A method for assessing the traffic capacity of tunnel sections based on intelligent connected vehicles.

Abstract

The application is specifically a tunnel group road section traffic capacity evaluation method based on intelligent networking, and relates to the technical field of traffic engineering, comprising: based on fusion data and digital twin model, combining tunnel engineering constraints, an AI-driven dynamic evaluation model is constructed to calculate the traffic capacity, congestion risk and bottleneck position of each road section of the tunnel group in real time. In the application, a multi-source data acquisition system is combined with an edge and cloud two-level fusion architecture to construct an LSTM-XGBoost double model architecture; the LSTM model can calculate and predict the traffic capacity of four time nodes within the next 10 minutes in real time, the XGBoost model accurately locates the bottleneck position and quantifies the contribution degree of four types of causes, while ensuring the accuracy through multidimensional verification, can cover the whole tunnel group, capture instantaneous traffic changes, and provide real-time data, prediction trends and panoramic decision basis of bottleneck causes for traffic management.

Description

Technical Field

[0001] This invention relates to the field of traffic engineering technology, and in particular to a method for assessing the traffic capacity of tunnel group road sections based on intelligent network connectivity. Background Technology

[0002] As a key node in the highway network, the capacity of tunnel clusters directly affects the efficiency and safety of regional traffic operations. Traditional methods for assessing tunnel capacity mainly rely on macroscopic traffic data collected by fixed detectors (such as loop detectors and video), which have the following limitations:

[0003] First, the data source is singular, making it difficult to cover the overall traffic status of both connected and non-connected vehicles. Furthermore, it cannot obtain dynamic changes in vehicle micro-behaviors (such as lane-changing intentions and acceleration fluctuations) and tunnel environments (such as visibility and CO concentration), resulting in a significant deviation between the assessment results and the actual scenario.

[0004] Secondly, the real-time performance is insufficient. Traditional methods are mostly based on historical statistical data for static assessment, which cannot capture the instantaneous changes in traffic flow and the impact of sudden events (such as vehicle breakdowns and lane occupancy) on traffic capacity, making it difficult to support dynamic traffic management decisions. Thirdly, the bottleneck location is vague. It can only identify congested road sections and cannot quantify the contribution of bottleneck causes (such as lane-changing conflicts and environmental factors), resulting in a lack of targeted traffic control strategies.

[0005] With the development of intelligent connected vehicle technology, technologies such as V2X communication, roadside sensors, and digital twins have provided new technical pathways for the comprehensive collection, real-time fusion, and dynamic evaluation of traffic data in tunnel clusters. The key to solving the problem of traffic efficiency in tunnel clusters lies in how to integrate multi-source intelligent connected vehicle data, construct a high-precision, high-real-time traffic capacity assessment model, and transform the assessment results into implementable traffic management strategies.

[0006] Therefore, a method for assessing the traffic capacity of tunnel groups based on intelligent connectivity is needed to address the aforementioned problems. Summary of the Invention

[0007] The purpose of this invention is to propose a method for assessing the traffic capacity of tunnel group road sections based on intelligent network connectivity in order to solve the above-mentioned problems.

[0008] To achieve the above objectives, the present invention adopts the following technical solution:

[0009] Methods for assessing the traffic capacity of tunnel group sections based on intelligent connected vehicles include:

[0010] The multi-source data fusion and digital twin construction steps are configured to integrate multi-source data into a holographic traffic flow model that maps physical scenes and digital mirrors in real time;

[0011] The dynamic capacity assessment model is configured to be built and calculated based on fused data and digital twin models, combined with tunnel engineering constraints, to construct an AI-driven dynamic assessment model that calculates the capacity, congestion risk and bottleneck location of each section of the tunnel group in real time.

[0012] The verification and correction steps for the evaluation results are configured to ensure the accuracy and reliability of the evaluation results through multi-dimensional verification. If the verification results do not meet the preset qualification standards, iterative optimization will be carried out until the accuracy requirements are met.

[0013] Preferably, the multi-source data acquisition and preprocessing step is configured to acquire CAV dynamic data, roadside sensor data, tunnel engineering parameter data and historical benchmark data of the entire tunnel group, and perform corresponding preprocessing on the acquired data to provide a high-quality data foundation for subsequent evaluation.

[0014] Preferably, the multi-source data fusion and digital twin construction specifically includes:

[0015] It adopts a two-tier architecture that integrates local edge computing with global cloud computing, wherein:

[0016] Local fusion at the edge: Using the roadside unit of each tunnel as the fusion node, the CAV data and roadside radar data of the road segment are integrated; the vehicle trajectory is corrected by combining the real-time detection results of the roadside radar, and the real-time vehicle position, speed, and distance local traffic flow data of the road segment are output.

[0017] Cloud-based global fusion: It integrates local data output from each edge node, environmental data collected by environmental monitoring instruments, and tunnel engineering parameter data across the entire domain; and outputs a global traffic flow feature matrix covering all road sections of the tunnel group.

[0018] Based on the fused global traffic flow feature matrix, a dynamically updatable digital twin model of the tunnel group is constructed, comprising three parts: geometric modeling, state mapping, and visualization.

[0019] Preferably, the construction and calculation of the dynamic traffic capacity assessment model specifically includes:

[0020] The following core features are extracted from the fused global traffic flow data: vehicle behavior features, traffic flow features, environmental features, road constraint features, event features, time features, CAV features, and historical features.

[0021] Preferably, the key indicators of the various features are as follows:

[0022] Vehicle behavior characteristics include average headway, acceleration fluctuations, and lane-changing frequency.

[0023] Traffic flow characteristics include real-time traffic volume, lane occupancy, and speed standard deviation.

[0024] Environmental characteristics: including visibility, road surface moisture, and CO concentration;

[0025] Road constraint features include: slope, curvature, lane width, and number of lanes;

[0026] Event characteristics: including event type, event duration, and number of lanes occupied;

[0027] Time characteristics: including time periods and holidays;

[0028] CAV characteristics include CAV coverage and CAV average speed.

[0029] Historical characteristics: including historical traffic capacity and historical frequency of congestion during the same period.

[0030] Preferably, the dual-model architecture combining LSTM time-series prediction and XGBoost bottleneck identification is used, wherein the LSTM time-series prediction model calculates real-time and short-term dynamic traffic capacity, and the XGBoost bottleneck identification model locates traffic capacity bottlenecks and their causes.

[0031] The LSTM time series prediction model first calculates the basic traffic capacity using the average headway as the core parameter, then obtains the actual traffic capacity through four types of correction coefficients: visibility, gradient, event, and CAV, and finally learns the time series change trend to predict future traffic capacity; the model training uses the mean absolute error as the loss function.

[0032] The XGBoost bottleneck identification model processing logic is as follows:

[0033] Input layer: Real-time feature vectors with input dimension

[22] ;

[0034] Output layer: Outputs bottleneck determination results and the contribution of the four types of causes;

[0035] Bottleneck determination: If the real-time capacity / traffic demand is ≤1.1 and the lane occupancy rate is ≥80%, it is determined to be a bottleneck;

[0036] Contribution of Causes: The formula is: Contribution of a certain type of cause = Sum of importance scores of the features corresponding to that type of cause / Sum of importance scores of all features × 100%;

[0037] The two work together to output a complete dynamic assessment result, and finally integrate and output a dynamic traffic capacity assessment report for the tunnel group, which includes the overall traffic capacity distribution, bottleneck list, and bottleneck cause analysis.

[0038] Preferably, the verification and correction of the evaluation results specifically includes:

[0039] Reference data independent of the evaluation data were selected, including reference data from traditional detectors and manually observed reference data, to ensure the objectivity and impartiality of the verification process.

[0040] Error analysis is performed on real-time traffic capacity to evaluate the error between the real-time traffic capacity output by the model and the actual traffic capacity obtained by traditional detectors / manual observation, specifically including absolute error and relative error;

[0041] Consistency checks are performed on bottleneck identification by comparing the bottleneck locations identified by the model with the congestion locations observed manually, and the consistency rate is calculated.

[0042] Preferably, if the verification result does not meet the qualification standard, iterative optimization is performed through the following three levels until the accuracy requirement is met:

[0043] Data-wise: For road sections with significant assessment errors, supplementary key data will be collected;

[0044] Model level: Adjust the feature weights and parameters of the model;

[0045] Iterative verification: After the model is corrected, the traffic capacity assessment results are recalculated, and the error analysis and consistency verification process is executed again.

[0046] Preferably, the evaluation results application and strategy optimization steps are configured to be based on three core evaluation outputs: real-time traffic capacity, short-term prediction results, and bottleneck location. These outputs are specifically applied to the traffic management of tunnel groups to improve the traffic efficiency and safety of tunnel groups, thus demonstrating the practical value of the evaluation method.

[0047] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0048] 1. This invention constructs an LSTM-XGBoost dual-model architecture by combining a multi-source data acquisition system with a two-level fusion architecture of edge and cloud. The LSTM model can calculate and predict the traffic capacity at four time points within the next 10 minutes in real time, while the XGBoost model accurately locates the bottleneck and quantifies the contribution of four types of causes. At the same time, the accuracy is ensured through multi-dimensional verification. It can cover the entire tunnel group and capture instantaneous traffic changes, providing a panoramic decision-making basis for traffic management based on real-time data, trend prediction, and bottleneck causes.

[0049] 2. This invention deeply applies the evaluation results to three core scenarios of tunnel group traffic management: entrance flow regulation uses V2X to issue reservation instructions to CAVs and variable message signs to guide non-connected vehicles to divert, thus avoiding congestion at the source; lane and speed guidance uses CAV lane suggestions and traffic light signs to optimize lane resource allocation and reduce differences in lane capacity; emergency response uses CAV speed / distance control and coordinated rescue to reduce the risk of accident escalation; after the strategy is implemented, the effect is monitored through multi-source data, and the data is fed back to the evaluation model to adjust feature weights and parameters, continuously improving the model accuracy and strategy effectiveness. Attached Figure Description

[0050] Further details, features, and advantages of this application are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which:

[0051] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0052] Several embodiments of this application will now be described in more detail with reference to the accompanying drawings to enable those skilled in the art to implement this application. This application may be embodied in many different forms and for various purposes and should not be limited to the embodiments set forth herein. These embodiments are provided to make this application thorough and complete, and to fully convey the scope of this application to those skilled in the art. The embodiments described do not limit this application.

[0053] Unless otherwise defined, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It will be further understood that terms such as those defined in commonly used dictionaries shall be interpreted as having a meaning consistent with their meaning in the relevant field and / or the context of this specification, and shall not be interpreted in an idealized or overly formal sense unless expressly defined herein.

[0054] Example 1

[0055] Its specific implementation method is combined with the appendix Figure 1 Please provide a detailed explanation.

[0056] Appendix Figure 1 The flowchart of the tunnel group road section traffic capacity assessment method based on intelligent network provided in the embodiments of the present invention shows the complete steps from multi-source data acquisition and preprocessing to the application of assessment results and strategy optimization.

[0057] In this embodiment, it includes:

[0058] The multi-source data acquisition and preprocessing steps are configured to acquire dynamic CAV data, roadside sensor data, tunnel engineering parameter data, and historical baseline data for the entire tunnel group, and perform corresponding preprocessing on the acquired data to provide a high-quality data foundation for subsequent evaluation. Specifically, this includes:

[0059] CAV dynamic data is collected in real time via V2X (DSRC / C-V2X) communication and includes three main categories: basic vehicle information, motion status data, and driving behavior data. Basic vehicle information includes vehicle ID, vehicle type (car / truck / bus), and vehicle dimensions (length / width). Motion status data includes high-precision GNSS position (error ≤ 1m), instantaneous speed (update frequency 1Hz), acceleration, heading angle, headway (distance to the vehicle in front), and turn signal / brake light status. Driving behavior data includes lane-changing intention (determined by turn signal triggering) and braking frequency (statistically counted by the number of times brake lights illuminate).

[0060] Roadside sensor data: Collected by roadside radars, video detectors, and environmental monitoring instruments deployed along the tunnel network, this data is mainly divided into traffic status data, environmental data, and event data. Traffic status data is used to statistically analyze the flow, speed, and queue length of non-connected vehicles, which can compensate for insufficient CAV coverage. Environmental data includes tunnel visibility (requiring special attention when ≤50m), CO concentration (reflecting ventilation status), road surface humidity (affecting braking distance in rainy weather), and lighting brightness. Event data is captured through radar and video recognition to identify sudden situations such as vehicle malfunctions (e.g., detecting stationary vehicles) and lane occupancy (e.g., emergency lane occupancy).

[0061] Tunnel engineering parameter data: Based on tunnel group design drawings and on-site surveys, these are all static physical parameters, serving as the core constraints of the evaluation model. Specifically, they include road parameters and facility parameters. Road parameters cover the alignment (radius of curvature), longitudinal slope (e.g., ±3%), cross-sectional width (lane width, emergency lane width), and number of lanes for each tunnel. Facility parameters include the length of acceleration / deceleration lanes at tunnel entrances / exits, the location of emergency refuge areas, and the rated efficiency of ventilation / lighting systems.

[0062] Historical baseline data: Collect data from traditional detectors (such as coils and video data) and manual observation data (such as peak-hour congestion duration and accident records) for the tunnel group over the past year. This type of data will serve as the basis for verifying the subsequent evaluation results and ensure that the evaluation accuracy meets the needs of the actual scenario.

[0063] Data preprocessing is used to remove noise and bias from the collected raw data, ensuring the consistency and usability of data from different sources. Preprocessing includes the following parts:

[0064] Time synchronization: Based on a unified timestamp (such as GPS time), align the collection time of CAV data and roadside sensor data, and control the time deviation within ≤100ms to avoid misjudgment of traffic conditions due to time asynchrony (such as mistaking the vehicle positions at different times as congestion at the same time).

[0065] Outlier removal: The 3σ principle or the isolated forest algorithm is used to filter out outlier data that does not conform to physical laws or is caused by interference. For example, the instantaneous speed of the CAV is 150 km / h, and its acceleration is 10... Data such as traffic flow displayed when there are no vehicles due to signal interference from roadside radar must be identified and removed to ensure data authenticity.

[0066] Standardized format: The formats of data from different sources are standardized and converted, including JSON format for CAV data, CSV format for roadside sensor data, and CAD format for engineering parameter data. All of these are converted into a unified spatiotemporal database format (such as PostGIS), and spatial coordinates are associated (such as mapping the vehicle's GNSS position to the road station number of the tunnel group), laying the foundation for subsequent fusion modeling.

[0067] The multi-source data fusion and digital twin construction steps are configured to integrate multi-source data into a holographic traffic flow model that maps physical scenes and digital mirrors in real time, providing a comprehensive, dynamic, and visual analysis platform for traffic capacity assessment.

[0068] Specifically, it includes:

[0069] It adopts a two-tier architecture that integrates local edge computing with global cloud computing, wherein:

[0070] Local fusion at the edge (segment level): Using the roadside unit (RSU) of each tunnel as the fusion node, the CAV data and roadside radar data of the segment are integrated. The position deviation of CAV caused by GPS signal attenuation in the tunnel is eliminated by Kalman filtering algorithm. The vehicle trajectory is corrected by combining the real-time detection results of roadside radar (for example, the original position error of CAV of 5m is corrected to ≤1.5m). Finally, the real-time position, speed, and distance of vehicles in the segment are output as local traffic flow data.

[0071] Cloud-based full-domain fusion (tunnel group level): Local data output from each edge node, environmental data collected by environmental monitoring instruments, and tunnel engineering parameter data are uniformly uploaded to the regional cloud platform and fused across the entire domain. The fusion process adopts a deep learning fusion algorithm: First, the spatial features of multi-source data (such as vehicle distribution of CAV and lane layout of tunnels) are extracted through a 3DCNN model. Then, different data are assigned differentiated weights through an attention mechanism (for example, CAV data accounts for 70% of the weight in micro-vehicle behavior analysis, and roadside video data accounts for 60% of the weight in non-connected vehicle statistics). Finally, a full-domain traffic flow feature matrix covering all sections of the tunnel group (including entrances, tunnel interiors, exits, and intersections) is output.

[0072] Based on the fused global traffic flow feature matrix, a dynamically updatable digital twin model of the tunnel group is constructed, comprising three parts: geometric modeling, state mapping, and visualization.

[0073] Geometric modeling: Recreate the physical structure of the tunnel group in digital space at a 1:1 scale, including tunnel alignment, lane distribution, and emergency facility locations, to ensure that the digital model perfectly matches the real-world scenario.

[0074] State mapping: Real-time traffic flow data (such as vehicle position and speed) and environmental data (such as visibility) are loaded into the digital model in real time to realize the synchronous movement of vehicles in the real world and vehicles in the digital model; dynamic mapping of the decrease in visibility in the tunnel in reality and the synchronous darkening of light in the tunnel in the digital model, to achieve real-time linkage between physical scene and digital mirror.

[0075] Visualization: Using WebGL or Unity engine, the traffic status of the entire tunnel group is displayed in 3D. For example, green indicates unobstructed lanes, yellow indicates slow-moving lanes, and red indicates congested lanes, allowing managers to intuitively view the traffic conditions of each section and providing visual support for subsequent assessment and decision-making.

[0076] The dynamic capacity assessment model is configured to be built and calculated based on fused data and digital twin models, combined with tunnel engineering constraints, to construct an AI-driven dynamic assessment model that calculates the capacity, congestion risk and bottleneck location of each section of the tunnel group in real time.

[0077] Specifically, it includes:

[0078] The following (8 categories) core features are extracted from the fused traffic flow data to ensure that the model conforms to both the data patterns and the actual constraints of tunnel engineering. These features include vehicle behavior features, traffic flow features, environmental features, road constraint features, event features, time features, CAV features, and historical features.

[0079] The key indicators for each feature are as follows:

[0080] Vehicle behavior characteristics include average headway, acceleration fluctuations, and lane-changing frequency.

[0081] Average headway: The average time difference (in seconds) of all adjacent CAVs passing through the same cross section within a certain road segment. It is calculated using high-precision CAV positioning data and timestamps (e.g., the times when CAV1 and CAV2 pass through tunnel cross section K12+300 are t1 and t2 respectively, and the headway h = t2 - t1. The average headway is the average of all valid h values ​​within 5 minutes).

[0082] Acceleration fluctuation: Standard deviation of CAV acceleration within a certain road segment (unit: This reflects the smoothness of vehicle movement (the greater the fluctuation, the more frequent the acceleration and deceleration, and the lower the traffic efficiency), and is calculated using acceleration data uploaded in real time by CAV.

[0083] Lane change frequency: The number of lane changes by CAV within a unit of time (5 minutes) in a certain road segment (unit: times / 5min), determined by the status and position changes of CAV turn signals (e.g., if the CAV turn signal is on and the lateral displacement exceeds 1.5m, it is determined as 1 lane change).

[0084] Traffic flow characteristics include real-time traffic volume, lane occupancy, and speed standard deviation.

[0085] Real-time traffic flow: The total number of standard passenger cars equivalent in a certain road section within 5 minutes (unit: pcu / 5min), calculated by fusing CAV data (converted to pcu by vehicle type, such as 1 truck = 2.5 pcu) with roadside video detector data;

[0086] Lane occupancy rate: The percentage of time vehicles occupy a lane within a certain road segment (unit: %), calculated using data from roadside loop detectors;

[0087] Speed ​​standard deviation: The standard deviation of the instantaneous speed of all vehicles in a certain road segment (unit: km / h), reflecting the consistency of vehicle speed (the larger the speed standard deviation, the greater the difference in vehicle speed, and the more likely a conflict will occur), is calculated by fusing CAV data and roadside radar data.

[0088] Environmental characteristics: including visibility, road surface moisture, and CO concentration;

[0089] Visibility: Horizontal visibility distance inside the tunnel (unit: m), collected in real time by a laser visibility meter (updated every 10 seconds);

[0090] Road surface moisture: The moisture content of the road surface inside the tunnel (unit: %), which is collected by a road surface moisture sensor (moisture content > 20% is considered a wet and slippery road surface);

[0091] CO concentration: Carbon monoxide concentration in the tunnel (unit: ppm). When the CO concentration is collected by an environmental monitoring instrument (CO concentration > 100 ppm), ventilation needs to be activated, which indirectly affects the efficiency of vehicle traffic.

[0092] Road constraint features include: slope, curvature, lane width, and number of lanes;

[0093] Gradient: The gradient value of the tunnel's longitudinal slope (unit: %, positive for uphill and negative for downhill), obtained based on the tunnel engineering design drawings;

[0094] Curvature: The radius of curvature of the tunnel alignment (unit: m), obtained based on the design drawings (curvature < 500m is a small radius curve, and the effect of vehicle turning and deceleration needs to be considered).

[0095] Lane width: The effective width of a single lane inside the tunnel (unit: m), obtained from on-site survey (standard lane width is 3.75m, and the traffic capacity decreases by 5%-8% for every 0.25m reduction in width).

[0096] Number of lanes: The number of effective lanes in the tunnel (unit: lanes), obtained based on design drawings (the more lanes, the higher the basic traffic capacity, but interference between lanes must be considered).

[0097] Event characteristics: including event type, event duration, and number of lanes occupied;

[0098] Event type: Category of emergencies (discrete value: 1=vehicle malfunction, 2=traffic accident, 3=lane occupancy, 4=no event), identified through joint identification of roadside video and radar;

[0099] Event duration: The time from the occurrence of the event to its completion (unit: min, updated in real time if not handled).

[0100] Lane occupancy count: The number of lanes occupied by the event (unit: lane, e.g., 1 lane occupied is counted as 1, 2 lanes occupied is counted as 2).

[0101] Time characteristics: including time periods (morning peak / off-peak / evening peak) and holidays;

[0102] Time Period Type: Traffic Time Period Division within a Day (Discrete Values: 1 = Morning Peak 7:00-9:00, 2 = Off-Peak 9:00-17:00, 3 = Evening Peak 17:00-19:00, 4 = Night 19:00-7:00);

[0103] Holiday indicator: Whether it is a holiday (Boolean value: 1 = holiday / weekend, 0 = workday).

[0104] CAV characteristics include CAV coverage and CAV average speed.

[0105] CAV coverage: The proportion of CAVs to the total number of vehicles in a certain road segment;

[0106] CAV average speed: The average speed of CAVs within a certain road segment (unit: km / h), calculated from CAV dynamic data.

[0107] Historical characteristics: including historical traffic capacity and historical frequency of congestion during the same period;

[0108] Historical traffic capacity during the same period: The average traffic capacity of this road segment during the same time period in the past month (such as the morning rush hour from 7:00 to 7:05 last week) (unit: pcu / h);

[0109] Historical congestion frequency: The number of times this road segment experienced congestion during the same period within the past month (unit: times).

[0110] A dual-model architecture combining LSTM time-series prediction and XGBoost bottleneck identification is adopted to achieve a comprehensive evaluation of real-time capacity calculation, future trend prediction, and bottleneck location. The LSTM time-series prediction model calculates real-time and short-term dynamic capacity, while the XGBoost bottleneck identification model locates capacity bottlenecks and their causes.

[0111] The processing logic of the LSTM time series prediction model is as follows:

[0112] Input layer:

[0113] The input is a time-series feature matrix with dimension 1. ,in:

[0114] The time window length (i.e., the number of historical time steps input) is set to 10 (corresponding to 5 minutes, 1 time step every 30 seconds, for a total of 10 time steps).

[0115] For each time step, the feature dimension is set to 22 (8 categories of features with a total of 22 specific indicators, such as 3 vehicle behavior features, 3 traffic flow features, ... 2 historical features, totaling 3+3+3+4+3+2+2+2=22).

[0116] Input data must first undergo standardization to eliminate dimensional differences;

[0117] Output layer:

[0118] The output is a traffic capability vector with dimension 1. ,in:

[0119] To determine the number of predicted time points, we take 4 (including 1 real-time traffic capacity and 3 short-term predicted traffic capacities, corresponding to the next 3 minutes, 7 minutes, and 10 minutes, respectively).

[0120] Output value unit: pcu / h (standard passenger car equivalent / hour), needs to be denormalized to restore to the actual traffic capacity value, the formula is:

[0121]

[0122] in, The standardized traffic capability value output by the model. , To train the maximum and minimum values ​​of the centralized traffic capacity.

[0123] The output of the LSTM model is the dynamic accessibility that integrates multiple features. Its calculation process consists of three steps: basic accessibility calculation, multi-factor correction, and time series prediction, as follows:

[0124] Using average headway as the core parameter, the theoretical capacity (i.e., basic capacity) under unconstrained conditions is calculated. The formula is as follows:

[0125]

[0126] in,

[0127] This represents the average headway.

[0128] Number of lanes;

[0129] The lane width correction factor (calculated based on lane width w, conforming to traffic engineering specifications) is given by the following formula:

[0130]

[0131] when When it equals 3.75 meters (standard lane width), It equals 1; When it is equal to 3.5 meters, It equals 0.9; When it is equal to 4 meters, It equals 1.1.

[0132] Basic traffic capacity The actual traffic capacity needs to be determined by adjusting four types of features: environment, road, event, and CAV. The corrected formula is as follows:

[0133]

[0134] in,

[0135] This is the visibility correction factor (calculated based on visibility V, quantifying the impact of visibility on vehicle speed);

[0136] This is the slope correction factor (calculated based on slope i, quantifying the impact of slope on vehicle dynamics).

[0137] Event correction factor (based on lane occupancy) (Calculate and quantify the impact of events on lane resource usage) ,in The effective number of lanes is multiplied by 0.9 to reserve a 10% safety margin (to avoid excessive vehicle density in the effective lanes).

[0138] CAV enablement factor (based on CAV coverage) (Calculate and quantify the optimization of traffic order by CAV)

[0139]

[0140] in CAV coverage (unit: %), with a coefficient of 0.003 indicating that for every 1% increase in CAV coverage, the traffic capacity increases by 0.3%.

[0141] LSTM learns from the past 5 minutes Based on temporal and characteristic trends (e.g., traffic flow increasing at 10 pcu / min, visibility decreasing at 5 m / min), predict traffic capacity for the next 3-10 minutes. ( =3, 7, 10 min).

[0142] The core of the prediction process is to minimize the error between the predicted and actual values. Model training uses the mean absolute error (MAE). As the loss function, the formula is:

[0143]

[0144] in,

[0145] The number of samples in the training set;

[0146] Let be the predicted traffic capacity of the j-th sample;

[0147] Let be the actual throughput capacity of the j-th sample.

[0148] The Adam optimizer (with an initial learning rate of 0.001, decaying by 10% every 5 epochs) is used for iterative training until the loss function converges, ensuring that the prediction accuracy meets engineering requirements.

[0149] The LSTM model ultimately outputs four traffic capability values, in the following format:

[0150] Real-time throughput (t=0min): 1690 pcu / h;

[0151] Short-term predicted traffic capacity (t=3min): 1580 pcu / h (due to the continuous decrease in visibility, the predicted traffic capacity is reduced);

[0152] Short-term predicted capacity (t=7min): 1450 pcu / h;

[0153] Short-term predicted traffic capacity (t=10min): 1720pcu / h (traffic capacity recovers after the predicted event is handled).

[0154] The XGBoost bottleneck identification model processing logic is as follows:

[0155] Input layer:

[0156] The input is a real-time feature vector with dimension . ,in =22 (consistent with the single-time-step feature dimension of LSTM, including 22 indicators in 8 categories such as vehicle behavior, traffic flow, and environment), and the input data must be standardized in the same way as LSTM.

[0157] Output layer:

[0158] The output is the bottleneck assessment result, which includes two types of output:

[0159] Bottleneck determination results (binary classification): ∈{0, 1} (0 = non-bottleneck, 1 = bottleneck);

[0160] Bottleneck Cause Contribution (Multidimensional Regression): Outputs the contribution of 4 types of causes (unit: %, totaling 100%), including lane-changing conflict contribution, truck lane occupation contribution, event impact contribution, and environmental factor contribution.

[0161] The core definition of a bottleneck is that the capacity of a certain road segment cannot meet the current traffic demand, resulting in traffic stagnation or slow traffic. XGBoost determines this by combining capacity, demand ratio, and lane occupancy threshold. The specific formula is as follows:

[0162]

[0163] in,

[0164] Real-time throughput capability for LSTM output;

[0165] The current traffic demand for this road segment (unit: pcu / h) is calculated from the real-time traffic flow (unit: pcu / 5min), i.e., hourly traffic flow.

[0166] Lane occupancy rate;

[0167] This indicates that the traffic capacity can only meet up to 1.1 times the demand (no redundancy). This indicates that the traffic flow is already saturated; if both conditions are met simultaneously, it is considered a bottleneck. =1 (determined as a bottleneck).

[0168] The logic for calculating the contribution of bottleneck causes is as follows:

[0169] The contribution of each factor to the bottleneck is quantified by feature importance scores. The core is to calculate the proportion of feature importance corresponding to a certain type of factor to the total importance, as shown in the following formula:

[0170]

[0171] in,

[0172] The contribution of the k-th type of cause (k=1,2,3,4, corresponding to lane-changing conflict, truck lane occupation, event impact, and environmental factors);

[0173] This is the feature set corresponding to the k-th type of cause (e.g., event impact corresponds to three features: event type, event duration, and number of lanes occupied; environmental factors correspond to three features: visibility, road surface humidity, and CO concentration).

[0174] The importance score for feature f is calculated using the XGBoost Gain metric, which is the total contribution of the feature to improving the classification / regression performance of the model across all decision trees. The higher the Gain value, the greater the impact of the feature on the results.

[0175] This is the set of all 22 input features.

[0176] The XGBoost model ultimately outputs the bottleneck determination and its contribution as a cause, in an example format:

[0177] Bottleneck determination result: 1 (the entrance section of Tunnel 3, K15+200, is the bottleneck);

[0178] Bottleneck causes and their contribution: lane-changing conflicts 8%, trucks occupying lanes 2%, incident impact 65% (vehicle breakdown occupying 1 lane), environmental factors 25% (visibility 80m).

[0179] The two work together to output a complete dynamic assessment result, and finally integrate and output a dynamic traffic capacity assessment report for the tunnel group, which includes three core contents: overall traffic capacity distribution, bottleneck list, and bottleneck cause analysis.

[0180] Overall traffic capacity distribution: Real-time traffic capacity and predicted value for the next 10 minutes for each tunnel and section (unit: pcu / h).

[0181] Bottleneck list: Location (station number) of all road sections identified as bottlenecks, bottleneck start time, and current capacity to demand ratio;

[0182] Bottleneck cause analysis: The contribution of four types of causes to each bottleneck is marked as the primary cause (contribution > 50%) and the secondary cause (contribution 10%-50%).

[0183] The verification and correction steps for the evaluation results are configured to ensure the accuracy and reliability of the evaluation results through multi-dimensional verification. If the verification results do not meet the preset qualification standards, iterative optimization will be carried out until the accuracy requirements are met.

[0184] Specifically, it includes:

[0185] Reference data independent of the evaluation data were selected, including reference data from traditional detectors and manually observed reference data, to ensure the objectivity and impartiality of the verification process.

[0186] Traditional detector benchmark data: Extract concurrent data from loop detectors and video analysis systems deployed along the tunnel network, such as traffic flow statistics from loop detectors and average speed from video recognition. This type of data can serve as a verification benchmark for macroscopic traffic conditions, used to compare and evaluate the accuracy of the macroscopic traffic capacity output by the model.

[0187] Manual observation baseline data: Organize on-site observation teams to conduct manual observations at key locations such as tunnel entrances, exits, and intersections. Obtain ground truth data by manually counting traffic flow, measuring actual vehicle speeds with speedometers, and recording congestion lengths with video recordings. This data serves as a verification baseline for micro-assessment results (such as vehicle headway and lane-changing frequency).

[0188] Error analysis is performed on real-time traffic capacity to evaluate the error between the real-time traffic capacity output by the model and the actual traffic capacity obtained by traditional detectors / manual observation. Specifically, this includes absolute error and relative error. The absolute error is the absolute difference between the assessed value and the actual value, and the relative error is the ratio of the absolute error to the actual value multiplied by 100%. The acceptable standard is set as a relative error ≤ 10%, that is, the deviation between the assessed value and the actual value must be controlled within an acceptable range to ensure the accuracy of real-time traffic capacity assessment.

[0189] A consistency check is performed on bottleneck identification by comparing the bottleneck locations identified by the model with the congestion locations observed manually, and calculating the consistency rate (i.e., the ratio of the number of consistent bottlenecks to the total number of bottlenecks multiplied by 100%). A pass / fail standard is set at a consistency rate ≥ 85% to ensure that the model can accurately capture congestion bottlenecks in real-world scenarios and avoid bottleneck location errors.

[0190] If the verification results do not meet the passing standard, iterative optimization will be performed through the following three levels until the accuracy requirements are met:

[0191] At the data level: For road sections with large assessment errors, supplementary key data should be collected; for example, if the assessment error of a tunnel exceeds the standard due to low CAV coverage, the density of roadside radar on that road section can be temporarily increased to make up for the lack of CAV data; if the lack of environmental data leads to assessment deviation, environmental monitoring instruments can be added to collect data such as visibility and humidity.

[0192] At the model level: Adjust the feature weights and parameters of the model; for example, if the impact of a certain tunnel slope on traffic capacity is underestimated, the weight of this feature in the LSTM model can be increased; if the bottleneck identification accuracy of the XGBoost model is insufficient, the tree depth, learning rate and other parameters of the model can be optimized to improve the feature recognition capability.

[0193] Iterative verification: After the model is corrected, the traffic capacity assessment results are recalculated, and the error analysis and consistency verification process is executed again. If the results are still not up to standard, the above correction steps are repeated until the relative error of the assessment results is ≤10% and the consistency rate is ≥85%, ensuring that the model accuracy meets the actual application requirements.

[0194] The evaluation results and strategy optimization steps are configured based on three core evaluation outputs: real-time traffic capacity, short-term prediction results, and bottleneck location. These are specifically applied to traffic management of tunnel groups to improve traffic efficiency and safety, demonstrating the practical value of the evaluation method.

[0195] Based on three core assessment outputs—real-time traffic capacity, short-term forecast results, and bottleneck location—this approach is specifically applied to traffic management within tunnel clusters, covering three main scenarios:

[0196] Scenario 1: Entrance Flow Regulation (Congestion Prevention): When the assessment results indicate that the downstream tunnel capacity is about to decrease, and the current entrance inflow exceeds the future capacity (for example, the capacity of Tunnel 2 exit will drop to 1200 pcu / h in 3 minutes, while the current entrance inflow has reached 1800 pcu / h), the entrance flow regulation strategy is immediately triggered. Implementation methods include: issuing reservation instructions to CAVs via the traffic control system (e.g., you need to enter Tunnel 2 in 2 minutes) to avoid concentrated vehicle influx; displaying temporary flow restrictions at the Tunnel 2 entrance to non-connected vehicles via variable message signs (VMS), suggesting detours via Tunnel 3, guiding some vehicles to divert, and ultimately controlling the entrance flow within the capacity of the downstream tunnel (e.g., below 1200 pcu / h), preventing congestion from the source.

[0197] Scenario 2: Lane and Speed ​​Guidance (Improving Efficiency): When the assessment results show significant differences in the capacity of different lanes within the tunnel (for example, lane 3 in tunnel 1 has a capacity of only 1000 pcu / h due to the high proportion of trucks (40%), while lanes 1-2 have a capacity of 1800 pcu / h), lane and speed guidance strategies are triggered. Implementation methods include: issuing instructions to CAVs suggesting they enter lane 1 or 2 to guide traffic flow towards lanes with higher capacity; illuminating the dedicated truck sign for lane 3 using lane control signals to prevent passenger cars from occupying truck lanes; and simultaneously issuing a suggested speed of 70 km / h to all vehicles on the entire road segment based on real-time capacity to reduce acceleration and deceleration conflicts caused by speed differences and improve overall traffic efficiency.

[0198] Scenario 3: Emergency Response (Risk Reduction): When the assessment results identify a sudden accident / malfunction in the tunnel that significantly reduces traffic capacity (e.g., a vehicle malfunction in the middle section of tunnel 3 occupies one lane, reducing traffic capacity by 50%), an emergency response strategy is triggered. Execution methods include: issuing a command to the upstream CAV to reduce speed to 40 km / h and maintain a 50m following distance to avoid rear-end collisions; issuing a command to the downstream CAV to accelerate away from the malfunction area to reduce vehicle congestion near the malfunction point; simultaneously coordinating with the tunnel monitoring center to notify rescue vehicles to handle the situation, and displaying the malfunction in the middle section of tunnel 3 to non-connected vehicles via VMS, urging them to avoid the area and reducing the risk of the accident escalating.

[0199] The optimization mechanism is as follows:

[0200] Effectiveness Evaluation: After the strategy is implemented, traffic status changes are monitored in real time through a multi-source data collection system to quantify the strategy's effectiveness. For example, when evaluating the entrance flow restriction strategy, it is monitored whether the congestion length at the tunnel 2 exit has decreased from 500m to less than 100m; when evaluating the lane guidance strategy, it is monitored whether the capacity difference between the lanes in tunnel 1 has decreased from 800pcu / h to less than 300pcu / h, thereby determining whether the strategy has achieved its expected goals.

[0201] Closed-loop feedback: The effect evaluation data is re-inputted into the dynamic capacity assessment model of step 3 as feedback data to optimize model parameters and feature design. For example, if the entrance flow restriction strategy increases tunnel capacity by 20%, the effect features of this strategy (such as flow restriction duration and capacity improvement rate) can be added to the model to improve the accuracy of future capacity prediction; if the emergency response strategy shortens the fault impact time, the emergency response speed can be incorporated as a new feature into the model to optimize the bottleneck cause analysis.

[0202] Continuous iteration: Through a cycle of evaluating results, assessing strategy effectiveness, optimizing the model based on feedback data, and re-evaluating, the accuracy of the evaluation model and the effectiveness of the control strategy are continuously improved. For example, after the initial implementation of the lane guidance strategy, if some vehicles are found not to change lanes as instructed, vehicle lane-changing intention features can be added to subsequent model optimizations, and the instruction issuance logic can be adjusted to gradually achieve long-term dynamic optimization of the tunnel group's capacity, continuously improving traffic efficiency and safety.

[0203] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0204] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

[0205] It should be noted that, in this document, the use of relational terms such as "first" and "second" is merely for distinguishing one entity or operation from another, and does not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "include," "contain," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the statement "includes a…" does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element.

[0206] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0207] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0208] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0209] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0210] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0211] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0212] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for evaluating the capacity of a tunnel group road section based on intelligent networking, characterized in that, Comprise: The multi-source data fusion and digital twin construction step is configured to integrate multi-source data into a holographic traffic flow model that maps the physical scene and digital mirror in real time; Specifically comprising: Adopting a two-level architecture of edge local fusion and cloud global fusion, wherein: Edge local fusion: taking the roadside unit of each tunnel as the fusion node, focusing on integrating CAV data and roadside radar data of the road section; combining the real-time detection results of roadside radar to correct the vehicle motion trajectory, outputting the real-time position, speed, and vehicle distance local traffic flow data of the vehicle on the road section; Cloud global fusion: integrating local data output by each edge node, environmental data collected by environmental monitors, and tunnel engineering parameter data, and performing global data fusion; outputting a global traffic flow feature matrix covering all road sections of the tunnel group; Based on the fused global traffic flow feature matrix, a dynamically updated tunnel group digital twin model is constructed, including geometric modeling, state mapping, and visual presentation; The dynamic traffic capacity evaluation model construction and calculation step is configured to construct an AI-driven dynamic evaluation model based on fused data and digital twin models, combined with tunnel engineering constraints, to calculate the traffic capacity, congestion risk, and bottleneck location of each road section of the tunnel group in real time; The evaluation result verification and correction step is configured to ensure the accuracy and reliability of the evaluation results through multi-dimensional verification. If the verification result does not meet the preset qualified standard, iterative optimization is performed until the accuracy requirement is met. 2.The intelligent network-based tunnel group road section capacity evaluation method according to claim 1, characterized in that, The multi-source data acquisition and preprocessing step is configured to obtain global data of the tunnel group, including CAV dynamic data, roadside sensor data, tunnel engineering parameter data, and historical reference data, and to preprocess the obtained data accordingly, providing a high-quality data foundation for subsequent evaluation. 3.The intelligent network-based tunnel group road section capacity evaluation method according to claim 1, characterized in that, Dynamic traffic capacity evaluation model construction and calculation, specifically including: The following core features are extracted from the fused global traffic flow data, including vehicle behavior features, traffic flow features, environmental features, road constraint features, event features, time features, CAV features, and historical features.

4. The intelligent network-based tunnel group road section capacity evaluation method according to claim 3, characterized in that, The key indicators of each type of feature are as follows: Vehicle behavior features: including average headway, acceleration fluctuation, and lane changing frequency; Traffic flow features: including real-time traffic volume, lane occupancy rate, and speed standard deviation; Environmental features: including visibility, road surface humidity, and CO concentration; Road constraint features: including slope, curvature, lane width, and number of lanes; Event features: including event type, event duration, and number of lane occupants; Time features: including time period and holidays; CAV features: including CAV coverage rate and CAV average speed; Historical features: including same-period historical traffic capacity and historical congestion frequency.

5. The intelligent network-based tunnel group road section capacity evaluation method according to claim 4, characterized in that, Adopting a dual-model architecture combining LSTM time series prediction and XGBoost bottleneck identification, wherein the LSTM time series prediction model calculates real-time and short-term dynamic traffic capacity, and the XGBoost bottleneck identification model locates traffic capacity bottlenecks and causes; Among them, the LSTM time series prediction model first calculates the basic traffic capacity with the average headway as the core parameter, then gets the actual traffic capacity through four types of correction coefficients of visibility, slope, event and CAV, and finally learns the time series trend to predict the future traffic capacity; The average absolute error is used as the loss function for model training. The XGBoost bottleneck identification model processing logic is as follows: Input layer: real-time feature vector with input dimension [22]; Output layer: output bottleneck judgment result and 4 types of contribution degree; Bottleneck judgment: if real-time traffic capacity / traffic demand ≤1.1 and lane occupancy rate ≥80%, it is judged as bottleneck; Cause contribution degree calculation: the calculation formula is contribution degree of a certain type of cause = total importance score of the feature corresponding to the cause / total importance score of all features ×100%; Both of them output complete dynamic evaluation results, output tunnel group dynamic traffic capacity evaluation report, including global traffic capacity distribution, bottleneck list and bottleneck cause analysis.

6. The intelligent network-based tunnel group road section capacity evaluation method according to claim 5, characterized in that, The verification and correction of the evaluation results include: Selecting benchmark data independent of the evaluation data, including traditional detector benchmark data and manual observation benchmark data, to ensure the objectivity and fairness of the verification process: Error analysis of real-time traffic capacity, to evaluate the error between the real-time traffic capacity output by the model and the actual traffic capacity obtained by traditional detectors / manual observation, including absolute error and relative error; Consistency test of bottleneck identification, to compare the bottleneck position identified by the model with the congestion position observed by manual observation, and calculate the consistency rate.

7. The intelligent network-based tunnel group road section capacity evaluation method according to claim 6, characterized in that, If the verification result does not meet the qualified standard, the following three levels are used for iterative optimization until the accuracy requirement is met: Data level: supplement key data for road sections with large evaluation errors; Model level: adjust the feature weight and parameter of the model; Iterative verification: recalculate the traffic capacity evaluation results after model correction, and perform error analysis and consistency test process again. 8.The intelligent network-based tunnel group road section capacity evaluation method according to claim 1, wherein, The evaluation result application and strategy optimization steps are configured based on three types of core evaluation outputs: real-time traffic capacity, short-term prediction results and bottleneck position, which are applied to the traffic management of tunnel group to improve the traffic efficiency and safety of tunnel group, and reflect the practical value of the evaluation method.

Images