Tunnel interflow overpass line shape suitability evaluation method for autonomous vehicle

Abstract

This invention relates to a method for evaluating the drivability of tunnel interchange alignments for autonomous vehicles, belonging to the field of autonomous vehicle technology. The method involves collecting alignment, vehicle, and traffic flow parameters; defining static and dynamic occlusion relationships and constructing parameterized occlusion sub-scenarios; establishing required stopping sight distances according to driving automation levels; constructing a sight distance failure function based on structural reliability theory to calculate the probability of sight distance failure at multiple levels; constructing a three-dimensional sight distance risk sample set; using an adaptive multi-prototype competitive learning algorithm for clustering; and adaptively determining risk categories and prototypes through low-frequency seed point deletion and sub-cluster density interval splitting; establishing a mapping between risk prototypes and drivability risk levels based on failure probability sorting; and outputting the highest risk level of each sub-scenarios as the overall drivability risk level. This invention comprehensively considers multiple occlusions, eliminates the need for pre-setting the number of levels, objectively quantifies sight distance risk, and provides a basis for alignment optimization and autonomous driving testing.

Description

Technical Field

[0001] This invention belongs to the field of autonomous vehicle technology, specifically relating to a method for evaluating the drivability of tunnel interchange alignments for autonomous vehicles. Background Technology

[0002] Tunnel interchanges are complex traffic facilities located within confined underground or mountain tunnel spaces, incorporating main lines, ramps, and merging / diverting connections. Due to factors such as tunnel walls, curved sections, ramp connection angles, merging / diverting noses, and obstruction from vehicles traveling in different lanes within the tunnel, vehicles are prone to experiencing continuously changing static and dynamic obstructions while traveling on the main line, ramps, and merging / diverting zones. For autonomous vehicles, these obstructions not only affect the driver's direct visibility but also limit the effective detection range of onboard sensors, causing the vehicle's actual line-of-sight to vary depending on the road alignment, traffic composition, vehicle speed, sensor field of view, and level of automation.

[0003] Existing road sight distance testing methods typically calculate stopping sight distance using deterministic design speed, reaction time, and braking parameters, and determine the outcome based on whether the geometric sight distance meets regulatory requirements. These methods are primarily geared towards human driving conditions and struggle to simultaneously characterize differences in autonomous driving system perception and response, driver takeover, preset deceleration, and sensor configuration. Furthermore, autonomous driving tests of tunnel interchanges often employ a limited number of typical scenarios or pre-defined risk thresholds, making the evaluation results susceptible to influence from scenario selection and manual grading rules.

[0004] Therefore, it is necessary to establish a method that can comprehensively consider the alignment information of tunnel interchanges, autonomous vehicle information, and traffic flow information. This method can calculate the line-of-sight failure probability using structural reliability theory under multiple types of occlusion sub-scenarios, and adaptively determine the number of risk categories and risk prototypes based on the actual distribution of line-of-sight risk samples. Furthermore, it can output the drivability risk level of the occlusion sub-scenarios and the tunnel interchange as a whole, thereby providing a basis for alignment design optimization, traffic organization, vehicle perception system configuration, and autonomous vehicle testing. Summary of the Invention

[0005] The purpose of this invention is to solve the problems that existing methods cannot simultaneously consider multiple types of occlusion, different levels of autonomous driving and parameter uncertainties, and that the number of risk levels and classification thresholds depend on manual presets. This invention provides a method for evaluating the drivability of tunnel interchange alignments for autonomous vehicles.

[0006] To achieve the above objectives, the technical solution of the present invention is: a method for evaluating the drivability of tunnel interchange alignments for autonomous vehicles, comprising:

[0007] Step S1: Collect the alignment parameters, autonomous vehicle parameters, and traffic flow parameters for the tunnel interchange scenario;

[0008] Step S2: Define multiple types of occlusion relationships based on the linear parameters. These relationships include static and dynamic occlusion scenarios. The static occlusion scenario consists of the tunnel wall structure, ramp connection angles, and vehicle positions as the main subjects of sight distance verification. The dynamic occlusion scenario consists of obstacle vehicles traveling on the left side of the main line, obstacle vehicles traveling on the right side of the main line, obstacle vehicles traveling in the merging and diverging zones and about to enter the main line or ramps, obstacle vehicle speeds, and the positions and speeds of vehicles as the main subjects of sight distance verification. Based on these multiple types of occlusion relationships, construct several occlusion sub-scenes in the virtual simulation platform and parameterize each sub-scene.

[0009] Step S3: Establish the required parking sight distance for different driving automation levels, construct the sight distance failure function using structural reliability theory, and calculate the sight distance failure probability of each occlusion sub-scenario under different driving automation levels.

[0010] Step S4: Construct three-dimensional line-of-sight risk samples based on the line-of-sight failure probabilities of each occluded sub-scene under different levels of driving automation. The line-of-sight risk sample set is composed of all three-dimensional line-of-sight risk samples. The line-of-sight risk sample set is clustered using an adaptive multi-prototype competitive learning algorithm. During the clustering process, a low-frequency seed point deletion operation is performed based on the number of wins of the seed point, and a sub-cluster splitting operation is performed based on the local density and density interval of the sub-cluster, so as to adaptively determine the number of line-of-sight risk categories and the risk prototypes corresponding to each line-of-sight risk category.

[0011] Step S5: Sort each sight risk category in ascending order of sight failure probability, and establish a mapping relationship between each risk prototype and the drivability risk level; for each occlusion sub-scenario under the evaluation condition, determine the drivability risk level of each occlusion sub-scenario based on the mapping relationship, and determine the highest drivability risk level among all occlusion sub-scenarios as the overall drivability risk level of the tunnel interchange under the evaluation condition, and output the drivability evaluation result.

[0012] Furthermore, in step S1,

[0013] The alignment parameters include the number of tunnel lanes, the length of the straight lanes, the length of the entrance ramp lanes, the length of the exit ramp lanes, the radius of the ramp curve, the length of the acceleration lane, the length of the deceleration lane, the position of the merging and diverging noses, the design speed of the tunnel mainline, the design speed of the tunnel ramps, the design speed of the acceleration lane, and the design speed of the deceleration lane.

[0014] The parameters of the autonomous vehicle include the level of automation, vehicle speed, perception-braking reaction time of the autonomous driving system, braking deceleration, preset braking deceleration, driver takeover reaction time, driver takeover time, and the configuration and deployment scheme of the on-board perception sensors. The configuration and deployment scheme of the on-board perception sensors includes the type of perception sensor, detection distance, horizontal field of view, vertical field of view, horizontal angular resolution, vertical angular resolution, installation position, and obstacle perception algorithm.

[0015] The traffic flow parameters include the traffic components, vehicles, and vehicle speeds.

[0016] Furthermore, in step S2, the obstacle vehicle includes large trucks and small vehicles; a combination of a set of linear parameters, autonomous vehicle parameters, and traffic flow parameters is defined as a set of evaluation conditions. Under the nth set of evaluation conditions, multiple occlusion sub-scenes are established according to the obstacle vehicle type, the obstacle vehicle location, and the obstacle vehicle movement state, and are numbered sequentially; the parameterization setting includes assigning values ​​to the vehicle position, driving speed, and sensor detection range in each occlusion sub-scene according to the autonomous vehicle parameters and traffic flow parameters collected in step S1.

[0017] Furthermore, in step S3, the required parking sight distance is established for different levels of driving automation, and the specific calculation formula is as follows:

[0018]

[0019] In the formula, The required stopping sight distance for Level j driving automation; The vehicle's speed; For Level 1 and 2 autonomous driving systems, the perception response time is required. For driver perception-braking reaction time; This refers to the longitudinal friction coefficient of the road surface. The longitudinal slope of the road; For Level 3, 4, and 5 autonomous driving systems, the perception-braking response time is specified. For the time for driver takeover; After issuing a takeover request to the Level 3 autonomous driving system, the driver performs the preset deceleration before taking over driving operations.

[0020] Furthermore, in step S3, the line-of-sight failure function is constructed using the following formula:

[0021]

[0022] In the formula, This refers to the achievable line-of-sight distance for L1 or L2 level autonomous driving in the current occluded scenario. This represents the achievable line-of-sight distance for Level 3 autonomous driving in the current occluded scenario. This refers to the achievable line-of-sight distance for L4 or L5 level autonomous driving in the current occluded scenario. The required stopping sight distance for Level j driving automation;

[0023] Based on the line-of-sight failure function, the probability of line-of-sight failure in each sub-scene is calculated. Group evaluation working condition The formula for the view distance failure probability corresponding to an occluded sub-scene is:

[0024]

[0025] In the formula, For the first Group evaluation working condition In a sub-scenario of occlusion, the probability of line-of-sight failure for L1 or L2 autonomous vehicles. This refers to the line-of-sight failure function value for the corresponding L1 or L2 level autonomous driving vehicle. For the first Group evaluation working condition In a sub-scenario of occlusion, the probability of line-of-sight failure for a Level 3 autonomous vehicle. This represents the line-of-sight failure function value for the corresponding Level 3 autonomous vehicle. For the first Group evaluation working condition In a sub-scenario of occlusion, the probability of line-of-sight failure for L4 or L5 autonomous vehicles. This refers to the line-of-sight failure function value for the corresponding L4 or L5 autonomous driving vehicle.

[0026] Furthermore, in step S4, the three-dimensional line-of-sight risk sample consists of the line-of-sight failure probability of L1 or L2 level driving automation corresponding to the same occlusion sub-scene. Level 3 driving automation line-of-sight failure probability Probability of line-of-sight failure at Level 4 or Level 5 driving automation It consists of three components, in the form of:

[0027]

[0028] Where n is the evaluation condition number, c is the occlusion sub-scene number, and the three-dimensional vectors corresponding to all N evaluation conditions and C occlusion sub-scenes in each group constitute the line-of-sight risk sample set. .

[0029] Furthermore, in step S4, the clustering process of the adaptive multi-prototype competitive learning algorithm includes:

[0030] Select from the line-of-sight risk sample set D Using the line-of-sight risk samples formed by different evaluation conditions and occlusion sub-scenes as initial seed points, an initial seed point set is constructed. ;

[0031] For the t-th sample update process, the view distance risk samples selected from the view distance risk sample set D Calculate the Euclidean distance between it and the current K seed points to determine the winning seed point. , Let j be the j-th seed point in the initial seed point set; make the winning seed point Update along the current sample direction, for the remaining opponent seed points Calculate the opponent's penalty weight , and according to Update, in which Learning rate for winning seed point, The penalty coefficient for the opponent's seed point;

[0032] After each full sample traversal is completed, a judgment is made. Whether it is valid, The seed point convergence threshold. This is the first iteration before the start of this round. The position of each seed point; if not found, continue updating the sample; if found, count the winning count of each type of seed point. When there is When a low-frequency seed point is found, delete the low-frequency seed point and update K. The threshold for the number of wins at a seed point is used to calculate the number of wins when there are no low-frequency seed points. Subclusters corresponding to each seed point Inner Local density of individual line-of-sight risk samples and density interval , For indicator functions, Let Euclidean distance be the distance between two line-of-sight risk samples. The radius of the local density neighborhood. For the first The average pairwise distance between each viewpoint risk sample within a sub-cluster; according to Select the subclusters to be split and copy their seed points. Then, the sample update and convergence judgment are re-executed until the newly added seed point loses more than 100% of its winning count during the competitive learning process. The remaining seed points are then deleted; the number of these retained seed points is determined as the risk prototype, and its quantity corresponds to the adaptively determined number of line-of-sight risk categories. To obtain the risk prototype set .

[0033] Furthermore, in step S5, for the first... Individual sight distance risk category The line-of-sight failure probability for this category is calculated using the following formula:

[0034]

[0035] In the formula, For the first The number of sight risk samples included in each sight risk category. For the first Group evaluation working condition The probability of view distance failure corresponding to each occluded sub-scene. For the first Probability of line-of-sight failure for each line-of-sight risk category;

[0036] According to the order of sight failure probability from smallest to largest for each sight risk category, The line-of-sight risk categories were reordered:

[0037]

[0038] The ranked sight distance risk categories are then assigned sequentially as driving suitability risk levels 1 to 1. Driving suitability risk levels, and ensuring that the risk level of each driving suitability risk level meets the following requirements:

[0039]

[0040] In the formula, The lowest level of driving suitability risk. The highest level of driving suitability risk;

[0041] And establish a mapping relationship between risk prototypes and driving suitability risk levels:

[0042]

[0043] In the formula, After ranking the representative line-of-sight failure probabilities, the [number]th Risk prototypes corresponding to each line-of-sight risk category.

[0044] Furthermore, in step S5, for the c-th occlusion sub-scene under the evaluation condition, the line-of-sight failure probability of the c-th occlusion sub-scene under different autonomous driving levels is obtained according to steps S1 to S3, and a three-dimensional line-of-sight risk sample to be evaluated is constructed:

[0045]

[0046] , , These represent the line-of-sight failure probabilities for Level 1 or Level 2, Level 3, and Level 4 or Level 5 driving automation, respectively, corresponding to the c-th occlusion sub-scene.

[0047] calculate The Euclidean distance between each risk prototype and the nearest risk prototype is determined by the following formula:

[0048]

[0049] The driving safety risk level corresponding to the most recent risk prototype is determined as the driving safety risk level of the c-th occlusion sub-scenario to be evaluated. ;

[0050] Obtain the driving suitability risk level for all C key line-of-sight occlusion sub-scenarios under the evaluation conditions. Then, determine the overall driving suitability risk level using the following formula:

[0051] .

[0052] Furthermore, in step S5, the overall drivability risk level is determined based on the established mapping relationship between the risk prototype and the drivability risk level. The corresponding line-of-sight risk category is determined, and the line-of-sight failure probability of this risk category under the autonomous driving level group to be evaluated is obtained; this line-of-sight failure probability is then compared with a preset line-of-sight failure probability threshold.

[0053] When the line-of-sight failure probability is less than the preset line-of-sight failure probability threshold, the drivability of the tunnel interchange for autonomous vehicles under the current evaluation condition is determined to meet the safe driving state.

[0054] When the line-of-sight failure probability is equal to the preset line-of-sight failure probability threshold, the drivability of the tunnel interchange for autonomous vehicles under the current evaluation condition is determined to be in a critical safety state.

[0055] When the line-of-sight failure probability is greater than the preset line-of-sight failure probability threshold, it is determined that the drivability of the tunnel interchange for autonomous vehicles under the current evaluation condition does not meet the safe driving state.

[0056] The preset line-of-sight failure probability threshold is determined based on the autonomous driving level to be evaluated, the safety requirements of the autonomous driving system, road operating conditions, or verified test data.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] (1) This invention incorporates the alignment parameters, autonomous vehicle parameters and traffic flow parameters of tunnel interchanges into the evaluation conditions, and simultaneously sets static obstructions such as tunnel walls and ramp connection angles as well as dynamic obstructions formed by different types of obstacle vehicles, which can more completely reflect the limited sight distance characteristics of tunnel interchanges in actual operation.

[0059] (2) This invention establishes demand parking sight distance and sight distance failure function functions for L1 or L2, L3 and L4 or L5 driving automation respectively, calculates the sight distance failure probability through structural reliability theory, and extends the traditional deterministic sight distance satisfaction test to a probabilistic risk assessment that considers the randomness of parameters.

[0060] (3) The present invention adopts an adaptive multi-prototype competitive learning algorithm, which deletes or splits risk prototypes based on the number of wins of seed points, local density of subclusters and density intervals. It does not require pre-fixing the number of final risk categories, which can reduce the subjectivity caused by manually setting the number of clusters and the hierarchical threshold.

[0061] (4) This invention determines the order of risk categories based on the probability of line-of-sight failure, establishes the correspondence between risk prototypes and drivability risk levels, and uses the highest risk level in the obstruction sub-scenario as the overall evaluation result, which facilitates the identification of key scenarios for controlling the drivability of tunnel interchanges and provides a quantitative basis for the selection of linear schemes, sensor deployment and traffic safety management. Attached Figure Description

[0062] Figure 1 This is a flowchart illustrating the overall implementation of the tunnel interchange alignment drivability assessment method for autonomous vehicles provided in this embodiment of the invention.

[0063] Figure 2 This is a flowchart of constructing multiple types of occlusion scenarios and calculating the probability of view distance failure in an embodiment of the present invention;

[0064] Figure 3 This is a flowchart illustrating the use of an adaptive multi-prototype competitive learning algorithm to determine the line-of-sight risk category and risk prototype in an embodiment of the present invention.

[0065] Figure 4 This is a flowchart in an embodiment of the present invention for establishing a risk prototype and a driving suitability risk level mapping, and outputting the driving suitability evaluation result based on the highest driving suitability risk level under the working condition to be evaluated. Detailed Implementation

[0066] The technical solution of the present invention will now be described in detail with reference to the accompanying drawings.

[0067] To make the features and advantages of this patent application more apparent and understandable, specific examples are provided below for detailed explanation:

[0068] To make the technical solution, implementation process, and beneficial effects of this patent application clearer and more explicit, the present invention will be further described below in conjunction with embodiments. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of the present invention.

[0069] Unless otherwise stated, the technical and scientific terms used in this specification have the meanings commonly understood by one of ordinary skill in the art to which this invention pertains. The term "evaluation condition" as used in this specification refers to evaluation conditions jointly determined by a set of tunnel interchange alignment parameter distributions, autonomous vehicle parameter distributions, and traffic flow parameter distributions; the term "c-th occlusion sub-scene under the n-th evaluation condition" refers to the c-th independent line-of-sight occlusion instance formed in the n-th evaluation condition according to the obstacle vehicle type, obstacle vehicle position, and obstacle vehicle motion state, and is uniquely identified by the index (n, c).

[0070] like Figure 1 As shown, the present invention provides a method for evaluating the drivability of tunnel interchange alignments for autonomous vehicles, comprising the following steps:

[0071] Step S1: Collect the alignment parameters, autonomous vehicle parameters, and traffic flow parameters for the tunnel interchange scenario;

[0072] Step S2: Define multiple types of occlusion relationships based on the linear parameters. These relationships include static and dynamic occlusion scenarios. The static occlusion scenario consists of the tunnel wall structure, ramp connection angles, and vehicle positions as the main subjects of sight distance verification. The dynamic occlusion scenario consists of obstacle vehicles traveling on the left side of the main line, obstacle vehicles traveling on the right side of the main line, obstacle vehicles traveling in the merging and diverging zones and about to enter the main line or ramps, obstacle vehicle speeds, and the positions and speeds of vehicles as the main subjects of sight distance verification. Based on these multiple types of occlusion relationships, construct several occlusion sub-scenes in the virtual simulation platform and parameterize each sub-scene.

[0073] Step S3: Establish the required parking sight distance for different driving automation levels, construct the sight distance failure function using structural reliability theory, and calculate the sight distance failure probability of each occlusion sub-scenario under different driving automation levels.

[0074] Step S4: Construct three-dimensional line-of-sight risk samples based on the line-of-sight failure probabilities of each occluded sub-scene under different levels of driving automation. The line-of-sight risk sample set is composed of all three-dimensional line-of-sight risk samples. The line-of-sight risk sample set is clustered using an adaptive multi-prototype competitive learning algorithm. During the clustering process, a low-frequency seed point deletion operation is performed based on the number of wins of the seed point, and a sub-cluster splitting operation is performed based on the local density and density interval of the sub-cluster, so as to adaptively determine the number of line-of-sight risk categories and the risk prototypes corresponding to each line-of-sight risk category.

[0075] Step S5: Sort each sight risk category in ascending order of sight failure probability, and establish a mapping relationship between each risk prototype and the drivability risk level; for each occlusion sub-scenario under the evaluation condition, determine the drivability risk level of each occlusion sub-scenario based on the mapping relationship, and determine the highest drivability risk level among all occlusion sub-scenarios as the overall drivability risk level of the tunnel interchange under the evaluation condition, and output the drivability evaluation result.

[0076] The specific implementation of each step of this invention is as follows:

[0077] Step S1 includes:

[0078] Step S11: Collect the alignment information of the tunnel interchange scenario. The alignment information of the tunnel interchange scenario includes the number of tunnel lanes, the length of the straight lanes, the length of the entrance ramp lanes, the length of the exit ramp lanes, the radius of the ramp curve, the length of the acceleration lane, the length of the deceleration lane, the position of the merging and diverging noses, the design speed of the tunnel mainline, the design speed of the tunnel ramps, the design speed of the acceleration lane, and the design speed of the deceleration lane.

[0079] Step S12: Collect information about autonomous vehicles. The information about autonomous vehicles includes the level of automation, vehicle speed, perception-braking reaction time of the autonomous driving system, braking deceleration, preset braking deceleration of the autonomous driving system, driver takeover reaction time, driver takeover time after disengaging from the autonomous driving system, and configuration and deployment scheme of on-board perception sensors.

[0080] The vehicle-mounted perception sensor configuration and deployment scheme includes the type of perception sensor, detection distance, horizontal field of view, vertical field of view, horizontal angular resolution, vertical angular resolution, installation location, and static and dynamic obstacle perception algorithms.

[0081] Step S13: Collect traffic flow information, which includes traffic components, vehicles, and vehicle speeds.

[0082] Specifically, the alignment information for tunnel interchange scenarios can be obtained from design documents, as-built data, road information models, or on-site measurements; information for autonomous vehicles can be obtained from vehicle technical documents, autonomous driving system configuration files, and real-vehicle test data; and traffic flow information can be obtained through traffic surveys, video detection, radar detection, or historical operational data. For parameters with random fluctuations, their value range, empirical distribution, or probability distribution are determined based on the collected samples, providing input for subsequent evaluation condition generation and reliability calculation.

[0083] The configuration and deployment scheme of onboard perception sensors for autonomous vehicles is used to determine the effective perception range of the vehicle under different occlusion conditions. The effective perception range is jointly constrained by sensor detection distance, horizontal field of view, vertical field of view, angular resolution, installation position, and obstacle perception algorithm. When multiple sensors are configured, the detectable areas of each sensor in the current scene can be fused to obtain a comprehensive perception area that the vehicle can use for line-of-sight verification.

[0084] like Figure 2 As shown, the implementation methods for steps S2 and S3 are as follows:

[0085] Step S21: Define multiple types of occlusion scenarios for tunnel interchanges:

[0086] The various types of occlusion scenarios in tunnel interchanges include static occlusion scenarios and dynamic occlusion scenarios; the static occlusion scenarios include the tunnel wall structure, ramp connection angles, and the position of the autonomous vehicle as the subject of sight distance verification.

[0087] The dynamic scenarios include obstacle vehicles traveling on the left side of the main line, obstacle vehicles traveling on the right side of the main line, obstacle vehicles traveling in the merging and diverging zones that are about to enter the main line or the ramp, the speed of the obstacle vehicles, and the driving position and speed of the autonomous driving vehicle as the subject of the line-of-sight test.

[0088] The obstacle vehicles include large trucks and small vehicles.

[0089] Step S22: Define a set of parameter distribution combinations of tunnel interchange alignment parameters, automated vehicle parameters, and traffic flow parameters as a set of evaluation conditions, and number all evaluation conditions sequentially from the first set of evaluation conditions to the second set of evaluation conditions. Group evaluation of working conditions, To evaluate the total number of operating conditions;

[0090] In the Under the group evaluation condition, multiple occlusion sub-scenes are established based on the type of obstacle vehicle, its location, and its movement state. These multiple occlusion sub-scenes are sequentially numbered from the first occlusion sub-scene to the second occlusion sub-scene. A scene that is partially obscured. The total number of occlusion sub-scenes set for each evaluation condition.

[0091] Step S23: Build multiple types of occlusion sub-scenes in the virtual simulation platform, and set relevant parameters for the sub-scenes of various evaluation conditions based on the autonomous vehicle parameters and traffic flow parameters collected in step S1.

[0092] The static occlusion scenario is used to characterize the tunnel walls, merging and diverging noses, ramp sidewalls, and geometric occlusion caused by ramp connection angles, whose positions remain basically unchanged during the evaluation period. The dynamic occlusion scenario is used to characterize the occlusion caused by the changing position and motion state of the obstacle vehicle relative to the inspection subject over time. Large trucks and small vehicles are modeled using corresponding external dimensions, motion trajectories, and speed parameters, respectively.

[0093] Each evaluation condition can be randomly sampled or stratified according to the probability distribution of the parameters to obtain multiple simulation samples for reliability analysis. For the same... The evaluation conditions, while keeping the distribution of the alignment parameters, vehicle parameters, and traffic flow parameters constant, are varied by changing the type of obstacle vehicle, its lane, longitudinal relative position, lateral relative position, direction of movement, and speed, to form the first to the second evaluation conditions. A scene with obscured elements.

[0094] The virtual simulation platform establishes a unified road coordinate system and parametrically models the main road, entrance ramps, exit ramps, acceleration lanes, deceleration lanes, and merging / diverging noses. When the autonomous vehicle runs along the predetermined path, ray detection or visibility analysis is performed based on the vehicle reference point, sensor installation location, sensor field of view, and the contours of obstructions to obtain the visible sections along the vehicle's travel path at each moment.

[0095] Step S3: Construct a line-of-sight failure function using structural reliability theory and calculate the line-of-sight failure probability for each sub-scene.

[0096] Step S31: Establish the required parking sight distance for autonomous vehicles based on the level of driving automation. :

[0097]

[0098] In the formula, The required stopping sight distance for Level j driving automation; The vehicle's speed; For Level 1 and 2 autonomous driving systems, the perception response time is required. For driver perception-braking reaction time; This refers to the longitudinal friction coefficient of the road surface. The longitudinal slope of the road; For Level 3, 4, and 5 autonomous driving systems, the perception-braking response time is specified. For the time for driver takeover; After issuing a takeover request to the Level 3 automated driving system, the driver performs a preset deceleration before taking over driving operations, typically 2.5 m / s². 2 ;

[0099] In the formula, the constant 3.6 is for unit conversion. , The unit is km / h, through Convert the units to m / s; the constant 9.81 is an engineering approximation of the standard gravitational acceleration, with units of m / s. In engineering calculations, the standard gravitational acceleration is typically taken as 9.81; the constant 19.62 is... Its unit is It is used to calculate braking distance when vehicle speed is expressed in m / s; the constant 254 is a combined conversion factor formed by gravitational acceleration and speed unit conversion. Applicable to When the unit is km / h, its value varies with the unit system used and it is not a dimensionless physical constant.

[0100] Step S32: Construct a line-of-sight failure function based on structural reliability theory. The formula is:

[0101]

[0102] In the formula, The achievable line-of-sight distance for L1 or L2 level autonomous driving in the current occluded scenario. This refers to the achievable line-of-sight distance for L3-level autonomous driving in the current occluded scenario. The achievable line-of-sight distance for L4 or L5 level autonomous driving in the current occluded scenario.

[0103] Step S33: Calculate the line-of-sight failure probability for each sub-scene based on the aforementioned line-of-sight failure function. Group evaluation working condition The formula for the view distance failure probability corresponding to an occluded sub-scene is:

[0104]

[0105] In the formula, For the first Group evaluation working condition In a sub-scenario of occlusion, the probability of line-of-sight failure for L1 or L2 autonomous vehicles. For the first Group evaluation working condition In a sub-scenario of occlusion, the probability of line-of-sight failure for a Level 3 autonomous vehicle. For the first Group evaluation working condition The probability of line-of-sight failure for L4 or L5 autonomous vehicles in a single occlusion sub-scenario.

[0106] In reliability analysis, a sight-distance failure event is defined as a function function where the sight-distance failure function is less than 0. Each random parameter is sampled according to the distribution determined in step S1, and the obtainable sight distance and required parking sight distance are repeatedly calculated in a virtual simulation scenario. The sight-distance failure probability can be obtained using Monte Carlo simulation, importance sampling, or a first-order reliability method. In this embodiment, when using Monte Carlo simulation, the ratio of the number of sight-distance failure samples to the total number of samples is used as the estimated value of the corresponding sight-distance failure probability.

[0107] Step S4: Construct a line-of-sight risk sample set, and use an adaptive multi-prototype competitive learning algorithm to cluster the line-of-sight risk sample set, adaptively determining the number of line-of-sight risk categories and the risk prototypes corresponding to each line-of-sight risk category.

[0108] like Figure 3 As shown, the implementation method of step S4 is as follows:

[0109] Step S41: Obtain the first corresponding to the different occlusion scenarios according to step S3. Group evaluation working condition The probabilities of line-of-sight failure for L1 or L2, L3, and L4 or L5 autonomous vehicles in each occluded sub-scene are calculated, and the probability of line-of-sight failure for each vehicle is constructed. Group evaluation working condition View distance risk samples corresponding to each occlusion sub-scene:

[0110]

[0111] Since the three line-of-sight failure probabilities are all within the same dimension range of 0 to 1, Euclidean distance can be directly used to measure the similarity between line-of-sight risk samples. The three components of each sample reflect the line-of-sight risk under low-level driving automation, conditional driving automation, and high-level driving automation, respectively, so that the clustering results can simultaneously preserve the risk differences of different driving automation levels for the same occluded sub-scene.

[0112] Step S42: From the line-of-sight risk sample set described in step S41 Selected from Using the line-of-sight risk samples formed by different evaluation conditions and occlusion sub-scenes as initial seed points, an initial seed point set is constructed. :

[0113]

[0114] In the formula, The initial number of seed points, which are used to represent the line-of-sight risk prototype in the line-of-sight risk sample set;

[0115] Set the learning rate for the winning seed point. Opponent's seed point penalty coefficient Seed point win count threshold Seed point convergence threshold Local density neighborhood radius and the maximum number of iterations .

[0116] Initial seed points are preferentially selected from different evaluation conditions and different occlusion sub-scenes to increase the coverage of the initial risk prototypes. The maximum-minimum distance principle can be further adopted to ensure that subsequently selected seed points maintain a large minimum distance from existing seed points. The learning rate of winning seed points controls the extent to which risk prototypes move towards the current sample, the penalty coefficient of opposing seed points controls the extent to which non-winning prototypes move away from the current sample, and the seed point winning count threshold is used to delete prototypes lacking sample support.

[0117] Step S43, in the During the next sample update, from the line-of-sight risk sample set Select current line-of-sight risk samples Calculate the Euclidean distance between the current line-of-sight risk sample and various sub-points, and determine the winning seed point closest to the current line-of-sight risk sample:

[0118]

[0119] Update the winning seed point along the current line-of-sight risk sample direction. For opponent seed points other than the winning seed point, calculate the... Opponent penalty weight for each opponent seed point:

[0120]

[0121] Update the opponent seed point along a direction away from the current line-of-sight risk sample:

[0122]

[0123] In the formula, This represents the current number of seed points. For the winning seed, For the first Each opponent's seed point, For the first The opponent penalty weight for each opponent seed point;

[0124] One iteration is completed after traversing all the sight distance risk samples in the sight distance risk sample set once.

[0125] The winning seed point moves along the current line-of-sight risk sample direction, gradually approaching the center of samples with similar risk characteristics; the opposing seed point is penalized according to distance relationships to maintain differentiation between different risk prototypes. After completing one traversal of all line-of-sight risk samples, the convergence of this round of competitive learning is determined based on the changes in seed point positions.

[0126] Step S44: After completing one iteration, determine whether the seed point set has converged according to the following formula:

[0127]

[0128] In the formula, This is the first iteration before the start of this round. The location of each seed point;

[0129] If the set of seed points does not meet the convergence condition, continue to execute step S43;

[0130] When the seed point set satisfies the convergence condition, the number of wins for each type of seed point in the current iteration round is counted. When there exists a low-frequency seed point that satisfies the following formula:

[0131]

[0132] Delete the low-frequency seed points and update the seed point count. The seed point set obtained after deleting low-frequency seed points is determined as the final risk prototype set;

[0133] When there are no low-frequency seed points with a winning count less than the seed point winning count threshold, calculate the local density and density interval of the sub-clusters corresponding to each type of seed point;

[0134] For the Subclusters corresponding to each seed point Calculate the first sub-cluster Local density of each line-of-sight risk sample:

[0135]

[0136] In the formula, This is an indicator function; it takes the value 1 when the condition inside the parentheses is true, and 0 otherwise. The Euclidean distance between two line-of-sight risk samples;

[0137] Calculate the first Density interval of subclusters:

[0138]

[0139] In the formula, For the first The average pairwise distance between each line-of-sight risk sample within a sub-cluster; for the line-of-sight risk sample with the highest local density, the corresponding distance is the maximum distance between that line-of-sight risk sample and other line-of-sight risk samples within its sub-cluster.

[0140] The sub-clusters to be split are determined according to the following formula:

[0141]

[0142] Copy the seed point corresponding to the sub-cluster to be split to form a new seed point, and update the number of seed points:

[0143]

[0144] After adding a seed point, repeat steps S43 and S44 until the newly added seed point is deleted during the competitive learning process because its winning count is less than the seed point winning count threshold.

[0145] Step S45: Determine the number of seed points retained in step S44 as the number of adaptive line-of-sight risk categories. The final retained seed points will be determined as the risk prototype set:

[0146]

[0147] Based on the Euclidean distance between each line-of-sight risk sample and each risk prototype, each line-of-sight risk sample is assigned to the line-of-sight risk category corresponding to the nearest risk prototype, thus forming... Individual line-of-sight risk categories:

[0148]

[0149] Local density characterizes the degree of clustering of samples within a subcluster within its neighborhood radius, while density interval characterizes the degree of separation between high-density samples and even higher-density samples. Subclusters to be split are selected using a joint criterion of sample count and density interval, prioritizing subclusters containing more samples and potentially multiple risk modalities to generate new seed points. If a new seed point consistently receives sufficient sample support during the re-competitive learning process, it is retained and forms a new risk category; if its winning count falls below a threshold and it is deleted, it indicates that the current sample set no longer supports the addition of stable risk categories, the algorithm terminates, and the final risk prototype set is output.

[0150] The algorithm can be implemented using software with vector operations and clustering capabilities, such as NumPy and SciPy in the Python programming language, along with related data processing libraries, to perform distance calculations, seed point updates, local density statistics, and sample class division. The parameters are set based on the number of samples, the Euclidean distance scale between samples, and convergence experiments.

[0151] Step S5: Determine the risk order of each sight distance risk category, establish the correspondence between risk prototypes and drivability risk levels, and output the drivability evaluation results of the tunnel interchange under the evaluation condition based on the highest drivability risk level under the evaluation condition.

[0152] like Figure 4 As shown, the implementation method of step S5 is as follows:

[0153] Step S51: Based on the line-of-sight risk clustering results obtained in step S4, perform the following steps: Individual sight distance risk category The probability of line-of-sight failure corresponding to each line-of-sight risk sample is statistically analyzed, and the probability of failure is calculated according to the following formula. Probability of line-of-sight failure for each line-of-sight risk category:

[0154]

[0155] In the formula, For the first The number of sight risk samples included in each sight risk category. The first one obtained in step S3 Group evaluation working condition The probability of view distance failure corresponding to each occluded sub-scene. For the first Probability of line-of-sight failure for each line-of-sight risk category;

[0156] Step S52: According to the order of sight failure probability from smallest to largest for each sight risk category, process the... The line-of-sight risk categories were reordered:

[0157]

[0158] The ranked sight distance risk categories are then assigned sequentially as driving suitability risk levels 1 to 1. Driving suitability risk levels, and ensuring that the risk level of each driving suitability risk level meets the following requirements:

[0159]

[0160] In the formula, The lowest level of driving suitability risk. The highest level of driving suitability risk;

[0161] Step S53: Based on the ranking results of sight distance risk categories obtained in step S52, establish a mapping relationship between the risk prototype corresponding to each sight distance risk category and the corresponding driving suitability risk level:

[0162]

[0163] In the formula, After ranking the representative line-of-sight failure probabilities, the [number]th Risk prototypes corresponding to each line-of-sight risk category;

[0164] After sorting by category, risk categories with higher probabilities of sight failure correspond to higher drivability risk levels. The number of drivability risk levels is determined by the final number of risk categories adaptively obtained in step S4. Therefore, it is not necessary to pre-define a fixed number of secondary, tertiary, or quinary risk levels. Once a one-to-one mapping is established between the risk prototype and the risk level, it can serve as a classification benchmark for subsequent samples to be evaluated.

[0165] Step S54: For the tunnel interchange obstruction sub-scenario to be evaluated, obtain the first sub-scenario under the evaluation condition according to steps S1 to S3. The probability of line-of-sight failure in each occluded sub-scene under different levels of autonomous driving is calculated, and a sample of line-of-sight risks to be evaluated is constructed:

[0166]

[0167] Calculate the first The Euclidean distance between each line-of-sight risk sample to be evaluated and each risk prototype is determined according to the following formula: The nearest risk prototype to the line-of-sight risk sample to be evaluated:

[0168]

[0169] The drivability risk level corresponding to the most recent risk prototype described in S53 is determined to be the [number]. The driving suitability risk level of each un-evaluated occlusion sub-scenario;

[0170] The line-of-sight risk samples to be evaluated use the same parameter definitions, probability calculation methods, and component arrangement order as the line-of-sight risk sample set. By calculating the Euclidean distance between the sample to be evaluated and each risk prototype, it is assigned to the line-of-sight risk category corresponding to the nearest risk prototype, and the driving suitability risk level of the occluded sub-scene is directly obtained.

[0171] Step S55: Based on step S54, obtain the driving suitability risk level for each key occlusion sub-scenario at each line of sight under the evaluation condition:

[0172]

[0173] The highest level of drivability risk is determined as the overall drivability risk level of the tunnel interchange to be evaluated:

[0174]

[0175] In the formula, The total number of key occlusion sub-scenes under the evaluation condition. For the first The corresponding driving suitability risk level for each key obstruction of sight distance sub-scenario. The overall drivability risk level of the tunnel interchange to be evaluated;

[0176] Among them, the key occlusion sub-scene refers to the occlusion sub-scene established in step S2 that creates a substantial line-of-sight limitation for the working condition to be evaluated. The highest risk level is adopted as the overall drivability risk level, reflecting the conservative principle in safety evaluation and avoiding the average dilution of high-risk key scenes by low-risk scenes; at the same time, the risk level of each sub-scene is still retained to identify the specific occlusion type and location that causes the overall high risk.

[0177] Step S56: Based on the risk prototype established in Step S53, the mapping relationship between the sight distance risk category and the drivability risk level is determined, and the sight distance risk category corresponding to the overall drivability risk level obtained in Step S55 is determined. The sight distance failure probability of the sight distance risk category calculated in Step S51 under the autonomous driving level group to be evaluated is obtained. The sight distance failure probability is compared with a preset sight distance failure probability threshold. When the sight distance failure probability is less than the preset sight distance failure probability threshold, it indicates that the drivability of the tunnel interchange for autonomous vehicles under the current evaluation condition meets the safe driving requirements. When the sight distance failure probability is equal to the preset sight distance failure probability threshold, it indicates that the drivability of the tunnel interchange for autonomous vehicles under the current evaluation condition is in a critically safe state. When the sight distance failure probability is greater than the preset sight distance failure probability threshold, it indicates that the drivability of the tunnel interchange for autonomous vehicles under the current evaluation condition does not meet the safe driving requirements.

[0178] In summary, this invention collects information on the alignment of tunnel interchanges, autonomous vehicles, and traffic flow to establish a parameterized sight distance testing scenario that includes both static and dynamic occlusion. It constructs required stopping sight distance and sight distance failure function functions for different levels of driving automation and uses structural reliability theory to form three-dimensional sight distance risk samples. Based on this, an adaptive multi-prototype competitive learning algorithm automatically deletes risk prototypes lacking sample support and splits risk subclusters with potential multimodal structures to obtain the number of risk categories and risk prototypes adapted to the sample distribution. Finally, it determines the risk order of each sight distance risk category, establishes the correspondence between risk prototypes and drivability risk levels, and outputs the drivability evaluation results of the tunnel interchange under the evaluated condition based on the highest drivability risk level.

[0179] Those skilled in the art will understand that the embodiments of this application can be implemented by computer programs, dedicated evaluation software, virtual simulation systems, or a combination of hardware and software. The program instructions for performing the above steps can be stored in a computer-readable storage medium and executed by a processor to complete evaluation condition generation, virtual scene parameterization, line-of-sight calculation, reliability analysis, risk clustering, and risk level output.

[0180] This application is described with reference to method flowcharts and algorithm flowcharts. Each step in the flowchart can be implemented by computer program instructions, and the execution platform, data interface, or calculation method can be changed according to the needs of actual engineering applications. Equivalent substitutions made to the data acquisition method, simulation platform, reliability calculation method, parameter setting method, and software implementation method without changing the technical logical relationships between the steps are all feasible forms of this invention.

[0181] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any simple modifications, equivalent substitutions, or improvements made by those skilled in the art based on the technical content disclosed in the present invention, as long as they do not depart from the essence of the technical solution of the present invention, should be included within the scope of protection of the present invention.

[0182] The above are preferred embodiments of the present invention. Any changes made to the technical solution of the present invention that do not exceed the scope of the technical solution of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for evaluating the drivability of tunnel interchange alignments for autonomous vehicles, characterized in that, include: Step S1: Collect the alignment parameters, autonomous vehicle parameters, and traffic flow parameters for the tunnel interchange scenario; Step S2: Define multiple types of occlusion relationships based on the linear parameters. These relationships include static and dynamic occlusion scenarios. The static occlusion scenario consists of the tunnel wall structure, ramp connection angles, and vehicle positions as the main subjects of sight distance verification. The dynamic occlusion scenario consists of obstacle vehicles traveling on the left side of the main line, obstacle vehicles traveling on the right side of the main line, obstacle vehicles traveling in the merging and diverging zones and about to enter the main line or ramps, obstacle vehicle speeds, and the positions and speeds of vehicles as the main subjects of sight distance verification. Based on these multiple types of occlusion relationships, construct several occlusion sub-scenes in the virtual simulation platform and parameterize each sub-scene. Step S3: Establish the required parking sight distance for different driving automation levels, construct the sight distance failure function using structural reliability theory, and calculate the sight distance failure probability of each occlusion sub-scenario under different driving automation levels. Step S4: Construct a three-dimensional line-of-sight risk sample based on the line-of-sight failure probability of each occluded sub-scene under different levels of driving automation, and construct a line-of-sight risk sample set from all three-dimensional line-of-sight risk samples. An adaptive multi-prototype competitive learning algorithm is used to cluster the line-of-sight risk sample set. During the clustering process, low-frequency seed point deletion is performed based on the number of times the seed point wins, and sub-cluster splitting is performed based on the local density and density interval of the sub-cluster, so as to adaptively determine the number of line-of-sight risk categories and the risk prototypes corresponding to each line-of-sight risk category. Step S5: Sort each sight risk category in ascending order of sight failure probability, and establish a mapping relationship between each risk prototype and the drivability risk level; for each occlusion sub-scenario under the evaluation condition, determine the drivability risk level of each occlusion sub-scenario based on the mapping relationship, and determine the highest drivability risk level among all occlusion sub-scenarios as the overall drivability risk level of the tunnel interchange under the evaluation condition, and output the drivability evaluation result.

2. The method for evaluating the drivability of tunnel interchange alignments for autonomous vehicles according to claim 1, characterized in that, In step S1, The alignment parameters include the number of tunnel lanes, the length of the straight lanes, the length of the entrance ramp lanes, the length of the exit ramp lanes, the radius of the ramp curve, the length of the acceleration lane, the length of the deceleration lane, the position of the merging and diverging noses, the design speed of the tunnel mainline, the design speed of the tunnel ramps, the design speed of the acceleration lane, and the design speed of the deceleration lane. The parameters of the autonomous vehicle include the level of automation, vehicle speed, perception-braking reaction time of the autonomous driving system, braking deceleration, preset braking deceleration, driver takeover reaction time, driver takeover time, and the configuration and deployment scheme of the on-board perception sensors. The configuration and deployment scheme of the on-board perception sensors includes the type of perception sensor, detection distance, horizontal field of view, vertical field of view, horizontal angular resolution, vertical angular resolution, installation position, and obstacle perception algorithm. The traffic flow parameters include the traffic components, vehicles, and vehicle speeds.

3. The method for evaluating the drivability of tunnel interchange alignments for autonomous vehicles according to claim 1, characterized in that, In step S2, the obstacle vehicles include large trucks and small vehicles; a combination of a set of linear parameters, autonomous vehicle parameters, and traffic flow parameters is defined as a set of evaluation conditions. Under the nth set of evaluation conditions, multiple occlusion sub-scenes are established according to the obstacle vehicle type, the obstacle vehicle location, and the obstacle vehicle movement state, and are numbered sequentially; the parameterization setting includes assigning values ​​to the vehicle position, driving speed, and sensor detection range in each occlusion sub-scene according to the autonomous vehicle parameters and traffic flow parameters collected in step S1.

4. The method for evaluating the drivability of tunnel interchange alignments for autonomous vehicles according to claim 1, characterized in that, In step S3, the required parking sight distance is established for different levels of driving automation. The specific calculation formula is as follows: In the formula, The required stopping sight distance for Level j driving automation; The vehicle's speed; For Level 1 and 2 autonomous driving systems, the perception response time is required. For driver perception-braking reaction time; This refers to the longitudinal friction coefficient of the road surface. The longitudinal slope of the road; For Level 3, 4, and 5 autonomous driving systems, the perception-braking response time is specified. For the time for driver takeover; After issuing a takeover request to the Level 3 autonomous driving system, the driver performs the preset deceleration before taking over driving operations.

5. The method for evaluating the drivability of tunnel interchange alignments for autonomous vehicles according to claim 1, characterized in that, In step S3, the line-of-sight failure function is constructed using the following formula: In the formula, This refers to the achievable line-of-sight distance for L1 or L2 level autonomous driving in the current occluded scenario. This represents the achievable line-of-sight distance for Level 3 autonomous driving in the current occluded scenario. This refers to the achievable line-of-sight distance for L4 or L5 level autonomous driving in the current occluded scenario. The required stopping sight distance for Level j driving automation; Based on the line-of-sight failure function, the probability of line-of-sight failure in each sub-scene is calculated. Group evaluation working condition The formula for the view distance failure probability corresponding to an occluded sub-scene is: In the formula, For the first Group evaluation working condition In a sub-scenario of occlusion, the probability of line-of-sight failure for L1 or L2 autonomous vehicles. This refers to the line-of-sight failure function value for the corresponding L1 or L2 level autonomous driving vehicle. For the first Group evaluation working condition In a sub-scenario of occlusion, the probability of line-of-sight failure for a Level 3 autonomous vehicle. This represents the line-of-sight failure function value for the corresponding Level 3 autonomous vehicle. For the first Group evaluation working condition In a sub-scenario of occlusion, the probability of line-of-sight failure for L4 or L5 autonomous vehicles. This refers to the line-of-sight failure function value for the corresponding L4 or L5 autonomous driving vehicle.

6. The method for evaluating the drivability of tunnel interchange alignments for autonomous vehicles according to claim 1, characterized in that, In step S4, the three-dimensional line-of-sight risk sample consists of the line-of-sight failure probability of L1 or L2 level driving automation corresponding to the same occlusion sub-scene. Level 3 driving automation line-of-sight failure probability Probability of line-of-sight failure at Level 4 or Level 5 driving automation It consists of three components, in the form of: Where n is the evaluation condition number, c is the occlusion sub-scene number, and the three-dimensional vectors corresponding to all N evaluation conditions and C occlusion sub-scenes in each group constitute the line-of-sight risk sample set. .

7. The method for evaluating the drivability of tunnel interchange alignments for autonomous vehicles according to claim 6, characterized in that, In step S4, the clustering process of the adaptive multi-prototype competitive learning algorithm includes: Select from the line-of-sight risk sample set D Using the line-of-sight risk samples formed by different evaluation conditions and occlusion sub-scenes as initial seed points, an initial seed point set is constructed. ; For the t-th sample update process, the view distance risk samples selected from the view distance risk sample set D Calculate the Euclidean distance between it and the current K seed points to determine the winning seed point. , Let j be the j-th seed point in the initial seed point set; make the winning seed point Update along the current sample direction, for the remaining opponent seed points Calculate the opponent's penalty weight , and according to Update, in which Learning rate for winning seed point, The penalty coefficient for the opponent's seed point; After each full sample traversal is completed, a judgment is made. Whether it is valid, The seed point convergence threshold. This is the first iteration before the start of this round. The position of each seed point; if not found, continue updating the sample; if found, count the winning count of each type of seed point. When there is When a low-frequency seed point is found, delete the low-frequency seed point and update K. The threshold for the number of wins at a seed point is used to calculate the number of wins when there are no low-frequency seed points. Subclusters corresponding to each seed point Inner Local density of individual line-of-sight risk samples and density interval , For indicator functions, Let Euclidean distance be the distance between two line-of-sight risk samples. The radius of the local density neighborhood. For the first The average pairwise distance between each viewpoint risk sample within a sub-cluster; according to Select the subclusters to be split and copy their seed points. Then, the sample update and convergence judgment are re-executed until the newly added seed point loses more than 100% of its winning count during the competitive learning process. The remaining seed points are then deleted; the number of these retained seed points is determined as the risk prototype, and its quantity corresponds to the adaptively determined number of line-of-sight risk categories. To obtain the risk prototype set .

8. The method for evaluating the drivability of tunnel interchange alignments for autonomous vehicles according to claim 7, characterized in that, In step S5, for the first Individual sight distance risk category The line-of-sight failure probability for this category is calculated using the following formula: In the formula, For the first The number of sight risk samples included in each sight risk category. For the first Group evaluation working condition The probability of view distance failure corresponding to each occluded sub-scene. For the first Probability of line-of-sight failure for each line-of-sight risk category; According to the order of sight failure probability from smallest to largest for each sight risk category, The line-of-sight risk categories were reordered: The ranked sight distance risk categories are then assigned sequentially as driving suitability risk levels 1 to 1. Driving suitability risk levels, and ensuring that the risk level of each driving suitability risk level meets the following requirements: In the formula, The lowest level of driving suitability risk. The highest level of driving suitability risk; And establish a mapping relationship between risk prototypes and driving suitability risk levels: In the formula, After ranking the representative line-of-sight failure probabilities, the [number]th Risk prototypes corresponding to each line-of-sight risk category.

9. The method for evaluating the drivability of tunnel interchange alignments for autonomous vehicles according to claim 8, characterized in that, In step S5, for the c-th occlusion sub-scene under the evaluation condition, the line-of-sight failure probability of the c-th occlusion sub-scene under different autonomous driving levels is obtained according to steps S1 to S3, and a three-dimensional line-of-sight risk sample to be evaluated is constructed: , , These represent the line-of-sight failure probabilities for Level 1 or Level 2, Level 3, and Level 4 or Level 5 driving automation, respectively, corresponding to the c-th occlusion sub-scene. calculate The Euclidean distance between each risk prototype and the nearest risk prototype is determined by the following formula: The driving safety risk level corresponding to the most recent risk prototype is determined as the driving safety risk level of the c-th occlusion sub-scenario to be evaluated. ; Obtain the driving suitability risk level for all C key line-of-sight occlusion sub-scenarios under the evaluation conditions. Then, determine the overall driving suitability risk level using the following formula: 。 10. The method for evaluating the drivability of tunnel interchange alignments for autonomous vehicles according to claim 9, characterized in that, In step S5, the overall drivability risk level is determined based on the established mapping relationship between the risk prototype and the drivability risk level. The corresponding line-of-sight risk category is determined, and the line-of-sight failure probability of this risk category under the autonomous driving level group to be evaluated is obtained; this line-of-sight failure probability is then compared with a preset line-of-sight failure probability threshold. When the line-of-sight failure probability is less than the preset line-of-sight failure probability threshold, the drivability of the tunnel interchange for autonomous vehicles under the current evaluation condition is determined to meet the safe driving state. When the line-of-sight failure probability is equal to the preset line-of-sight failure probability threshold, the drivability of the tunnel interchange for autonomous vehicles under the current evaluation condition is determined to be in a critical safety state. When the line-of-sight failure probability is greater than the preset line-of-sight failure probability threshold, it is determined that the drivability of the tunnel interchange for autonomous vehicles under the current evaluation condition does not meet the safe driving state. The preset line-of-sight failure probability threshold is determined based on the autonomous driving level to be evaluated, the safety requirements of the autonomous driving system, road operating conditions, or verified test data.

Images