Tunnel traffic flow prediction and control method and system

Abstract

The application belongs to the technical field of traffic control, and discloses a tunnel traffic flow prediction and control method and system; the method comprises the following steps: a tunnel traffic management and control scheme based on edge computing is used, a sensor network is arranged in the tunnel, and an initial weight and a redundancy activation rule are preset; an edge node collects and processes data, generates a traffic and sensor vector, and inputs a flow prediction model; through an output degradation mechanism combined with PPO reinforcement learning, a real-time weight is dynamically adjusted and an optimized prediction is fed back; a three-dimensional tunnel model is constructed, an optimal control strategy is generated based on an NSGA-III algorithm, and the coordinated control of an induction system, tunnel equipment and a signal lamp is realized; the application improves the reliability of data collection, the accuracy of flow prediction and the robustness of the control strategy in an extreme scenario, and realizes the key breakthrough of tunnel traffic management and control from post-response to real-time intelligence.

Description

Technical Field

[0001] This invention relates to the field of traffic control technology, and more specifically, to a method and system for predicting and controlling tunnel traffic flow. Background Technology

[0002] With the acceleration of urbanization and the continuous growth of traffic flow, tunnels, as key nodes in the transportation network, face challenges in ensuring safe and efficient operation.

[0003] Chinese patent application CN111179601A discloses a tunnel traffic operation control method: Step S1, collecting traffic operation data within the tunnel and assessing the warning level of various traffic operation scenarios triggered within the tunnel based on the traffic operation data. This traffic operation data includes traffic flow data, vehicle behavior data, event data, and external data. Step S2, determining the control level and corresponding traffic operation scenario based on the warning level of each traffic operation scenario. Step S3, determining the corresponding control plan based on the control level and traffic operation scenario, and implementing traffic control according to the control plan. This control plan consists of multiple control measures. This tunnel traffic operation control method can assess the traffic operation status under various traffic operation scenarios triggered within the tunnel, enabling managers to promptly grasp the traffic operation warning level within the tunnel. Managers can then formulate targeted control plans based on the warning level to effectively and promptly address operational risks, reduce the probability of accidents, and improve tunnel operation efficiency.

[0004] While the above methods can meet the needs of most scenarios, research and practical application of these methods and existing technologies have revealed at least the following shortcomings:

[0005] The failure to consider the performance degradation of sensors in extreme environments makes it impossible to compensate for the degradation of sensor output in real time, leading to increased errors in traffic flow prediction.

[0006] In view of this, the present invention proposes a method and system for predicting and controlling tunnel traffic flow to solve the above problems. Summary of the Invention

[0007] To overcome the aforementioned deficiencies of the prior art and achieve the above objectives, the present invention provides the following technical solution: a method for predicting and controlling tunnel traffic flow, comprising the following steps:

[0008] A sensor network is deployed inside the tunnel, with preset initial sensor weights and redundant activation rules for extreme scenarios.

[0009] The sensor network collects and transmits sensor data and traffic data to edge nodes, analyzes them to obtain traffic vectors, and uses the sensor data and initial weights as input to the fusion model to obtain sensor vectors.

[0010] The initial weights, sensor vectors, and traffic vectors are used as inputs to the traffic flow prediction model to obtain traffic flow predictions for future time windows.

[0011] By combining sensor vectors, traffic vectors, and traffic flow predictions, and using the sensor output degradation mechanism triggered by edge nodes to simulate extreme scenarios, the sensor output is dynamically compensated. The real-time weights are then adjusted based on the compensation results using the PPO reinforcement learning algorithm.

[0012] Based on the real-time weights fed back to the fusion model and the traffic prediction model, the traffic prediction values ​​are updated, a three-dimensional tunnel model is constructed, and then the optimal control strategy for the guidance system, tunnel control system and upstream and downstream traffic lights is generated by combining the real-time weights with the NSGA-III algorithm.

[0013] Furthermore, the method for simulating extreme scenarios using a sensor output degradation mechanism triggered by edge nodes includes:

[0014] When the sensor network detects sensor data or traffic data exceeding a preset threshold, it triggers a sensor output degradation mechanism to simulate the scene and correct the parameters of the sensor output.

[0015] Furthermore, the method for scene simulation and parameter correction of the sensor output by the output degradation mechanism includes:

[0016] For camera sensors, a light intensity degradation model is built based on the attenuation coefficient, detection distance, and noise to correct for actual lighting conditions;

[0017] For radar sensors: For laser sensors, the effective detection range is calculated based on the scattering coefficient, absorption coefficient, transmission power, receiving sensitivity, and target reflectivity; for millimeter-wave sensors, the corrected signal-to-noise ratio is calculated based on the original signal-to-noise ratio, combined with rainfall and detection range.

[0018] For radar sensors, signal power loss is also calculated based on reflection coefficient, dynamic time delay, and multipath effect.

[0019] For traffic detection sensors, the corrected signal voltage is calculated based on the actual signal voltage, external magnetic field strength, interference frequency, and preset calibration coefficients, and the updated signal-to-noise ratio is calculated based on the corrected signal voltage.

[0020] Based on the aforementioned output degradation mechanism, the data obtained from scene simulation and parameter correction of sensor output is used for supervised learning training of various sensors.

[0021] Furthermore, methods for obtaining traffic vectors include:

[0022] Traffic flow can be obtained by counting the number of pulses passing through a vehicle using geomagnetic coils;

[0023] Traffic detection sensors collect the time difference of vehicles passing through the tunnel to calculate the average speed of vehicles in the corresponding lane.

[0024] The vehicle speed distribution of each lane is detected, speed intervals are divided according to preset speed intervals, and the proportion of vehicles in each lane in each interval is counted to obtain the vehicle speed distribution.

[0025] Calculate the average vehicle speed and combine it with the traffic flow to obtain the traffic density.

[0026] The actual average time for vehicles to pass through the tunnel is statistically analyzed, and the ratio of the actual average time to the free-flow time is calculated to obtain the travel time ratio.

[0027] Traffic vectors are obtained by concatenating traffic density, vehicle speed distribution, and travel time ratio.

[0028] Furthermore, if the average speed of vehicles in a lane is lower than R% of the preset free-flow speed, it is marked as a potential congestion; the NSGA-III algorithm is directly triggered to generate the optimal control strategy based on the historical weights of the sensors, and the weights and optimal control strategy are updated in real time based on the updated sensor data; where R is a preset threshold.

[0029] Furthermore, methods for obtaining sensor vectors include:

[0030] Acquire sampling data from each sensor. For camera sensors, obtain the mean square error between the sampled data and the processed data. Calculate the peak signal-to-noise ratio based on the mean square error and the maximum possible value of the pixel value.

[0031] By using the sampling data from camera sensors as input to the camera detection model, the detection confidence of the camera sensors can be obtained.

[0032] For radar sensors, the number of point clouds per unit area is calculated to obtain the point cloud density. The point cloud density is compared with a preset density threshold, and areas below the preset density threshold are marked as low quality. The marking results of all areas are statistically analyzed.

[0033] By using Kalman filtering, a unique ID is assigned to each physical entity, i.e., target, detected by radar sensors. The occurrence and loss of each target ID in continuous frame data of radar sensors are recorded, and the corresponding loss rate is recorded.

[0034] For environmental sensors, the historical data average is statistically analyzed, and the deviation rate between the real-time data and the historical data average is calculated.

[0035] The peak signal-to-noise ratio, detection confidence, labeling results, loss rate, and bias rate, along with the corresponding initial weights, are used as inputs to the fusion model to obtain the sensor vector.

[0036] Furthermore, methods for adjusting the real-time weights based on the compensation results include:

[0037] Define the state space: The state space consists of state vectors, which include environmental parameters, sensor vectors, traffic vectors, traffic flow predictions, and the deviation between the traffic flow predictions and the actual traffic flow.

[0038] Define the action space: The action space consists of action vectors, which include the weight adjustment of sensor nodes. The weight adjustment of each sensor satisfies the corresponding preset weight adjustment range, and the sum of the weights of all sensors is 1.

[0039] Design the reward function: The reward function is obtained by combining the deviation between the traffic flow prediction value and the actual traffic flow, the prediction stability within the preset time period, the sensor energy consumption, and the fluctuation rate of the sensor fusion weight.

[0040] Simulate extreme tunnel scenarios and compensate for changes in sensor output based on a sensor output degradation mechanism;

[0041] Sensor data, environmental data, and historical data are used as inputs to the Actor network, and a multi-layer Transformer+CNN is used as an intermediate layer to obtain the dynamic weight vector of each sensor.

[0042] Sensor data, environmental data, historical data, and the error between the fused traffic prediction and the actual value are used as inputs to the Critic network, and the long-term benefit of the current weight allocation strategy is output through the state value function.

[0043] The weight allocation strategy with the highest long-term return is selected as the final adjustment strategy to obtain the corresponding dynamic weights.

[0044] Furthermore, methods for generating the optimal control strategy include:

[0045] Pre-define optimization objectives and constraints;

[0046] Preset chromosome coding scheme: Set the gene structure of the chromosome, including weighted genes and strategy genes;

[0047] Randomly generate real-time weights and control strategy parameters for N sets of sensors. The control strategy parameters include control commands for the induction system, tunnel control system, and upstream and downstream traffic lights. Encode the real-time weights and control strategy parameters of the N sets of sensors into the gene structure of N1 sets of chromosomes. Eliminate individuals that violate the constraints to ensure that the initial population consists entirely of feasible solutions.

[0048] The population is divided into M front layers according to Pareto dominance, arranged in ascending order of the number of dominated members. Edge nodes obtain the first L front layers and upload them.

[0049] Based on the target space formed by the first L front layers, a uniformly distributed set of reference points is generated in each dimension of the preset optimization objective. The number of reference points is E times the number of chromosomes in the first L front layers. The position of the reference point set is periodically updated according to the change of the objective function range.

[0050] The parent chromosome parameters are crossed to generate offspring. For weighted genes, arithmetic crossover is used to ensure that the weights sum to 1 after crossover; for strategy genes, discrete crossover is used to randomly swap the parent parameter values.

[0051] The offspring are subjected to random perturbation through mutation; for the weight genes, each weight is subjected to random perturbation and then renormalized; for the policy genes, Gaussian noise is added within the parameter range.

[0052] Calculate the Euclidean distance between each offspring individual and the reference point, associate the individuals with the reference point that minimizes the Euclidean distance, retain the top Q individuals associated with the reference point into the offspring, and directly retain the top P individuals of the parent generation's optimal frontier into the offspring; thus obtaining the offspring population.

[0053] Repeat the iteration on the offspring individuals until the preset termination condition is met.

[0054] Furthermore, methods for obtaining the first L leading edges include:

[0055] Step A1, Domination Rule: For two individuals A and B, if A's target value is not lower than B's in all preset optimization objectives, and A is better than B in at least one objective, then A is said to dominate B, and B's dominated number is incremented by 1; Count the dominated number of all individuals according to the above domination rule.

[0056] Step A2: Extract all individuals in the population that are not dominated by any other individual, i.e., individuals with a dominated number of 0, to form the first frontier layer;

[0057] Step A3: Repeat step A1 to extract the number of dominated individuals among all remaining individuals after the first front layer, and extract individuals with a number of dominated individuals of 0 to form the second front layer;

[0058] Step A4: Repeat step A3 to extract currently undominated individuals one by one to form the next frontier layer, until all individuals are assigned to the corresponding frontier layer or the preset maximum layer number L is reached.

[0059] A tunnel traffic flow prediction and control system, implementing the tunnel traffic flow prediction and control method, includes:

[0060] Network setup module: Deploy a sensor network inside the tunnel, preset initial sensor weights and redundant activation rules for extreme scenarios;

[0061] Data acquisition module: The sensor network collects and transmits sensor data and traffic data to edge nodes, analyzes the data to obtain traffic vectors, and uses the sensor data and initial weights as input to the fusion model to obtain sensor vectors.

[0062] Traffic flow prediction module: The initial weights, sensor vectors, and traffic vectors are used as inputs to the traffic flow prediction model to obtain traffic flow prediction values ​​for future time windows.

[0063] Compensation Analysis Module: Combining sensor vectors, traffic vectors, and flow prediction values, it simulates extreme scenarios using the sensor output degradation mechanism triggered by edge nodes, dynamically compensates the sensor output, and obtains real-time weights based on the compensation results through the PPO reinforcement learning algorithm;

[0064] Intelligent control module: Based on real-time weights, it feeds back to the fusion model and traffic prediction model, updates the traffic prediction value, constructs a three-dimensional model of the tunnel, and then combines the real-time weights with the NSGA-III algorithm to generate the optimal control strategy for controlling the guidance system, the tunnel control system and the upstream and downstream traffic lights.

[0065] The technical effects and advantages of the tunnel traffic flow prediction and control method and system of the present invention are as follows:

[0066] This invention establishes a core technological advantage lacking in traditional solutions by constructing a sensor output degradation mechanism at edge nodes and combining it with PPO reinforcement learning to dynamically compensate for sensor performance degradation in extreme scenarios. In extreme scenarios such as dense fog, heavy rain, and accidents, it can identify sensor anomalies in real time, improve the accuracy of multi-source data fusion, and reduce traffic prediction errors by dynamically adjusting fusion weights. The edge computing architecture supports localized real-time processing, shortening response time in extreme scenarios and ensuring the rapid generation of reliable control strategies even when sensor performance degrades. Through a closed loop of "model simulation - weight compensation - edge decision-making," this invention improves the reliability of data acquisition, the accuracy of traffic prediction, and the robustness of control strategies in extreme scenarios, achieving a key breakthrough in tunnel traffic management from post-event response to real-time intelligent response. Attached Figure Description

[0067] Figure 1 This is a schematic diagram of the tunnel traffic flow prediction and control method of the present invention;

[0068] Figure 2 This is a diagram of the tunnel traffic flow prediction and control system architecture of the present invention. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] Example 1

[0071] Please see Figure 1 As shown in the figure, this embodiment provides a method for predicting and controlling tunnel traffic flow, including the following steps:

[0072] A sensor network is deployed within the tunnel, with initial sensor weights and redundant activation rules for extreme scenarios pre-set. For example, during heavy rain, a combination of lidar and millimeter-wave radar is activated (the activation weight of lidar and millimeter-wave radar increases accordingly, e.g., activation weight = 1.5 * initial weight); during dense fog, a combination of thermal imaging camera and radar is activated; and during accidents, a redundant sensor cluster is activated. By deploying a sensor network within the tunnel and pre-setting initial sensor weights and redundant activation rules for extreme scenarios, multi-modal sensors (such as lidar, cameras, and geomagnetic coils) can collect multi-dimensional data such as traffic flow and environment in real time. The initial weights ensure efficient fusion of multi-source data in normal scenarios, while the redundant activation rules dynamically activate backup sensor clusters in extreme scenarios such as heavy rain, dense fog, and accidents, ensuring the integrity and reliability of data collection. Both provide stable and high-precision input data for subsequent traffic prediction models, supporting edge nodes to dynamically adjust fusion weights based on PPO reinforcement learning, thereby improving traffic prediction accuracy. At the same time, they provide real-time and reliable sensor status information for the generation of intelligent guidance and control strategies, and combine the NSGA-III algorithm to optimize control parameters such as traffic light timing and lane allocation, so as to achieve rapid response and global optimal control in extreme scenarios, ultimately enhancing the prediction accuracy, emergency robustness and traffic efficiency of the tunnel traffic system.

[0073] The sensor network collects and transmits sensor data and traffic data to edge nodes, analyzing them to obtain traffic vectors. Sensor data and initial weights are used as input to a fusion model to obtain sensor vectors. The sensor network collects and transmits sensor data and traffic data to edge nodes, which, after analysis and processing, generate traffic vectors reflecting real-time traffic conditions, providing intuitive traffic flow characteristic input for the traffic flow prediction model. Simultaneously, raw sensor data and preset initial weights are input to the fusion model, and sensor vectors are generated through weighted fusion, noise filtering, and other processing, effectively integrating multi-source sensor information and suppressing the performance degradation of single sensors in complex environments. Together, these components provide multi-dimensional, high-precision input data for the traffic flow prediction module, supporting edge nodes in accurately predicting future traffic flow within time windows based on models such as LSTM / Transformer. Furthermore, they provide real-time, reliable traffic condition data for the generation of intelligent guidance and control strategies, assisting the NSGA-III algorithm in dynamically optimizing control parameters such as signal timing, lane allocation, and guidance information dissemination, achieving a closed-loop process from data collection to prediction and decision-making, significantly improving the real-time performance and intelligence level of tunnel traffic management.

[0074] Methods for obtaining traffic vectors include:

[0075] Traffic flow can be obtained by counting the number of pulses passing through a vehicle using geomagnetic coils;

[0076] Traffic detection sensors collect the time difference of vehicles passing through the tunnel to calculate the average speed of vehicles in the corresponding lane.

[0077] The system detects and obtains the vehicle speed distribution of each lane, divides the speed into intervals according to preset speed intervals, and counts the proportion of vehicles in each lane within each interval to obtain the vehicle speed distribution. If the average speed of vehicles in a lane is lower than the preset free-flow speed (referring to the highest average speed at which vehicles can drive stably under ideal traffic conditions, without traffic congestion, traffic accidents, good weather, etc.) by R%, where R is a preset threshold, such as 50%, which is based on traffic flow theories, such as the Greenshields model, car-following theory, and empirical settings derived from actual observation data, then it is marked as potential congestion.

[0078] Calculate the average vehicle speed and combine it with the traffic flow to obtain the traffic density.

[0079] The actual average time for vehicles to pass through the tunnel is statistically analyzed, and the ratio of the actual average time to the free-flow time is calculated to obtain the travel time ratio.

[0080] Traffic vectors are obtained by concatenating traffic density, vehicle speed distribution, and travel time ratio.

[0081] Training methods for fusion models include:

[0082] S sets of fusion training data were collected in advance. The fusion training data included sensor data, initial weights, and sensor vectors.

[0083] Sensor data and initial weights are used as input to the fusion model, and sensor vectors are used as output. The goal is to minimize the error between the output sensor vector and the actual sensor vector. The network parameters of the fusion model are optimized using a natural heuristic optimization algorithm to obtain the network parameters that minimize the error between the output sensor vector and the actual sensor vector. The fusion model constructed with the corresponding network parameters is used as the trained fusion model.

[0084] Methods for obtaining sensor vectors include:

[0085] Acquire sampling data from each sensor. For camera sensors, obtain the mean square error between the sampled data and the processed data. Calculate the peak signal-to-noise ratio based on the mean square error and the maximum possible value of the pixel value, such as 255.

[0086] By using the sampling data from camera sensors as input to the camera detection model, the detection confidence of the camera sensors can be obtained.

[0087] Training methods for camera detection models include:

[0088] K sets of detection training data are collected in advance, including sampling data from camera sensors and detection confidence levels.

[0089] Using sampled data from camera sensors as input to the camera detection model and detection confidence as output, the network parameters of the camera detection model are optimized through a natural heuristic optimization algorithm with the goal of minimizing the error between the output detection confidence and the actual detection confidence. The network parameters that minimize the error between the output detection confidence and the actual detection confidence are obtained, and the camera detection model constructed with the corresponding network parameters is used as the trained camera detection model.

[0090] For radar sensors, the number of point clouds per unit area is calculated to obtain the point cloud density. The point cloud density is compared with a preset density threshold, and areas below the preset density threshold are marked as low quality. The marking results of all areas are statistically analyzed.

[0091] By using Kalman filtering, a unique ID is assigned to each physical entity detected by each radar-type sensor, i.e., the detected target. The occurrence and loss of each target ID in the continuous frame data of the radar-type sensor are recorded, and the corresponding loss rate is recorded.

[0092] For environmental sensors, the historical data average is statistically analyzed, and the deviation rate between the real-time data and the historical data average is calculated.

[0093] The peak signal-to-noise ratio, detection confidence, labeling results, loss rate, and bias rate, along with the corresponding initial weights, are used as inputs to the fusion model to obtain the sensor vector.

[0094] The initial weights, sensor vectors, and traffic vectors are used as inputs to the traffic flow prediction model to obtain traffic flow predictions for future time windows.

[0095] Training methods for traffic prediction models include:

[0096] Y sets of detection training data are collected in advance. The detection training data includes initial weights, sensor vectors and traffic vectors, as well as traffic prediction values ​​for future time windows.

[0097] The initial weights, sensor vectors, and traffic vectors are used as inputs to the traffic flow prediction model, and the traffic flow prediction values ​​for future time windows are used as the outputs of the traffic flow prediction model. With the goal of minimizing the error between the output traffic flow prediction values ​​for future time windows and the actual traffic flow values, the network parameters of the traffic flow prediction model are optimized using a natural heuristic optimization algorithm. The network parameters that minimize the error between the detection confidence of the output traffic flow prediction model and the actual detection confidence are obtained, and the traffic flow prediction model constructed with the corresponding network parameters is used as the trained traffic flow prediction model.

[0098] By combining sensor vectors, traffic vectors, and traffic flow predictions, extreme scenarios are fitted and simulated using a sensor output degradation mechanism triggered by edge nodes. Dynamic compensation is then applied to the sensor output, and real-time weights are obtained by adjusting the compensation results using the PPO reinforcement learning algorithm. These steps enable real-time compensation for sensor output degradation in complex environments. This process significantly optimizes the accuracy and robustness of multi-source data fusion, providing more reliable input for traffic flow prediction models. Simultaneously, the real-time weights, as a quantitative representation of sensor states, directly participate in the generation of intelligent guidance and control strategies. This helps the NSGA-III algorithm dynamically balance sensor reliability and control strategy efficiency in multi-objective optimization, achieving precise control of guidance systems, tunnel equipment, and traffic lights. Especially in accidents or severe weather, this can shorten emergency response time and effectively improve the safety and efficiency of tunnel traffic.

[0099] Methods for dynamically adjusting fusion weights include:

[0100] Define the state space: The state space consists of state vectors, which include environmental parameters, sensor vectors, traffic vectors, traffic flow predictions, and the deviation between the traffic flow predictions and the actual traffic flow.

[0101] Define the action space: The action space consists of action vectors, which include the weight adjustment of sensor nodes. The weight adjustment of each sensor satisfies the corresponding preset weight adjustment range, and the sum of the weights of all sensors is 1.

[0102] Design the reward function: Combine the deviation between the traffic flow prediction and the actual traffic flow, the prediction stability within a preset time period (e.g., the deviation does not exceed the preset deviation threshold for 10 consecutive minutes), the sensor energy consumption (e.g., the total power consumption of the sensors decreases), and the fluctuation rate of the sensor fusion weight (e.g., the sensor fusion weight is too low, resulting in missed detection).

[0103] Simulate extreme tunnel scenarios and compensate for changes in sensor output based on a sensor output degradation mechanism;

[0104] Methods for compensating for changes in sensor output include:

[0105] Acquire tunnel BIM model data to simulate tunnel scenarios, including road surface material, lighting system, tunnel wall texture, number of lanes, speed limit signs, entrance and exit coordinates, vehicle type distribution (such as passenger car / freight car ratio), initial traffic flow density (such as peak / off-peak parameters), and sensor installation locations (such as the installation height and angle of lidar on the tunnel top / sidewall).

[0106] Preset multimodal extreme scenario data; such as dense fog scenario, such as setting the ambient visibility to 5-20 meters and air humidity >90%, to simulate the infrared radiation characteristics of thermal imaging cameras; accident scenario, such as triggering a vehicle collision event at a designated location, generating dynamic interference such as the stopped accident vehicle, scattered debris, and exhaust emissions (affecting sensor signals), and simultaneously simulating sudden changes in traffic flow when rescue vehicles intervene.

[0107] For each type of sensor, an output degradation mechanism under extreme scenarios is established. Based on the output degradation mechanism, degradation simulations are performed on different types of sensors, and the sensors are trained based on the updated parameters obtained from the degradation simulations.

[0108] Methods for simulating degradation in different types of sensors include:

[0109] For camera sensors, such as tunnel cameras, a light intensity degradation model is built based on the attenuation coefficient, detection distance, and noise. The attenuation coefficient is positively correlated with PM2.5 concentration, and the noise term is negatively correlated with the signal-to-noise ratio.

[0110] In radar sensors: For laser sensors, such as lidar, the effective detection range is calculated based on the scattering coefficient, absorption coefficient, transmit power, receive sensitivity, and target reflectivity; for millimeter-wave sensors, such as millimeter-wave radar, the corrected signal-to-noise ratio is calculated based on the original signal-to-noise ratio, combined with rainfall and detection range.

[0111] For radar sensors, signal power loss is also calculated based on reflection coefficient, dynamic time delay, and multipath effect.

[0112] For traffic detection sensors, such as geomagnetic sensors and piezoelectric sensors, the corrected signal voltage is calculated based on the actual signal voltage, external magnetic field strength, interference frequency and preset calibration coefficients, and the updated signal-to-noise ratio is calculated based on the corrected signal voltage.

[0113] The sensors are trained using the updated parameters obtained by simulating scenarios and correcting parameters of the sensor output based on the output degradation mechanism.

[0114] A high-precision simulation scenario was constructed based on the tunnel BIM model, pre-setting multi-modal extreme scenario data such as dense fog and accidents. Physical degradation models were established for different sensors, including cameras and radar, enabling real-time simulation of sensor performance degradation under extreme conditions at edge nodes. The PPO reinforcement learning algorithm at the edge was trained based on the sensor parameters updated by the degradation simulation. This allows the system to learn the sensor output patterns under extreme scenarios in advance, and then dynamically adjust the fusion weights during actual operation, ensuring the accuracy of multi-source data fusion. Even in dense fog and accident scenarios, the recognition rate of abnormal sensor data remains high. This process provides the traffic prediction model with input data that more closely resembles real extreme environments, reducing prediction errors in sudden scenarios. Simultaneously, it provides reliable sensor status data for the generation of intelligent guidance and control strategies, helping the NSGA-III algorithm to quickly respond to sudden changes in traffic flow, shortening emergency response time in extreme scenarios, and significantly improving the safety and traffic efficiency of the tunnel traffic system.

[0115] Sensor data, environmental data, and historical data are used as inputs to the Actor network, and a multi-layer Transformer+CNN is used as an intermediate layer to obtain the dynamic weight vector of each sensor.

[0116] Sensor data, environmental data, historical data, and the error between the fused traffic prediction and the actual value are used as inputs to the Critic network, and the long-term benefit of the current weight allocation strategy is output through the state value function.

[0117] The weight allocation strategy with the highest long-term return is selected as the final adjustment strategy to obtain the corresponding dynamic weights.

[0118] Based on feedback from real-time weights to the fusion model and traffic prediction model, the traffic prediction values ​​are updated, a 3D tunnel model is constructed, and then, combined with the real-time weights, the NSGA-III algorithm generates the optimal control strategy for controlling the guidance system, tunnel control system, and upstream and downstream traffic lights. The feedback mechanism of real-time weights to the fusion model and traffic prediction model dynamically optimizes the fusion accuracy of multi-source sensor data, reduces the error of the updated traffic prediction values ​​in extreme scenarios, and provides more accurate input for short-term trend analysis of tunnel traffic flow. The 3D digital twin model of the tunnel constructed with real-time weights can map vehicle trajectories, sensor states, and equipment operation in real time, providing visualized spatial and state constraints (such as lane conflict risk assessment and equipment power threshold verification) for the NSGA-III algorithm. Based on this, the optimal control strategy generated by the NSGA-III algorithm through multi-objective optimization (prediction error, response time, system energy consumption) can achieve coordinated control of the guidance system, tunnel equipment, and upstream and downstream traffic lights, improving traffic efficiency in normal scenarios and shortening emergency response time in extreme scenarios. This forms a closed loop of "data fusion optimization → accurate prediction → intelligent decision-making," significantly enhancing the real-time performance, robustness, and global optimality of tunnel traffic management.

[0119] Methods for generating optimal control strategies include:

[0120] The optimization objectives and constraints are preset. For example, the optimization objectives include minimizing prediction error, minimizing emergency response time, and minimizing system energy consumption. The constraints include communication delay being lower than a preset delay threshold, equipment power being lower than a preset power threshold, and the optimization strategy not causing lane conflict or speeding risk.

[0121] Preset chromosome coding scheme: Set the gene structure of the chromosome, including weighted genes and strategy genes;

[0122] Randomly generate real-time weights and control strategy parameters for N sets of sensors. The control strategy parameters include control commands for the induction system (such as the update frequency of the induction screen), control commands for the tunnel control system (such as the start threshold of the ventilation system), and control commands for upstream and downstream traffic lights (such as the traffic light cycle). Encode the real-time weights and control strategy parameters of the N sets of sensors into the gene structure of N1 sets of chromosomes. Eliminate individuals that violate the constraints to ensure that the initial population consists entirely of feasible solutions. N and N1 are numerically equal.

[0123] The population is divided into M front layers according to Pareto dominance, arranged in ascending order of the number of dominated members. Edge nodes obtain the first L front layers and upload them.

[0124] Methods for obtaining L frontier layers include:

[0125] Step A1, Domination Rule: For two individuals A and B, if A's target value is not lower than B's in all preset optimization objectives, and A is better than B in at least one objective, then A is said to dominate B, and B's dominated number is incremented by 1; Count the dominated number of all individuals according to the above domination rule.

[0126] Step A2: Extract all individuals in the population that are not dominated by any other individual, i.e., individuals with a dominated number of 0, to form the first frontier layer;

[0127] Step A3: Repeat step A1 to extract the number of dominated individuals among all remaining individuals after the first front layer, and extract individuals with a number of dominated individuals of 0 to form the second front layer;

[0128] Step A4: Repeat step A3 to extract currently undominated individuals one by one to form the next frontier layer, until all individuals are assigned to the corresponding frontier layer or the preset maximum layer number L is reached.

[0129] Based on the target space formed by the first L front layers, a uniformly distributed set of reference points is generated in each dimension of the preset optimization objective. The number of reference points is 1.5 times the number of chromosomes in the first L front layers. The position of the reference point set is periodically updated according to the change of the objective function range.

[0130] The parent chromosome parameters are crossed to generate offspring. For weighted genes, arithmetic crossover is used to ensure that the weights sum to 1 after crossover; for strategy genes, discrete crossover is used to randomly swap the parent parameter values.

[0131] The offspring are subjected to random perturbation through mutation; for the weight genes, each weight is subjected to random perturbation and then renormalized; for the policy genes, Gaussian noise is added within the parameter range.

[0132] Calculate the Euclidean distance between each offspring individual and the reference point, associate the individuals with the reference point that minimizes the Euclidean distance, retain the top Q individuals associated with the reference point into the offspring, and directly retain the top P individuals of the parent generation's optimal frontier into the offspring; thus obtaining the offspring population.

[0133] Repeat the iteration on the offspring individuals until the preset termination condition is met (such as the volatility of the objective function being less than 5% for 10 consecutive generations, or reaching the preset maximum number of iterations).

[0134] Example 2

[0135] This embodiment provides a quantum-photonic hybrid computing architecture applied to Embodiment 1, including:

[0136] Deploy quantum annealing machines (such as D-Wave) in the cloud for offline training of PPO policy networks and NSGA-III multi-objective optimization models;

[0137] The objective function is optimized by qubit mapping, and then the optimization objective function is solved. For example, the sensor weight allocation problem is transformed into the Ising model for solution.

[0138] Edge nodes integrate photon matrix computing units (such as Lightmatter chips) to enable real-time generation of PPO policy inference and NSGA-III solutions, thereby reducing latency.

[0139] Through the aforementioned quantum-photon hybrid computing architecture, quantum computing is responsible for global model optimization, while photonic computing handles local real-time decision-making. This enables the formation of a collaborative computing network across the cloud, edge, and device, resolving the real-time bottleneck of complex algorithms and improving response speed.

[0140] Example 3

[0141] Please see Figure 2 As shown, this embodiment provides a tunnel traffic flow prediction and control system, including:

[0142] Network setup module: Deploy a sensor network inside the tunnel, preset initial sensor weights and redundant activation rules for extreme scenarios;

[0143] Data acquisition module: The sensor network collects and transmits sensor data and traffic data to edge nodes, analyzes the data to obtain traffic vectors, and uses the sensor data and initial weights as input to the fusion model to obtain sensor vectors.

[0144] Traffic flow prediction module: The initial weights, sensor vectors, and traffic vectors are used as inputs to the traffic flow prediction model to obtain traffic flow prediction values ​​for future time windows.

[0145] Compensation Analysis Module: Combining sensor vectors, traffic vectors, and flow prediction values, it simulates extreme scenarios using the sensor output degradation mechanism triggered by edge nodes, dynamically compensates the sensor output, and obtains real-time weights based on the compensation results through the PPO reinforcement learning algorithm;

[0146] Intelligent control module: Based on real-time weights, it feeds back to the fusion model and traffic prediction model, updates the traffic prediction value, constructs a three-dimensional model of the tunnel, and then combines the real-time weights with the NSGA-III algorithm to generate the optimal control strategy for controlling the guidance system, tunnel control system and upstream and downstream traffic lights.

[0147] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0148] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for predicting and controlling tunnel traffic flow, characterized in that, Includes the following steps: A sensor network is deployed inside the tunnel, with preset initial sensor weights and redundant activation rules for extreme scenarios. The sensor network collects and transmits sensor data and traffic data to edge nodes, analyzes them to obtain traffic vectors, and uses the sensor data and initial weights as input to the fusion model to obtain sensor vectors. The initial weights, sensor vectors, and traffic vectors are used as inputs to the traffic flow prediction model to obtain traffic flow predictions for future time windows. By combining sensor vectors, traffic vectors, and traffic flow predictions, an edge-node-triggered sensor output degradation mechanism is used to simulate extreme scenarios, dynamically compensating for the sensor output. Real-time weights are then adjusted based on the compensation results using the PPO reinforcement learning algorithm. Define the state space: The state space consists of state vectors, which include environmental parameters, sensor vectors, traffic vectors, traffic flow predictions, and the deviation between the traffic flow predictions and the actual traffic flow. Define the action space: The action space consists of action vectors, which include the weight adjustment of sensor nodes. The weight adjustment of each sensor satisfies the corresponding preset weight adjustment range, and the sum of the weights of all sensors is 1. Design the reward function: The reward function is obtained by combining the deviation between the traffic flow prediction value and the actual traffic flow, the prediction stability within the preset time period, the sensor energy consumption, and the fluctuation rate of the sensor fusion weight. Simulate extreme tunnel scenarios and compensate for changes in sensor output based on a sensor output degradation mechanism; Sensor data, environmental data, and historical data are used as inputs to the Actor network, and a multi-layer Transformer+CNN is used as an intermediate layer to obtain the dynamic weight vector of each sensor. Sensor data, environmental data, historical data, and the error between the fused traffic prediction and the actual value are used as inputs to the Critic network, and the long-term benefit of the current weight allocation strategy is output through the state value function. Select the weight allocation strategy with the highest long-term return as the final adjustment strategy to obtain the corresponding dynamic weights; Based on the real-time weights fed back to the fusion model and the traffic prediction model, the traffic prediction values ​​are updated, a three-dimensional tunnel model is constructed, and then the optimal control strategy for the guidance system, tunnel control system and upstream and downstream traffic lights is generated by combining the real-time weights with the NSGA-III algorithm.

2. The tunnel traffic flow prediction and control method according to claim 1, characterized in that, The method for simulating extreme scenarios using a sensor output degradation mechanism triggered by edge nodes includes: When the sensor network detects sensor data or traffic data exceeding a preset threshold, it triggers a sensor output degradation mechanism to simulate the scene and correct the parameters of the sensor output.

3. The tunnel traffic flow prediction and control method according to claim 2, characterized in that, The output degradation mechanism includes methods for scene simulation and parameter correction of sensor output, including: For camera-type sensors, a light intensity degradation model is built based on the attenuation coefficient, detection distance, and noise to correct for actual lighting conditions; For radar sensors: For laser sensors, the effective detection range is calculated based on the scattering coefficient, absorption coefficient, transmission power, receiving sensitivity, and target reflectivity; for millimeter-wave sensors, the corrected signal-to-noise ratio is calculated based on the original signal-to-noise ratio, combined with rainfall and detection range. For radar sensors, signal power loss is also calculated based on reflection coefficient, dynamic time delay, and multipath effect. For traffic detection sensors, the corrected signal voltage is calculated based on the actual signal voltage, external magnetic field strength, interference frequency, and preset calibration coefficients, and the updated signal-to-noise ratio is calculated based on the corrected signal voltage. Supervised learning training is performed on various sensors based on the data obtained from the degradation simulation.

4. The tunnel traffic flow prediction and control method according to claim 1, characterized in that, Methods for obtaining traffic vectors include: Traffic flow can be obtained by counting the number of pulses passing through a vehicle using geomagnetic coils; Traffic detection sensors collect the time difference of vehicles passing through the tunnel to calculate the average speed of vehicles in the corresponding lane. The vehicle speed distribution of each lane is detected, speed intervals are divided according to preset speed intervals, and the proportion of vehicles in each lane in each interval is counted to obtain the vehicle speed distribution. Calculate the average vehicle speed and combine it with the traffic flow to obtain the traffic density. The actual average time for vehicles to pass through the tunnel is statistically analyzed, and the ratio of the actual average time to the free-flow time is calculated to obtain the travel time ratio. Traffic vectors are obtained by concatenating traffic density, vehicle speed distribution, and travel time ratio.

5. The tunnel traffic flow prediction and control method according to claim 4, characterized in that, If the average speed of vehicles in a lane is lower than R% of the preset free-flow speed, it is marked as a potential congestion. The NSGA-III algorithm is directly triggered to generate the optimal control strategy based on the historical weights of the sensors, and the weights and optimal control strategy are updated in real time based on the updated sensor data.

6. The tunnel traffic flow prediction and control method according to claim 1, characterized in that, Methods for obtaining sensor vectors include: Acquire sampling data from each sensor. For camera sensors, obtain the mean square error between the sampled data and the processed data. Calculate the peak signal-to-noise ratio based on the mean square error and the maximum possible value of the pixel value. By using the sampling data from camera sensors as input to the camera detection model, the detection confidence of the camera sensors can be obtained. For radar sensors, the number of point clouds per unit area is calculated to obtain the point cloud density. The point cloud density is compared with a preset density threshold, and areas below the preset density threshold are marked as low quality. The marking results of all areas are statistically analyzed. By using Kalman filtering, a unique ID is assigned to each physical entity, i.e., target, detected by radar sensors. The occurrence and loss of each target ID in continuous frame data of radar sensors are recorded, and the corresponding loss rate is recorded. For environmental sensors, the historical data average is statistically analyzed, and the deviation rate between the real-time data and the historical data average is calculated. The peak signal-to-noise ratio, detection confidence, labeling results, loss rate, and bias rate, along with the corresponding initial weights, are used as inputs to the fusion model to obtain the sensor vector.

7. The tunnel traffic flow prediction and control method according to claim 1, characterized in that, Methods for generating optimal control strategies include: Pre-define optimization objectives and constraints; Preset chromosome coding scheme: Set the gene structure of the chromosome, including weighted genes and strategy genes; Randomly generate real-time weights and control strategy parameters for N sets of sensors. The control strategy parameters include control commands for the induction system, tunnel control system, and upstream and downstream traffic lights. Encode the real-time weights and control strategy parameters of the N sets of sensors into the gene structure of N1 sets of chromosomes. Eliminate individuals that violate the constraints to ensure that the initial population consists entirely of feasible solutions. The population is divided into M front layers according to Pareto dominance, arranged in ascending order of the number of dominated members. Edge nodes obtain the first L front layers and upload them. Based on the target space formed by the first L front layers, a uniformly distributed set of reference points is generated in each dimension of the preset optimization objective. The number of reference points is E times the number of chromosomes in the first L front layers. The position of the reference point set is periodically updated according to the change of the objective function range. The parent chromosome parameters are crossed to generate offspring. For weighted genes, arithmetic crossover is used to ensure that the weights sum to 1 after crossover; for strategy genes, discrete crossover is used to randomly swap the parent parameter values. The offspring are subjected to random perturbation through mutation; for the weight genes, each weight is subjected to random perturbation and then renormalized; for the policy genes, Gaussian noise is added within the parameter range. Calculate the Euclidean distance between each offspring individual and the reference point, associate the individuals with the reference point that minimizes the Euclidean distance, retain the top Q individuals associated with the reference point into the offspring, and directly retain the top P individuals of the parent generation's optimal frontier into the offspring; thus obtaining the offspring population. Repeat the iteration on the offspring individuals until the preset termination condition is met.

8. The tunnel traffic flow prediction and control method according to claim 7, characterized in that, Methods for obtaining the top L leading edges include: For two individuals A and B, if A's target value is not lower than B's in all preset optimization objectives and is better than B in at least one objective, then A is said to dominate B, and B's number of dominated individuals is increased by 1; the number of dominated individuals is counted according to the above rules. Extract all individuals in the population that are not dominated by any other individual, i.e., individuals with a dominated number of 0, to form the first front layer; Extract the number of dominated individuals from the remaining individuals after the first front layer, repeat the above process for all individuals, extract individuals with a dominated number of 0, and form the second front layer; repeat this step, successively extracting currently undominated individuals to form the next front layer, until all individuals are assigned to the corresponding front layer or the preset maximum number of layers L is reached.

9. A tunnel traffic flow prediction and control system, implementing the tunnel traffic flow prediction and control method according to any one of claims 1-8, characterized in that, include: Network setup module: Deploy a sensor network inside the tunnel, preset initial sensor weights and redundant activation rules for extreme scenarios; Data acquisition module: The sensor network collects and transmits sensor data and traffic data to edge nodes, analyzes the data to obtain traffic vectors, and uses the sensor data and initial weights as input to the fusion model to obtain sensor vectors. Traffic flow prediction module: The initial weights, sensor vectors, traffic vectors and historical sensor vectors are used as inputs to the traffic flow prediction model to obtain traffic flow prediction values ​​for future time windows. Compensation Analysis Module: Combining sensor vectors, traffic vectors, and flow prediction values, it simulates extreme scenarios using the sensor output degradation mechanism triggered by edge nodes, dynamically compensates the sensor output, and obtains real-time weights based on the compensation results through the PPO reinforcement learning algorithm; Intelligent control module: Based on real-time weights, it feeds back to the fusion model and traffic prediction model, updates the traffic prediction value, constructs a three-dimensional model of the tunnel, and then combines the real-time weights with the NSGA-III algorithm to generate the optimal control strategy for controlling the guidance system, the tunnel control system and the upstream and downstream traffic lights.

Images