dimensional convolution kernel uses a multi-scale dilated convolution structure to improve feature extraction efficiency, including: The real-time traffic distribution heat map of the road network is divided into multi-level grid units, each of which stores the traffic speed variance and density gradient statistics; In the feature extraction stage, the dilated convolution kernel is used to expand the spatial perception range, and in the feature reconstruction stage, the transposed convolution layer is used to restore local detail features; A spatial attention gating mechanism is introduced to perform dynamic weighted fusion of feature maps.

6. The electronic information traffic flow automatic control system according to claim 2 is characterized in that: The spatial graph convolution layer adopts a dynamic adjacency encoding mechanism, including: Define the dynamic association vector between the intersection node and the adjacent road segments, including vehicle turning probability, lane connection relationship and historical traffic efficiency; The dynamic correlation vector is converted into the bias weight parameters of the graph convolution kernel through a nonlinear mapping layer; The bias weight parameters are integrated into the graph convolution operation process to generate a node feature representation with spatial adaptability.

7. The electronic information traffic flow automatic control system according to claim 1 is characterized in that: The dynamic profit adjustment mechanism is implemented based on an online reinforcement learning strategy, including: Collect the strategy selection records and equilibrium solution distribution in the historical game process as training data sets; Construct a deep Q-learning network to fit the changing patterns of the payoff functions of the game participants; Update network parameters online through asynchronous policy gradient descent method, and adjust the utility function weights of each participant in real time; When an unexpected traffic event is detected, an emergency reconstruction operation of the profit function is triggered.

8. The electronic information traffic flow automatic control system according to claim 3 is characterized in that: The error modeling of the state transfer equation adopts the interval envelope analysis technology, including: modeling the vehicle motion uncertainty as an interval parameter, with the upper and lower bounds determined by the standard deviation of the sensor measurement error; performing orthogonal decomposition on the interval parameter to separate the system deterministic component and the environmental random interference component.

Images