Dynamic traffic dispersing control method and system based on traffic flow situation prediction

Abstract

The application belongs to the technical field of traffic management, and particularly relates to a dynamic dredging control method and system based on traffic flow situation prediction, which comprises the following steps: collecting multiple source physical parameters, reconstructing road node dynamic effective capacity based on the multiple source physical parameters, obtaining a dynamic physical resistance difference value according to the difference between the corresponding physical saturation parameters of adjacent nodes, replacing the physical distance parameters with the dynamic physical resistance difference value to generate a correction weight, inputting a graph adjacency matrix containing the correction weight characteristics and historical traffic flow time series data matrix into a graph convolution calculation layer to output predicted traffic flow, comparing the predicted traffic flow with the dynamic effective capacity to generate a capacity overflow difference value, and issuing a control instruction to execute a flood prevention flow interception control action to prevent road network congestion from spreading. The application can integrate topographic mechanical attenuation characteristics and construct a safety warning mechanism that meets the anti-collision demand.

Description

Technical Field

[0001] This invention relates to the field of traffic management technology. More specifically, this invention relates to a dynamic traffic flow control method and system based on traffic flow trend prediction. Background Technology

[0002] In modern urban traffic management, traffic flow prediction and congestion management in complex areas such as road networks and underpasses and tunnels are crucial for ensuring smooth vehicle traffic. During morning and evening rush hours or in severe weather such as heavy rain, tunnels and intersections are prone to congestion due to traffic accumulation and slippery road surfaces. In order to detect congestion trends in advance and divert vehicles, the industry generally adopts traffic situation prediction and signal management based on historical traffic data and spatiotemporal graph network algorithms.

[0003] Chinese patent document CN112927510B discloses a traffic flow prediction method. It extracts time-series segments from traffic flow data along the time axis, defines the traffic network as an undirected graph, performs graph convolution operations on the graph to obtain spatial relationships between nodes, performs standard convolution operations on the time dimension to obtain temporal relationships, and finally inputs the spatial relationships into a conditional random field layer to obtain the prediction result. However, this patent document mainly filters prediction errors by adding computational layers. When constructing the underlying network topology, it still relies on the absolute physical distance between road nodes, failing to consider the actual increase in vehicle braking distance under severe weather conditions. Therefore, it struggles to accurately reflect the sharp decline in actual road capacity during rainy weather.

[0004] Chinese patent document CN104778837B discloses a multi-timescale prediction method for road traffic operation status. By analyzing the characteristics of road traffic parameters at different time scales, it uses exponential smoothing, weighted average, and Kalman filtering algorithms to predict road traffic operation status at different time scales. However, this patent relies entirely on mathematical deduction of one-dimensional time series, without establishing any spatial topological relationships. It fails to perceive the spatial influence between different intersections and underpasses in urban three-dimensional road networks, resulting in poor prediction performance when dealing with complex traffic congestion spread scenarios.

[0005] A paper titled "Traffic Flow Prediction Based on Dynamic Graph Multi-Time Vision Attention," published in the Journal of Guangdong University of Technology in 2025, presented a traffic flow prediction method. Based on spatiotemporal traffic flow data, it uses a dynamic graph learning module to build a dynamic graph to extract dynamic relationship information between traffic nodes and employs an attention network to capture long-term dependencies in time-series data for prediction. However, this paper primarily relies on the algorithm model to blindly learn the spatial characteristics between traffic nodes, failing to identify the underlying physical causes of node congestion changes. Furthermore, the model still cannot provide accurate early warnings when faced with extreme situations such as heavy rain or downhill sections that are not present in the historical training data.

[0006] In existing technologies, although some solutions attempt to predict traffic flow using multi-layer neural networks, dynamic graph models, or time series algorithms, most of these solutions rely solely on historical data for simple numerical calculations or use fixed geographical distances to determine the degree of influence between road nodes. Because urban road networks, especially underpasses and tunnels, exhibit significant gradients, and heavy rainfall leads to a substantial reduction in tire-to-ground friction, increasing braking distances, the actual number of vehicles a road can accommodate is drastically reduced. This makes it difficult for traditional methods relying on static features to accurately reflect changes in road carrying capacity under the combined effects of weather and terrain. Furthermore, in actual traffic management, traditional congestion mitigation often involves adjusting traffic lights only after queues have formed or traffic flow has exceeded set limits. This passive response strategy, failing to anticipate the reduction in the actual road capacity, easily results in delayed traffic control instructions, ultimately causing congestion and paralysis at urban intersections and tunnels due to reverse traffic overflow. Summary of the Invention

[0007] To address the technical problem that existing traffic flow prediction and diversion technologies fail to effectively integrate the three-dimensional road geometry of cities with the physical attenuation law of underlying forces under extreme weather conditions, resulting in distorted prediction of the effective dynamic traffic capacity of roads in scenarios with overlapping severe terrain and weather, and consequently causing serious delays in traffic diversion and diversion instructions, which can easily lead to overflow deadlock at urban underpasses and ground intersections, this invention provides solutions in the following aspects.

[0008] In a first aspect, the present invention provides a dynamic traffic flow control method based on traffic flow situation prediction, comprising acquiring traffic data, constructing a road network topology and predicting traffic flow, and executing traffic flow control based on the prediction results. The method is characterized by: acquiring real-time traffic velocity, real-time traffic flow, average vehicle wheelbase, real-time road surface friction coefficient, and three-dimensional road longitudinal slope angle; constructing a dynamic effective capacity including average vehicle wheelbase, reaction distance, and braking distance based on the real-time traffic velocity, average vehicle wheelbase, real-time road surface friction coefficient, and three-dimensional road longitudinal slope angle; and calculating the ratio between real-time traffic flow and dynamic effective capacity. Take the physical saturation parameter; obtain the dynamic physical resistance difference based on the difference in the physical saturation parameters of adjacent nodes, and generate the correction weight between adjacent nodes based on the dynamic physical resistance difference; construct a multi-layer superimposed spatiotemporal graph convolutional neural network architecture, and obtain the multi-step predicted traffic flow of each node in the future target time period based on the graph adjacency matrix containing the correction weight and the historical traffic flow time series data matrix; obtain the capacity overflow difference data packet based on the multi-step predicted traffic flow and dynamic effective capacity of each node, obtain the overflow prevention and diversion ratio based on the capacity overflow difference data packet and map it into the phase duration compensation parameter, and perform overflow prevention, interception and diversion control.

[0009] Preferably, obtaining the real-time traffic flow speed, real-time traffic volume, and average wheelbase of the vehicle includes: calling an anisotropic magnetoresistive geomagnetic sensor array, setting a trigger threshold to convert the local magnetic field intensity disturbance of the sensor into a square wave pulse; extracting the real-time traffic volume by accumulating the number of pulse rising edges per unit time; identifying the time difference of the main wave peak representing the front and rear axles in a single pass through the magnetic field waveform, performing a physical product operation with the real-time traffic flow speed, and calculating the average wheelbase of the vehicle.

[0010] Preferably, obtaining the real-time road surface friction coefficient includes: applying an AC sweep voltage to a piezoelectric ceramic element pre-anchored below the asphalt surface layer to excite mechanical vibration; acquiring an impedance frequency response curve and comparing it with a pre-stored dry reference curve to determine the road surface water depth based on the water film thickness impedance mapping relationship; and importing the road surface water depth and real-time traffic flow velocity into a water film friction attenuation model to calculate and extract the real-time road surface friction coefficient.

[0011] Preferably, the step of constructing a dynamic effective capacity based on real-time traffic flow velocity, average vehicle wheelbase, real-time road surface friction coefficient, and three-dimensional road longitudinal slope angle, including the average vehicle wheelbase, reaction distance, and braking distance, includes: obtaining a formula for calculating the average headway, where the average headway is equal to the sum of the average vehicle wheelbase, reaction distance, and braking distance; wherein the reaction distance is the product of real-time traffic flow velocity and reaction time constant; using the reciprocal of the average headway as the traffic density, and multiplying the real-time traffic flow velocity by the traffic density to obtain the dynamic effective capacity.

[0012] Preferably, the dynamic effective capacity includes: In the formula, For dynamic effective capacity; Real-time traffic flow; This represents the average wheelbase of the vehicle. The reaction time constant; The vehicle's reaction distance; This refers to the vehicle's braking distance. Indicates the average distance between vehicle heads; Indicates traffic density.

[0013] Preferably, the braking distance is calculated using the following formula: In the formula, This refers to the vehicle's braking distance. Real-time traffic flow; This refers to the real-time road surface friction coefficient; It is the gravitational acceleration constant; This refers to the longitudinal slope angle of a three-dimensional road.

[0014] Preferably, the construction of a dynamic effective capacity based on real-time traffic flow speed, average vehicle wheelbase, real-time road surface friction coefficient, and three-dimensional road longitudinal slope angle, including the average vehicle wheelbase, reaction distance, and braking distance, further includes: synchronously setting underlying physical boundary conditions, which are triggered by monitoring parameters. When the logical discriminant condition is met, the dynamic effective capacity is directly reduced to zero, and a tunnel access restriction control command is issued; where, This refers to the real-time road surface friction coefficient; This refers to the longitudinal slope angle of a three-dimensional road.

[0015] Preferably, generating corrected weights between adjacent nodes based on the dynamic physical resistance difference includes: In the formula, To adjust the weights; It is a natural exponential function; It is a function for maximizing the value; For downstream nodes Real-time traffic flow; For downstream nodes Dynamic effective capacity; upstream node Real-time traffic flow; upstream node Dynamic effective capacity; Indicates upstream node With downstream nodes The difference in dynamic physical resistance.

[0016] Preferably, the overflow diversion ratio is obtained based on the capacity overflow difference data packet and mapped to a phase duration compensation parameter. The overflow diversion control is then executed, including: extracting and generating a spatial distance array containing the absolute distance from each node to each upstream diversion node based on the geographical coordinate information of each node in the road network topology; backtracking and calling the spatial distance array to allocate the overflow diversion ratio of each upstream ground intersection and elevated off-ramp according to the inverse distance rule; calling the industrial Ethernet communication interface to directly map the overflow diversion ratio to the phase duration compensation parameter of traffic signal control, encapsulating it into a control command message and sending it to the upstream traffic signal controller to execute the red light phase extension action, and synchronously linking the variable message sign at the corresponding ramp entrance to project the diversion map.

[0017] In a second aspect, the present invention provides a dynamic traffic flow situation prediction-based control system, including a processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned dynamic traffic flow situation prediction-based control method is implemented.

[0018] By adopting the above technical solution, the above-mentioned dynamic traffic flow situation prediction-based diversion and control method is generated into a computer program and stored in a memory so that it can be loaded and executed by a processor. In this way, a terminal device can be made based on the memory and the processor for convenient use.

[0019] The beneficial effects of this invention are as follows:

[0020] This invention overcomes the limitations of previous city-level situation prediction models that relied solely on a two-dimensional planar coordinate system. By leveraging spatial mechanics analysis and vehicle kinetic energy conservation deduction, it maps the increased braking distance caused by rainfall and road slope into a quantitative attenuation characteristic of road carrying capacity. This allows the graph convolutional computation layer to abandon fixed threshold constraints and establish the early warning triggering mechanism based on the real microscopic laws of wheel-to-ground frictional impedance attenuation. This feature extraction method uses physical parameter dimensionality reduction to replace massive amounts of sudden weather training samples, enabling traffic control systems to generate reliable early warning judgments based on three-dimensional spatial physical laws calculation models even when facing heavy rain combined with three-dimensional elevation differences. Simultaneously, it utilizes… The difference in physical saturation parameter replaces the physical distance parameter in the spatial feature extraction formula of the existing spatiotemporal graph convolutional network, giving the graph adjacency matrix the ability to actively perceive downstream microscopic physical attenuation. This allows the updated network layer to extract congestion reverse propagation features in advance on the feature aggregation path, mitigating the impact of sudden weather changes on the prediction algorithm. Finally, based on comparing the capacity overflow difference data packets, a forward-defense type of road network traffic interception is implemented, avoiding the feedback delay defects caused by passive diversion based on fixed queue lengths. By using computational analysis and physical boundary dual verification instructions to replace the experience-based control scheme, the injected traffic is restricted before road nodes become overloaded, ensuring the operational order of the city's core three-dimensional transportation. Attached Figure Description

[0021] Figure 1 This is a flowchart illustrating the dynamic traffic flow situation prediction-based traffic management and control method of the present invention. Detailed Implementation

[0022] 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, not all, of the embodiments of the present invention. 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.

[0023] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0024] This invention discloses a dynamic traffic flow pattern prediction-based diversion and control method, referring to... Figure 1 This includes steps S1-S5:

[0025] S1. Obtain real-time traffic flow speed, real-time traffic volume, current average vehicle wheelbase, real-time road surface friction coefficient, and three-dimensional road longitudinal slope angle.

[0026] It should be noted that, in response to the interaction between the confined space of the underpass tunnel and the meteorological environment, the traffic situation simulation of this invention needs to be based on the physical environmental interference baseline. Since the road surface is slippery in rainy weather, the tire grip decreases, and mechanical attenuation features cannot be extracted by image recognition. Therefore, this invention reduces the priority of panoramic video image data transmission, calls an anisotropic magnetoresistive geomagnetic sensor array and non-invasive road weather station equipment, and extracts and integrates physical attenuation features into the motion boundary assessment system.

[0027] Real-time traffic flow velocity is obtained by using an anisotropic magnetoresistive geomagnetic sensor array. With real-time traffic flow and the average wheelbase of the vehicle Specifically, the hardware topology includes dual geomagnetic nodes buried at fixed intervals along the longitudinal direction of the lane. When the stable geomagnetic field on the Earth's surface is disturbed by the ferromagnetic material of the vehicle chassis, the local magnetic field of the sensor is distorted and outputs a corresponding voltage disturbance waveform. By setting a trigger threshold, the magnetic field intensity disturbance is converted into a square wave pulse, and then the real-time traffic flow is extracted by accumulating the number of pulse rising edges per unit time. Due to the differences in ferromagnetic mass distribution across various parts of the vehicle, multiple local peaks appear in the waveform of a single passage through the magnetic field. Here, the time difference of the main peak representing the front and rear axles is identified and correlated with the real-time traffic flow velocity. The average wheelbase of the vehicle is calculated by performing a physical product operation. .

[0028] Furthermore, non-intrusive road weather station equipment was used to obtain the real-time road friction coefficient. Specifically, an AC sweep voltage is applied to a piezoelectric ceramic element pre-anchored beneath the asphalt surface to excite mechanical vibration. Because surface water acts as an additional mass layer adhering to the sensor surface, the resonant frequency of the piezoelectric ceramic shifts to a lower frequency and the impedance peak decreases. By acquiring the impedance frequency response curve and comparing it with a pre-stored dry baseline curve, the surface water depth can be calibrated based on the water film thickness impedance mapping relationship. This surface water depth is then correlated with real-time traffic flow velocity. Import the water film friction decay model and calculate and extract the real-time road surface friction coefficient. Simultaneously, a request to access the coordinates of the detection nodes is sent to the map server, and the returned data packet is parsed to obtain the corresponding three-dimensional road longitudinal slope angle. .

[0029] S2. Construct a dynamic effective capacity that includes the vehicle's average wheelbase, reaction distance, and braking distance.

[0030] It should be noted that when faced with the superposition of road surface water and road longitudinal slope, the vehicle braking performance deteriorates, leading to an increased risk of collision. This invention replaces the original front-end spacing parameter with the sum of vehicle size parameters and safe braking distance, and adjusts the theoretical maximum traffic flow parameter of a single lane downward to construct a space margin that meets the collision avoidance requirements. This parameter configuration rule effectively reduces the probability of rear-end collisions caused by static distance calculations exceeding the braking limit, converts static capacity into dynamic effective capacity boundaries, and ensures that the prediction model conforms to kinematic constraints.

[0031] Specifically, the basic theory of macroscopic traffic flow is introduced to define dynamic effective capacity. Equal to real-time traffic flow speed Traffic density The physical product; extracting the average length occupied by vehicles arranged continuously in physical space as the average headway. Set traffic density The value is the average front-to-back distance. The reciprocal of, that is Substituting the above reciprocal relationship into the basic calculation formula completes the conversion of basic flow rate to spacing, yielding the simplified dynamic effective capacity. The form:

[0032]

[0033] In the formula, For dynamic effective capacity; Real-time traffic flow; This represents the average distance between the front ends of the vehicles.

[0034] The logic behind this calculation lies in the average headway. Smaller traffic density The larger the value, the better for the same real-time traffic flow rate. The higher the number of vehicles that can pass through per unit of time, the greater the dynamic effective capacity. The larger the capacity, the more it realizes the transformation from micro-vehicle spacing to macro-lane capacity.

[0035] Furthermore, a limit safety boundary model was established to explore the hazards of gravitational potential energy in underpass ramps, and the physical slippage caused by rainwater accumulation was incorporated into the average headway. In the computational system, the geometric constraint space is specified to include the sum of three independent physical lengths: the vehicle's own dimensions, the driver's reaction coasting area, and the mechanical braking coasting area. Therefore, obtaining the average front-end distance is crucial. The calculation formula:

[0036]

[0037] In the formula, This represents the average distance between the front ends of the vehicles. This represents the average wheelbase of the vehicle. The vehicle's reaction distance; This refers to the vehicle's braking distance.

[0038] Among them, the reaction time constant is set. At this time, vehicles maintain real-time traffic flow speed. If the displacement is uniform linear motion, then the reaction distance is calculated. The calculation formula:

[0039]

[0040] In the formula, The vehicle's reaction distance; Real-time traffic flow; is the reaction time constant.

[0041] It should be further added that the reaction time constant... The reaction time constant, used to characterize the sum of physiological and physical delays from the driver's visual perception of an abnormality ahead to the initial establishment of pressure by the mechanical braking system, is defined according to highway design specifications for conventional road sections. The baseline value is set between 2.0s and 2.5s. In this embodiment, it is set to 2.5s to meet the calculation requirements of the system response.

[0042] Among them, the vehicle mass is set. and the gravitational acceleration constant Braking distance is calculated based on space mechanics and the kinetic energy theorem. Physical derivation: When the controlled vehicle is in a position carrying the three-dimensional longitudinal slope angle of the road When performing emergency braking under dangerous downhill conditions, the effective resistance that the vehicle experiences along the tangential direction of the road surface is... It includes two independent vectors: the frictional resistance derived from the road surface normal pressure and the downward sliding component of gravity along the slope. Therefore, the vertical gravity component of the road surface and the real-time road surface friction coefficient are extracted. The physical product constructs the positive braking work term, while subtracting the negative sliding effect caused by the gravity component parallel to the downhill slope, yielding the force balance calculation formula:

[0043]

[0044] In the formula, As an effective resistance; This refers to the real-time road surface friction coefficient; For vehicle quality; It is the gravitational acceleration constant; This refers to the longitudinal slope angle of a three-dimensional road.

[0045] According to the work-energy theorem, the work done by the effective resistance against displacement during vehicle braking is equal to the amount of kinetic energy lost by the vehicle. Therefore, the effective resistance... Braking distance The physical product is used as the total work done, and is established to equal the vehicle mass. With real-time traffic flow The kinetic energy theorem equation, which is half the product of squares, is as follows:

[0046]

[0047] In the formula, As an effective resistance; This refers to the vehicle's braking distance. For vehicle quality; This represents real-time traffic flow speed.

[0048] Furthermore, within the physical derivation framework of kinetic energy loss, an independent rigid body dynamic model of the vehicle mass is established to isolate the complex suspension deformation dissipation factors: assuming that the mechanical braking system pressure instantaneously reaches its peak state, causing complete wheel lock-up and slippage, the vehicle resists forward kinetic energy with pure sliding friction resistance, without considering the braking compensation from the anti-lock braking system. A simultaneous substitution calculation is performed to determine the effective resistance... The polynomial characteristic is replaced on the left side of the kinetic energy theorem equation, and the vehicle mass is simultaneously stripped from both sides of the equation. The parameters are then evaluated and algebraically simplified to obtain the braking distance formula that integrates the longitudinal slope angle and the road surface friction coefficient:

[0049]

[0050] In the formula, This refers to the vehicle's braking distance. Real-time traffic flow; This refers to the real-time road surface friction coefficient; The acceleration due to gravity is constant; in this embodiment, it is taken as a value of . ; This refers to the longitudinal slope angle of a three-dimensional road.

[0051] The calculation logic for this braking distance formula lies in the real-time road friction coefficient. Reduce or decrease the longitudinal slope angle of the three-dimensional road Increasing the value of the denominator results in a decrease in the calculated braking distance. An increase in the value of the tire grip indicates a significant increase in the vehicle's gliding distance in wet or downhill conditions, reflecting the non-linear decay of tire grip.

[0052] It should be further explained that, for the extreme state of frictional resistance attenuation caused by water accumulation, the underlying physical boundary conditions are simultaneously set, and when the monitoring parameters are triggered... When the logical discriminant condition is met, it indicates that the friction provided by the tires is insufficient to counteract the downhill component of gravity. This parameter mapping suggests that the vehicle's mechanical braking has failed, at which point the dynamic effective capacity will be... The system is set to zero directly, without any subsequent calculations or judgments, and a tunnel traffic restriction control command is issued directly.

[0053] Furthermore, under normal natural decay conditions, the obtained response distance... Braking distance with terrain-constrained features Substituting these values ​​into the original formula for calculating the average frontage, we obtain the reconstructed average frontage, which includes the reaction lag distance and braking slip distance. Substitute the recombined average headway equation back into the dynamic effective capacity equation to obtain the dynamic effective capacity. The calculation formula:

[0054]

[0055] In the formula, For dynamic effective capacity; Real-time traffic flow; This represents the average wheelbase of the vehicle. The reaction time constant; It is the gravitational acceleration constant; This refers to the real-time road surface friction coefficient; This refers to the longitudinal slope angle of a three-dimensional road.

[0056] The calculation logic of this dynamic effective capacity formula lies in the fact that the increase in road surface water depth causes a decrease in the real-time road surface friction coefficient. The decrease in magnitude leads to a smaller difference between the physical friction force and the gravitational component, causing the denominator to expand and increase overall, directly affecting the dynamic effective capacity. The numerical nonlinear decay decreases; this process accurately maps the mechanism of tunnel congestion and queuing overflow in the physical world, transforming the change in road segment geometry into a quantitative reduction in carrying capacity.

[0057] S3. Calculate the physical saturation parameter based on real-time traffic flow and dynamic effective capacity, and generate corrected weights based on the physical saturation parameters of adjacent nodes.

[0058] It should be noted that existing spatiotemporal graph convolutional neural networks ignore the heterogeneity of node bearing capacity. Conventional graph convolutional layers, which only use Euclidean distance to evaluate node correlation, are prone to errors. For example, when faced with traffic disruption caused by tunnel flooding, the physical distance between upstream and downstream nodes remains unchanged, but the downstream node's ability to accept vehicles has been significantly reduced. Therefore, this invention reduces the initial convergence speed feature of the road network graph topology and replaces the physical distance parameter with dynamic physical resistance difference, thereby expanding the model's feature extraction range from two-dimensional coordinate parameters to three-dimensional force and blockage parameters, giving the road network graph topology the characteristic of perceiving downstream impedance backtracking.

[0059] Specifically, based on the acquired dynamic effective capacity Select any node in the road network topology and extract the real-time traffic flow of that node. With the corresponding dynamic effective capacity Perform a division operation to obtain the physical saturation parameter of the node. The calculation formula:

[0060]

[0061] In the formula, This is a parameter for physical saturation. Real-time traffic flow; This refers to the dynamic effective capacity.

[0062] Furthermore, regarding the upstream nodes among adjacent nodes in the road network topology... and downstream nodes Extract upstream nodes Real-time traffic flow With upstream nodes Dynamic effective capacity The upstream node is calculated. Physical saturation parameter Simultaneously extract downstream nodes Real-time traffic flow With downstream nodes Dynamic effective capacity The downstream node was calculated. Physical saturation parameter ; downstream nodes Physical saturation parameter minus upstream node The physical saturation parameter is combined with a maximum value function to preserve the positive blocking overflow effect value, thus obtaining the dynamic physical drag difference. The calculation formula:

[0063]

[0064] In the formula, This represents the difference in dynamic physical resistance. It is a function for maximizing the value; For downstream nodes Real-time traffic flow; For downstream nodes Dynamic effective capacity; upstream node Real-time traffic flow; upstream node Dynamic effective capacity.

[0065] Existing spatiotemporal graph convolutional networks extract spatial feature dependency graph adjacency matrices, and the basic weights are calculated using the following formula:

[0066]

[0067] In the formula, Indicates upstream node With downstream nodes The basic weights between them; Indicates upstream node With downstream nodes Physical distance between them; This represents the distance attenuation constant.

[0068] Furthermore, a mathematical replacement procedure for the road network topology is executed, replacing the physical distance parameter in the calculation formula of the basic weights on which the spatial features are extracted from the existing spatiotemporal graph convolutional network with the derived dynamic physical resistance difference. Generate corrected weights to address terrain variations. The formula for calculation is:

[0069]

[0070] In the formula, To adjust the weights; It is a natural exponential function; It is a function for maximizing the value; For downstream nodes Real-time traffic flow; For downstream nodes Dynamic effective capacity; upstream node Real-time traffic flow; upstream node Dynamic effective capacity.

[0071] The calculation logic for this corrected weight is based on the fact that when the underpass encounters water accumulation and causes downstream nodes... Dynamic effective capacity During reduction, nodes An increase in the physical saturation parameter leads to a larger difference in dynamic physical drag, which, through a negative exponential function mapping, results in a calculated correction weight. The weights become smaller; this means that the upstream graph convolutional feature extraction layer obtains the downstream blocking features in advance, thereby weakening the connectivity weights and realizing the dynamic adjustment mechanism of the edge weights of the road network graph topology by abnormal meteorological elements.

[0072] S4. Construct a multi-layered spatiotemporal graph convolutional neural network architecture, and obtain multi-step predicted traffic flow for each node in the future target time period based on the graph adjacency matrix containing the corrected weights and the historical traffic flow time series data matrix.

[0073] It should be noted that existing spatiotemporal graph convolutional neural networks suffer from a severe decline in prediction accuracy when training with a lack of extreme meteorological disaster samples. The architecture design of this invention transforms meteorological mechanics rules into the underlying topological weight features of the graph structure. When sudden changes in meteorological conditions occur, the information transmission path can be automatically adjusted according to the changes in the graph structure weights. Physical laws are used to supplement the training data features, thereby significantly improving the robustness of the model's predictions in extreme scenarios.

[0074] Specifically, a multi-layered spatiotemporal graph convolutional neural network architecture is constructed; the graph adjacency matrix corresponding to the modified weight features is used as the spatial information propagation skeleton, and each detection node transmits historical traffic flow data to establish a historical traffic flow time series data matrix; wherein, the dimension of the historical traffic flow time series data matrix is ​​defined as a three-dimensional tensor. In the formula, This represents the total number of detection nodes in the road network topology. This represents the number of time steps contained within the historical time sliding window. In this embodiment, the feature dimension of the node is... That is, only traffic flow is used as the feature input; the historical traffic flow time series data matrix and the graph adjacency matrix are jointly input into the graph convolutional computation layer to perform feature aggregation operation, and then the time-dependent evolution logic is handled by the hidden layer state update mechanism of the gated recurrent unit; in order to fully extract complex spatiotemporal features, the specific architecture of the multi-layer stacked spatiotemporal graph convolutional neural network is as follows: graph convolutional layers and gated recurrent units (GRUs) are alternately stacked to form a spatiotemporal feature extraction module (ST-Block). In this embodiment, a total of 2 such spatiotemporal feature extraction modules are stacked; in terms of specific hyperparameters and implementation details, the graph convolution order of the graph convolutional layer is Chebyshev multinomial. The model order is set to 2, the number of hidden units in the gated recurrent unit is set to 64, and the ReLU activation function is used. During model training, mean squared error (MSE) is used as the loss function, an alternating training method is employed, and the Adam optimizer is used for parameter updates. The initial learning rate is set to 0.001, the batch size is 32, and the number of training epochs is 100. Finally, a fully connected layer outputs the multi-step predicted traffic flow for each node within the future target time period. In this embodiment, the prediction time scale (sampling and prediction step size) is set to 5 minutes, and the future target time period is set to the next 30 minutes. Multi-step prediction is used to provide a time base for subsequent capacity overflow calculations, and the final predicted tensor output by the fully connected layer has a dimension of [missing value]. .

[0075] S5. Based on the multi-step predicted traffic flow of each node in the future target time period and the dynamic effective capacity of the corresponding node, obtain the capacity overflow difference data packet, then obtain the anti-overflow diversion ratio and map it into the phase duration compensation parameter, and execute the anti-overflow interception and diversion control.

[0076] It should be noted that traditional traffic management strategies lag far behind the on-site queue length alarm mechanism, resulting in a failure to effectively prevent the spread of congestion. This invention relies on the aforementioned dynamic effective capacity to obtain early warning time windows in advance. Before the tunnel's carrying capacity reaches the critical value, it restricts the right-of-way by lowering the local traffic efficiency index of the upstream intersection, ensuring the traffic stability of the core section of the underpass tunnel and effectively transforming post-event traffic management into pre-emptive interception and control.

[0077] Specifically, the multi-step predicted traffic flow for each node within a future target time period is extracted and compared with the corresponding node's dynamic effective capacity using an algebraic difference operation. When the multi-step predicted traffic flow exceeds the dynamic effective capacity, the difference between the multi-step predicted traffic flow and the dynamic effective capacity is used as the capacity overflow difference, and a capacity overflow difference data packet is output. The capacity overflow difference data packet specifically includes the following fields: the target node ID where the capacity overflow occurred, the unified timestamp of the data packet generation, the specific value of the capacity overflow difference, and the emergency warning level label field. After generating the data packet, each node calls the communication interface of the roadside edge computing device, based on M... The QTT IoT communication protocol transmits and aggregates the aforementioned data packets to the cloud server of the regional traffic control center for global overall calculation. Based on the geographical coordinates of each node in the road network topology, it extracts and generates a spatial distance array containing the absolute distances from each node to its upstream diversion nodes. It then backtracks and calls this spatial distance array to allocate the overflow diversion ratios for each upstream ground intersection and elevated off-ramp according to an inverse distance ratio rule. Finally, it calls the industrial Ethernet communication interface to directly map these overflow diversion ratios to the phase duration compensation parameters of the traffic signal control. This direct mapping is specifically implemented through a mapping function, which is as follows: In the formula, The upstream generation of the mapping Compensation parameters for the duration of red light phase in traffic signal control at each node. To allocate to the first according to the inverse distance rule The overflow prevention and diversion ratio of each node, For the first The basic green light duration for each node within the current signal control cycle. To control the intensity adjustment coefficient, this embodiment sets , This represents the function of rounding down; the phase duration compensation parameter is encapsulated into a control command message and sent to the upstream traffic signal controller to execute the red light phase extension action, and the variable message sign at the corresponding ramp entrance is simultaneously linked to project the diversion map, thereby completing the entire process of overflow prevention, diversion and diversion control.

[0078] This invention also discloses a dynamic traffic flow situation prediction-based control system, including a processor and a memory. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the dynamic traffic flow situation prediction-based control method according to this invention is implemented.

[0079] The system also includes other components well known to those skilled in the art, such as communication buses and communication interfaces, the settings and functions of which are known in the art and will not be described in detail here.

Claims

1. A dynamic traffic flow pattern prediction-based traffic management and control method, comprising acquiring traffic data, constructing a road network topology and predicting traffic flow, and executing traffic management and control based on the prediction results, characterized in that, include: Obtain real-time traffic flow speed, real-time traffic volume, average wheelbase of vehicles, real-time road surface friction coefficient, and three-dimensional road longitudinal slope angle; Obtaining the real-time road surface friction coefficient includes: applying an AC sweep voltage to a piezoelectric ceramic element pre-anchored below the asphalt surface layer to excite mechanical vibration; acquiring an impedance frequency response curve and comparing it with a pre-stored dry reference curve to determine the road surface water depth based on the water film thickness impedance mapping relationship; and importing the road surface water depth and real-time traffic flow velocity into a water film friction attenuation model to calculate and extract the real-time road surface friction coefficient. Based on real-time traffic flow velocity, average vehicle wheelbase, real-time road surface friction coefficient, and three-dimensional road longitudinal slope angle, a dynamic effective capacity is constructed, including the average vehicle wheelbase, reaction distance, and braking distance. In the formula, For dynamic effective capacity; Real-time traffic flow; This represents the average wheelbase of the vehicle. The reaction time constant; The vehicle's reaction distance; This is the vehicle's braking distance, and ; This refers to the real-time road surface friction coefficient; It is the gravitational acceleration constant; The longitudinal slope angle of the three-dimensional road; Indicates the average distance between vehicle heads; Indicates traffic density; Calculate the ratio between real-time traffic flow and dynamic effective capacity to obtain the physical saturation parameter; obtain the dynamic physical resistance difference based on the difference in the physical saturation parameters of adjacent nodes, and generate the correction weight between adjacent nodes based on the dynamic physical resistance difference. A multi-layered spatiotemporal graph convolutional neural network architecture is constructed. Based on the graph adjacency matrix with corrected weights and the historical traffic flow time series data matrix, the multi-step predicted traffic flow of each node in the future target time period is obtained. Based on the multi-step predicted traffic flow and dynamic effective capacity of each node, a capacity overflow difference data packet is obtained. Based on the capacity overflow difference data packet, the overflow prevention and diversion ratio is obtained and mapped to the phase duration compensation parameter, and overflow prevention, interception and diversion control is performed.

2. The dynamic traffic flow situation prediction-based traffic management and control method according to claim 1, characterized in that, Obtaining the real-time traffic flow speed, real-time traffic volume, and average wheelbase of vehicles includes: The anisotropic magnetoresistive geomagnetic sensor array is invoked, and a trigger threshold is set to convert the local magnetic field intensity disturbance of the sensor into a square wave pulse; Real-time traffic flow is extracted by accumulating the number of pulse rising edges per unit time. Identify the time difference between the main wave peaks representing the front and rear axles in a single pass through the magnetic field waveform, perform a physical product operation with the real-time traffic flow velocity, and calculate the average wheelbase of the vehicle.

3. The dynamic traffic flow situation prediction-based diversion and control method according to claim 1, characterized in that, The dynamic effective capacity, constructed based on real-time traffic flow velocity, average vehicle wheelbase, real-time road surface friction coefficient, and three-dimensional road longitudinal slope angle, including the average vehicle wheelbase, reaction distance, and braking distance, also includes: Synchronously set the underlying physical boundary conditions, when the monitoring parameters are triggered. When the logical discriminant condition is met, the dynamic effective capacity is directly reduced to zero, and a tunnel access restriction control command is issued; where, This refers to the real-time road surface friction coefficient; This refers to the longitudinal slope angle of a three-dimensional road.

4. The dynamic traffic flow situation prediction-based diversion and control method according to claim 1, characterized in that, Based on the dynamic physical resistance difference, a corrected weight is generated between adjacent nodes, including: ； In the formula, To adjust the weights; It is a natural exponential function; It is a function for maximizing the value; For downstream nodes Real-time traffic flow; For downstream nodes Dynamic effective capacity; upstream node Real-time traffic flow; upstream node Dynamic effective capacity; Indicates upstream node With downstream nodes The difference in dynamic physical resistance.

5. The dynamic traffic flow situation prediction-based diversion and control method according to claim 1, characterized in that, Based on the capacity overflow difference data packet, the overflow prevention diversion ratio is obtained and mapped to a phase duration compensation parameter. Overflow prevention interception and diversion control is then performed, including: Based on the geographic coordinates of each node in the road network topology, extract and generate a spatial distance array containing the absolute distances from each node to each upstream branch node; The spatial distance array is called back, and the overflow diversion ratio of each upstream ground intersection and elevated off-ramp is allocated according to the inverse distance ratio rule; The overflow diversion ratio is directly mapped to the phase duration compensation parameter of the traffic signal control by calling the industrial Ethernet communication interface. It is then encapsulated into a control command message and sent to the upstream traffic signal controller to execute the red light phase extension action, and synchronously linked with the variable message sign at the corresponding ramp entrance to project the diversion map.

6. A dynamic traffic flow pattern prediction-based traffic management and control system, characterized in that, include: A processor and a memory, wherein the memory stores computer program instructions that, when executed by the processor, implement the dynamic traffic flow situation prediction-based control method according to any one of claims 1-5.

Images