A Highway Lane Control Method and System Based on Real-Time Traffic Flow

Abstract

The present invention discloses a highway lane control method based on real-time traffic flow, comprising the following steps: fusing and sensing through a distributed fiber optic vibration sensor array and a millimeter wave radar; constructing a spatio-temporal graph convolutional network model to dynamically map the road topology into a lane connection graph with variable weights; adopting a hierarchical reinforcement learning architecture to trigger an emergency lane dynamic management mode; deploying a digital twin sand table for multi-resolution simulation verification, and when the actual traffic flow deviates from the predicted value by more than 15%, starting a causal inference engine to trace back the decision-making link and update the agent policy library. The present invention realizes global dynamic lane optimization through multi-source perception fusion and a hierarchical reinforcement learning architecture. The simulation data shows that the morning peak traffic volume is increased by 29.3%, and the traffic efficiency in accident scenarios is improved by 28.6%; by adopting digital twin and a causal inference engine, the policy update response time is shortened to 30 seconds, and the congestion propagation distance is reduced by 35%.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent traffic management, and specifically provides a highway lane control method and system based on real-time traffic flow. Background Art

[0002] Traditional highway lane control mostly uses fixed speed limit signs or manual intervention, which has the following defects:

[0003] Response lag: Static strategies based on historical data cannot respond to sudden traffic changes in real time (such as accidents, holiday peaks).

[0004] Insufficient global coordination: Isolated regulation of a single lane is likely to cause downstream chain congestion (see the comparative document CN114170793A).

[0005] Inefficient emergency resources: The emergency lane is often illegally occupied, and the false alarm rate of traditional microwave detection is high (comparative document CN115512539A).

[0006] The prior art such as CN113870562B proposes emergency lane control based on queue length, but does not solve the problems of multi-lane coordination and long-term speed stability.

[0007] The prior art relies on a single sensor or static strategy, while the present invention realizes global dynamic optimization through multi-source perception fusion (vibration + radar) and a hierarchical reinforcement learning architecture (long-term prediction + real-time regulation);

[0008] In the emergency lane management of the present invention, a laser clearance scanner (infrared thermal imaging) and a causal inference engine are introduced to solve the problem of high false alarm rate of traditional microwave detection. Summary of the Invention

[0009] The purpose of the present invention is to provide a highway lane control method and system based on real-time traffic flow to solve the problems raised in the above background art.

[0010] To achieve the above object, the present invention provides the following technical solutions: A highway lane control method based on real-time traffic flow, comprising the following steps: (a) Through the fusion perception of a distributed fiber optic vibration sensor array and a millimeter wave radar, the vehicle trajectories, vehicle type distributions, and vibration anomaly events of the entire road section are obtained in real time, and a multi-modal traffic situation matrix is generated; (b) Construct a spatio-temporal graph convolutional network model, dynamically map the road topology structure into a lane connection graph with variable weights, and update the edge weights between nodes based on the real-time traffic flow gradient; (c) Adopt a hierarchical reinforcement learning architecture, the upper-level meta-agent predicts the optimal lane configuration strategy set within the next 1-6 hours according to the historical traffic pattern, and the lower-level lane agent adjusts the combination of lane function modes and speed limit values at a 30-second time granularity based on the Q-learning algorithm; (d) Trigger the dynamic management mode of the emergency lane. When the average speed of the main road is lower than 25 km / h and lasts for more than 5 minutes, and there is no accident within 500 meters downstream, activate the laser clearance monitoring device and open the emergency lane; (e) Deploy a digital twin sand table for multi-resolution simulation verification. When the actual traffic flow deviates from the predicted value by more than 15%, start the causal inference engine to trace back the decision link and update the agent strategy library. The causal inference engine uses a Bayesian network model to trace back the abnormal nodes according to the historical decision path and real-time data. The nodes include the historical decision path, real-time traffic flow deviation, and external events such as weather, and the edge weights are dynamically updated by the conditional probability table.

[0011] The method for generating the multi-modal traffic situation matrix in the step (a) includes: extracting the energy characteristics of the 0.5-5 Hz frequency band in the vibration signal through wavelet transform, and calculating the anomaly event confidence of each lane segment according to the following formula:

[0012] Confidence=

[0013] Where: represents the energy integral value of the high-frequency band (3-5 Hz), represents the energy integral value of the low-frequency band (0.5-2 Hz), is a smoothing coefficient, and its value is 0.01;

[0014] When the confidence exceeds 0.7, it is marked as a potential anomaly event and triggers the directional capture of the pan-tilt camera.

[0015] The reward function of the hierarchical reinforcement learning in the step (c) is designed as:

[0016]

[0017] Where: is a weight coefficient, is the travel time error weight coefficient, is the congestion propagation attenuation coefficient, is the speed limit deviation penalty coefficient; satisfying , , , , represents the downstream congestion entropy value, is the actual travel time, in seconds, calculated based on real-time traffic flow, is the predicted travel time, in seconds, generated by the historical data model of the upper-layer meta-agent, is the set speed limit, is the actual average vehicle speed, is the congestion entropy attenuation coefficient, used to adjust the influence weight of the congestion entropy value on the reward, with a value range of [0.1, 1.0]. After 500 Monte Carlo simulations, the weight coefficients 0.3, 0.5, and 0.2 can reduce the travel time error by 18% and reduce congestion propagation by 35% through simulation verification;

[0018] When the emergency lane occupancy is detected, a penalty term is introduced:

[0019]

[0020] Among them: is the penalty factor, with a value range of 0.1 - 0.5, is the occupancy duration, is the maximum allowable occupancy threshold, set to 30 minutes.

[0021] The calculation method of the downstream congestion entropy value includes:

[0022] The road is divided into road units at intervals of 500 meters. First, calculate the speed variation coefficient of each unit:

[0023]

[0024] Among them: refers to the set of instantaneous speeds of all vehicles in the i-th road unit, is the average value of the instantaneous speeds of vehicles in the i-th road unit, with a sampling window of 30 seconds, is the standard deviation of the speeds in the ii-th road unit, calculated based on the same time window;

[0025] Based on the information entropy theory, a congestion propagation model is constructed:

[0026]

[0027] Among them: represents the total length of the influence area, which is the sum of the length of the current section and the length of the associated sections within 2 kilometers downstream. When When it is triggered, it conducts coordinated control of the entire road network. A low CV value, such as 0.05, indicates small speed fluctuations. For example, if all vehicles maintain a speed of 50 ± 2 km / h, it means the traffic flow is stable and close to the free flow state. A high CV value, such as 0.3, indicates large speed differences. For example, if the vehicle speeds vary drastically between 20 - 80 km / h, it indicates the existence of anomalies such as congestion, accidents, or frequent lane changes.

[0028] A highway lane control system based on real - time traffic flow, comprising:

[0029] Perception module:

[0030] Optical fiber vibration sensors, millimeter - wave radars, and edge computing units;

[0031] The edge computing unit is data - connected to the millimeter - wave radar and the optical fiber vibration sensor through a 5G - V2X communication module;

[0032] The edge computing unit uses NVIDIA Jetson AGX Orin (computing power 275 TOPS) and supports real - time processing of multi - sensor data; Execution module: Electromagnetic road stud array; Decision - making module: Federated learning server cluster, adopting a differential privacy aggregation strategy; Verification module: Laser clearance scanner and variable speed limit sign. The laser clearance scanner (wavelength 905 nm) integrates an infrared thermal imaging module, can distinguish stationary obstacles and pedestrians, and the false detection rate is <0.1%.

[0033] The electromagnetic road stud array includes:

[0034] Magnetic adsorption unit, the adjustable range of magnetic field strength is 0.1 - 1.5 Tesla. When the magnetic field strength is 1.0 Tesla, it can resist the action of an 8 - level wind and maintain the stability of the road stud position; Multi - spectral LED unit, including dual - band of 590nm amber and 630nm red, and the color gamut covers 98% of sRGB; Anti - collision warning module, when the vehicle crossing the line speed > 20 km / h, it triggers an audible and visual alarm. The audible and visual alarm module supports decibel adjustment (80 - 105 dB), and the default setting is 90 dB.

[0035] Compared with the prior art, the beneficial effects of the present invention are:

[0036] The present invention realizes global dynamic lane optimization through multi - source perception fusion and hierarchical reinforcement learning architecture. The simulation data shows that the morning peak traffic volume increases by 29.3%, and the traffic efficiency in accident scenarios increases by 28.6%; By adopting digital twin and causal inference engine, the strategy update response time is shortened to 30 seconds, and the congestion propagation distance is reduced by 35%. Description of the Drawings

[0037] Figure 1 It is a schematic diagram of the method steps of the present invention;

[0038] Figure 2 Schematic diagram of the system framework of the present invention;

[0039] Figure 3 Schematic diagram of the system process of the present invention;

[0040] Figure 4 Data fusion flow chart of the present invention. Specific implementation manners

[0041] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0042] Please refer to Figures 1 - 4 , the present invention provides a technical solution: a highway lane control method and system based on real-time traffic flow.

[0043] Embodiment 1:

[0044] Multi-source perception data fusion

[0045] Data acquisition:

[0046] Distributed fiber optic vibration sensors (model OSEN-FT-01) are deployed every 20 meters along the central median strip, with a sampling rate of 1 kHz and a sensitive frequency band of 0.5 - 5 Hz. Vehicle types (such as trucks, cars), vehicle speeds, and abnormal events (such as sudden braking, collisions) are identified by detecting ground vibration signals;

[0047] A 79 GHz millimeter-wave radar is installed on the roadside, with a horizontal field of view of ±60° and a detection distance of 200 m, and target coordinates and speed information are output.

[0048] The vibration signal processing process performs discrete wavelet decomposition on the fiber optic vibration signal: selects the Daubechies 4 wavelet basis function; decomposition level: 3 layers;

[0049] Approximation coefficients cA and detail coefficients cD are obtained;

[0050] Calculate the high-frequency energy ;

[0051] Calculate the low-frequency energy ;

[0052] Calculate the anomaly confidence level according to the formula:

[0053]

[0054] When the confidence level > 0.7, trigger the pan-tilt camera to capture and save images, and verify abnormal events (such as accidents, breakdowns).

[0055] Radar data processing flow:

[0056] Perform spatio-temporal alignment on the radar point cloud data to generate a vehicle trajectory point set ;

[0057] Use the density clustering algorithm (DBSCAN) for trajectory classification. The Euclidean distance formula of the density clustering algorithm:

[0058]

[0059] Neighborhood radius ;

[0060] The minimum number of neighborhood points minPts = 5;

[0061] Output the number of vehicles, average speed, and acceleration distribution in each lane.

[0062] The data fusion logic establishes vibration-radar association rules:

[0063]

[0064] Data association rules:

[0065] Align the vibration signal and the radar trajectory spatio-temporally, and establish a vibration-radar association matrix (such as matching the abnormal vibration signal area with the vehicle emergency braking position detected by the radar).

[0066] Generate a multi-modal traffic situation matrix, including the following dimensions:

[0067] Lane-level traffic density;

[0068] Vehicle type distribution ratio (truck ratio, small car ratio);

[0069] Abnormal event location and confidence level;

[0070] Real-time speed gradient (upstream and downstream speed difference).

[0071] Dynamic lane topology modeling and traffic flow prediction

[0072] Spatio-temporal graph convolutional network (ST-GCN):

[0073] Map the road topology structure into a graph structure, where the nodes represent lane units and the edges represent lane connection relationships.

[0074] Dynamic weight update: Adjust the edge weights according to the real-time traffic flow gradient (such as a sudden increase in upstream traffic). The formula is:

[0075]

[0076] Among them, is the traffic flow gradient, is the lane connection edge;

[0077] Output: A lane connection graph with variable weights, identifying high congestion risk areas.

[0078] Hierarchical reinforcement learning decision-making:

[0079] Upper-level meta-agent:

[0080] Input:

[0081] Historical traffic patterns (such as weekday morning peak traffic flow), weather data, event calendar.

[0082] Model:

[0083] Based on the LSTM network, predict the optimal lane configuration strategy set for the next 1 - 6 hours (such as the opening time of tidal lanes, the pre-opening period of emergency lanes).

[0084] Lower-level lane agent:

[0085] Input:

[0086] Real-time traffic situation matrix, upper-level policy constraints.

[0087] Algorithm:

[0088] Real-time optimization based on Q-learning, adjusting the combination of lane functions (such as variable lane direction switching) and speed limit values every 30 seconds.

[0089] Reward function design:

[0090]

[0091] Weight combinations α, β, γ Improved traffic efficiency Reduced congestion propagation (0.3,0.5,0.2) 29.3% 35% (0.2,0.6,0.2) 24.1% 28% (0.4,0.4,0.2) 18.7% 22%

[0092] This table is a comparison table of experimental data.

[0093] Dynamic management of emergency lanes

[0094] Triggering conditions:

[0095] Main road congestion: The average speed of the main road < 25 km / h for 5 consecutive minutes (based on the Greenshields model to avoid misjudgment due to instantaneous fluctuations);

[0096] Downstream safety: No accidents within 500 meters downstream (verified in real time through cameras and radars).

[0097] Perform operations: Activate the laser clearance scanner (wavelength 905 nm), combine with infrared thermal imaging to detect obstacles in the emergency lane, with a false detection rate <0.1%. Based on 1000 hours of real road test data, a total of 12,345 obstacles were detected, and the number of false alarms was 12 times;

[0098] Open the emergency lane and switch to the red flashing mode (3 Hz) through electromagnetic road studs, and synchronously update the variable speed limit sign;

[0099] If a vehicle illegally occupies (the speed of crossing the line > 20 km / h), trigger a 105 dB audible and visual alarm (frequency 2 - 4 kHz).

[0100] Example 2:

[0101] Congestion entropy calculation and control

[0102] Divide a 10 km section into 20 500 - meter units;

[0103] Calculate the speed variation coefficient of each unit in real - time (Sampling window 30 seconds);

[0104] When the total congestion entropy is reached, trigger the following inter - lock control:

[0105] The speed limit of the upstream unit is decreased by 10% (e.g., 100 → 90 km / h);

[0106] Parameter Set value Vehicle following model Intelligent Driver Model (IDM) Maximum acceleration 1.5 m / s² Safe headway 1.8 s Number of Monte Carlo simulations 500 times

[0107] This table is the simulation parameter data.

[0108] Example 3: Implementation of electromagnetic road studs

[0109] The spacing between road studs is 5 meters, the magnetic unit uses neodymium iron boron permanent magnet (N52 grade), and the magnetic field strength after power - on is 1.0 T;

[0110] Multi - spectral LED unit display strategy:

[0111] Green: The lane is open for normal traffic;

[0112] Red flashing (3 Hz): The lane is closed;

[0113] Amber constant on: The tidal lane is enabled;

[0114] Scenario Traditional method (traffic volume per hour) This invention (traffic volume per hour) Improvement rate Morning rush hour 2,150 vehicles 2,780 vehicles +29.3% Accident handling 1,890 vehicles 2,430 vehicles +28.6%

[0115] This table is the simulation test (SUMO platform) data.

[0116] It should be noted that, in this article, relational terms such as first and second are only used 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 term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent in such process, method, article or device.

Claims

1. A highway lane control method based on real-time traffic flow, characterized in that: The following steps are involved: (a) Through the fusion perception of distributed optical fiber vibration sensor array and millimeter wave radar, the vehicle trajectories, vehicle type distribution and vibration abnormal events of the entire road section are obtained in real time to generate a multi-modal traffic situation matrix; (b) Construct a spatiotemporal graph convolutional network model to dynamically map the road topology into a lane connection graph with variable weights, and update the edge weights between nodes based on real-time traffic gradients; (c) Using a hierarchical reinforcement learning architecture, the upper meta-agent predicts the optimal lane configuration strategy set within the next 1-6 hours based on historical traffic patterns, and the lower lane agent adjusts the lane function mode and speed limit value combination at a 30-second time granularity based on the Q-learning algorithm; (d) Trigger the dynamic management mode of the emergency lane. When the average vehicle speed on the main road is less than 25 km / h for more than 5 minutes and there is no accident within 500 meters downstream, the laser clearance monitoring device is activated and the emergency lane is opened; (e) Deploy a digital twin sandbox for multi-resolution simulation verification. When the actual traffic flow deviates from the predicted value by more than 15%, the causal reasoning engine is started to backtrack the decision link and update the intelligent agent strategy library. The causal reasoning engine uses a Bayesian network model to backtrack abnormal nodes based on historical decision paths and real-time data. The nodes include historical decision paths, real-time traffic deviations and external events, and the edge weights are dynamically updated by the conditional probability table.

2. A highway lane control method based on real-time traffic flow according to claim 1, characterized in that: The method for generating the multimodal traffic situation matrix in step (a) includes: The energy characteristics of the 0.5-5 Hz frequency band in the vibration signal are extracted by wavelet transform, and the confidence of abnormal events in each lane section is calculated according to the following formula: in: Represents the energy integral value of the high frequency band, Indicates the energy integral value of the low frequency band, is the smoothing coefficient, the value is 0.01; When the confidence level exceeds 0.7, it is marked as a potential abnormal event and triggers the PTZ camera to take directional photos.

3. A highway lane control method based on real-time traffic flow according to claim 1, characterized in that: The reward function of the hierarchical reinforcement learning in step (c) is designed as: in: is the weight coefficient, is the travel time error weight coefficient, Congestion propagation attenuation coefficient, is the speed limit deviation penalty coefficient; , , , , represents the downstream congestion entropy value, Actual travel time, in seconds, calculated based on real-time traffic flow. To predict the travel time, in seconds, generated by the historical data model of the upper meta-agent, To set a speed limit, is the actual average vehicle speed, is the congestion entropy attenuation coefficient, which is used to adjust the weight of the impact of the congestion entropy value on the reward, and the value range is [0.1, 1.0]; When emergency lane occupation is detected, a penalty term is introduced: in: is the penalty factor, ranging from 0.1 to 0.

5. is the occupation duration, is the maximum allowed occupancy threshold, set to 30 minutes.

4. A highway lane control method based on real-time traffic flow according to claim 3, characterized in that: The method for calculating the downstream congestion entropy value includes: Divide the road units into 500-meter intervals, and first calculate the speed variation coefficient of each unit: in: Refers to the set of instantaneous speeds of all vehicles in the ith road unit, is the average instantaneous speed of vehicles in the i-th road unit, with a sampling window of 30 seconds. is the standard deviation of speed within the ii-th road unit, calculated based on the same time window; Congestion propagation model based on information entropy theory: in: Indicates the total length of the affected area. The coordinated control of the entire road network is triggered when the CV value is low. A low CV value indicates that the speed fluctuation is small, and all vehicles maintain 50±2 km / h, indicating that the traffic flow is stable and close to the free flow state. A high CV value indicates that the speed difference is large, and the vehicle speed changes dramatically between 20-80 km / h, indicating congestion, accidents or frequent lane changes.

5. A highway lane control system based on real-time traffic flow, used to implement a highway lane control method based on real-time traffic flow as described in any one of claims 1 to 4, characterized in that: include: Perception module: optical fiber vibration sensor, millimeter wave radar and edge computing unit, the edge computing unit is connected to the millimeter wave radar and optical fiber vibration sensor data through the 5G-V2X communication module; Execution module: electromagnetic spike array; Decision module: Federated learning server cluster, using differential privacy aggregation strategy; Verification module: laser clearance scanner and variable speed limit sign. The laser clearance scanner integrates an infrared thermal imaging module, which can distinguish between stationary obstacles and pedestrians, with a false detection rate of <0.1%; The edge computing unit uploads the preprocessed multimodal traffic situation matrix to the federated learning server cluster through the 5G-V2X communication module, and the federated learning server cluster broadcasts lane control instructions to the electromagnetic road spike array through 5G-V2X. The federated learning server cluster sends control signals to the laser clearance scanner and the variable speed limit sign through the edge computing unit, and the laser clearance scanner and the variable speed limit sign transmit the real-time status back to the federated learning server through 5G-V2X.

6. A highway lane control system based on real-time traffic flow according to claim 5, characterized in that: The electromagnetic spike array comprises: Magnetic adsorption unit, the magnetic field strength can be adjusted from 0.1 to 1.5 Tesla. When the magnetic field strength is 1.0 Tesla, it can resist the 8-level wind force to maintain the stable position of the road spikes. Multi-spectrum LED unit, including 590nm amber and 630nm red dual bands, color gamut covers 98% of sRGB; The anti-collision warning module triggers an audible and visual alarm when the vehicle crosses the line at a speed greater than 20 km / h. The audible and visual alarm module supports decibel adjustment, and the default setting is 90 dB.

Images