A method for generating a high-value simulation test scene of an intelligent connected vehicle

Abstract

The application provides a method for generating high-value simulation test scenes of intelligent networked vehicles. The method includes constructing simulation test scene elements including road environment, traffic participants, meteorological lighting, traffic control and communication, and driving behavior conditions, forming a standardized scene element framework covering static road base, dynamic traffic participation, natural environment conditions, networked communication control, and vehicle behavior decision-making; constructing a simulation test scene feature system; based on the standardized scene element framework, taking the feature elements in the simulation test scene feature system as decision variables, and generating high-value simulation test scenes based on an improved genetic algorithm; and applying the high-value simulation test scenes to intelligent networked vehicle simulation closed-loop testing. The application realizes accurate and efficient generation of high-value simulation test scenes, continuously optimizes scene quality through feedback loops, improves the relevance and efficiency of intelligent networked vehicle simulation testing, and reduces research and testing costs.

Description

Technical Field

[0001] This invention relates to the field of intelligent connected vehicle simulation testing technology, and specifically to a method for generating high-value simulation testing scenarios for intelligent connected vehicles. Background Technology

[0002] With the rapid development of intelligent connected vehicle technology, the safety and reliability verification of its autonomous driving functions has become a core industry requirement. Simulation testing, as a crucial link in the R&D process of intelligent connected vehicles, offers advantages such as high testing efficiency, low cost, reproducible scenarios, and no safety risks, and has become a core method to replace some real-vehicle testing. Currently, the generation of simulation test scenarios for intelligent connected vehicles mostly adopts random generation, manual design, or simple scenario reuse methods, which have the following shortcomings: First, the construction of scene elements is incomplete, and the interaction between multiple dimensions such as vehicles, roads, and environment is not fully considered, resulting in the scene being out of touch with real driving conditions; Second, the lack of systematic analysis of scenario characteristics makes it impossible to accurately identify high-value test scenarios (i.e., scenarios that are of key significance for the performance verification of autonomous driving systems and can expose potential defects). A large number of redundant scenarios occupy test resources and reduce test efficiency. Third, the scene generation algorithm has limitations. Traditional heuristic algorithms are prone to problems such as premature convergence and local optima, making it difficult to efficiently generate high-value scenes that meet testing requirements. Fourth, simulation testing and scene optimization are disconnected, lacking an effective feedback loop mechanism. Test results cannot be fed back to the scene generation process in a timely manner, making it difficult to continuously improve scene quality.

[0003] Therefore, there is an urgent need for a generation method that can comprehensively construct scene elements, systematically sort out scene characteristics, efficiently generate high-value scenes, and achieve closed-loop optimization of test feedback. This method would solve the technical problems of poor scene targeting, low test efficiency, and difficulty in iteratively improving scene quality in existing technologies, and support the efficient and comprehensive verification of autonomous driving systems for intelligent connected vehicles. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a method for generating high-value simulation test scenarios for intelligent connected vehicles. This method aims to achieve accurate and efficient generation of high-value simulation test scenarios, while continuously optimizing scenario quality through a feedback loop. This enhances the relevance and efficiency of intelligent connected vehicle simulation testing and reduces R&D testing costs.

[0005] The technical solution adopted in this invention is a method for generating high-value simulation test scenarios for intelligent connected vehicles; This includes constructing simulation test scenario elements for road environment, traffic participants, weather and lighting, traffic control and communication, and driving behavior conditions, forming a standardized scenario element framework covering all dimensions of static road base, dynamic traffic participation, natural environmental conditions, network communication control, and vehicle behavior decision-making. Construct a simulation test scenario feature system, including three types of simulation scenario features: scenario simulation accuracy, scenario real-time performance, and scenario reproducibility; Based on a standardized scenario element framework, the feature elements in the simulation test scenario feature system are used as decision variables, and a high-value simulation test scenario is generated based on an improved genetic algorithm. High-value simulation test scenarios are applied to closed-loop simulation testing of intelligent connected vehicles, and real-time data on scenario operation status, test indicator achievement, and actual feedback data on system defect exposure are collected; the high-value simulation test scenarios are then optimized.

[0006] Furthermore, the road environment elements include road type, number of lanes, road longitudinal slope, road surface adhesion coefficient, lane width, and curve radius of curvature data; The traffic participant elements include data on the vehicle's initial speed, vehicle's initial position, number of background vehicles, background vehicle speed, background vehicle initial position, number of pedestrians, pedestrian crossing speed, and pedestrian initial position. The meteorological lighting elements mentioned include weather type, visibility, light intensity, rainfall intensity, snowfall intensity, wind speed, wind direction, and road surface water thickness data; Traffic control and communication elements include traffic sign type, traffic light timing, communication delay, communication packet loss rate, and communication distance data; Driving behavior and operating condition elements include the vehicle's driving intention, longitudinal acceleration, lateral acceleration, lane change lateral deviation rate, following safety distance, and following time distance data.

[0007] Furthermore, the construction of the simulation test scenario feature system includes scenario complexity, scenario coverage, scenario danger, scenario realism, scenario non-redundancy, scenario real-time performance, and scenario reproducibility. The aforementioned scene complexity includes decomposing scene complexity into multiple secondary evaluation indicators, for ; For each secondary evaluation indicator, experts scored it from 1 to 5 points according to the scenario, resulting in the original score vector for each scenario. Then perform subjective weight calculation: Reconstruct the judgment matrix by having experts score the pairwise importance of each indicator; then calculate the largest eigenvalue of the judgment matrix. The subjective weights are obtained by normalizing the corresponding feature vectors. ; Perform a consistency check; ,like If so, then the weights are valid; Next, the objective weights are calculated, including the standardization of the evaluation index scores for all scenarios. Let the total number of scenarios be m, and the total number of secondary evaluation indicators for scenario complexity C be n, to obtain the standardization matrix. ;

[0008] Calculate the information entropy of each secondary evaluation indicator.

[0009] in , ; Calculate objective weights ; Then calculate the combined weights and the final score. The expression for calculating the combined weights is as follows: .

[0010] Furthermore, for a single scenario, a weighted score is calculated, expressed as follows:

[0011] For all scenarios Perform linear normalization and map to The range is used to obtain the value of the complexity C for each scenario, as shown in the following expression:

[0012] The scenario coverage rate R is the degree to which the simulation scenario covers the functional requirements of intelligent connected vehicles. Based on the functional requirements document, the matching rate between the scenario and the requirement points is calculated, with a value range of [0, 100].

[0013] Furthermore, the matching rate between the calculation scenario and the requirement points is calculated as follows: First, a list of functional requirements is compiled, listing all functional requirements of the system under test. Each function is then further subdivided into sub-requirements to obtain a set of requirements. ,common One demand point; Scenario and requirement matching determination: For each simulation scenario, determine whether it can cover each requirement point to obtain a matching vector;

[0014] in This indicates that the scenario can cover the requirements. ; Then calculate the coverage rate and map the coverage rate values ​​to... The coverage value for each scene is obtained from the interval, as shown in the following expression:

[0015] The scenario hazard is defined as the probability of a smart connected vehicle facing collision or loss of control safety risks in a simulated scenario, quantified by combining virtual TTC and THW; the expression is as follows:

[0016] The value range is [0, 100].

[0017] Furthermore, by combining virtual TTC and THW, the probability of collision and loss of control safety risks faced by intelligent connected vehicles is quantified. The specific steps are as follows: First, data collection is performed: During the simulation, the minimum values ​​of the time headway (TTC) and distance between the target vehicle and the intelligent vehicle (THW) are collected in real time and denoted as follows: and ; Where TTC is the distance between the two vehicles / relative speed, the smaller the value, the higher the risk; THW is the distance between the two vehicles / vehicle speed, the higher the value, the higher the risk. The expression is as follows:

[0018] The higher the D value, the higher the risk; Next, apply range constraints. If the calculated result exceeds the [0, 100] interval, truncation is performed, as shown in the following expression:

[0019] The scene realism refers to the similarity between the simulated scene and real road data. The feature vector similarity is calculated using a dynamic time warping algorithm, with values ​​ranging from [0, 100]. The steps are as follows: First, feature vectors are constructed, then DTW distance is calculated. Given the distance matrix, we use dynamic programming to find the shortest regular path and obtain the DTW distance. The smaller the distance, the more similar the trajectories; Then, the DTW distance for all scenarios is linearly normalized and mapped to the [0,100] interval; The scenario non-redundancy refers to the degree to which the simulation scenario does not overlap with other high-quality scenarios, thus avoiding waste of testing resources; it includes calculating the cosine similarity of the scenario feature vectors, extracting key features of the scenario (such as the number of traffic participants, road type, driving intention, vehicle speed range, etc.), and representing each scenario as a standardized vector. Then, similarity calculations are performed with other high-quality scenes, using vectors of high-quality scenes already included in the test library. Using the baseline as a reference, calculate the cosine similarity between the current scene and the baseline; The scene simulation accuracy refers to the degree of deviation between the physical model and the environmental model in the simulation scene and the real scene, including the root mean square error (RMSE) of the simulation data and the real test data, A=100-10×RMSE, with a value of [0,100] (RMSE≤10). The real-time performance of the scenario refers to the real-time response capability during the simulation scenario's runtime, reflecting the processing efficiency of the simulation platform; this is calculated by converting it into a frame rate compliance rate. The reproducibility of the scenario refers to the degree of consistency of the simulation scenario when it is repeatedly run under the same initial conditions. By repeatedly running the scenario, the coefficient of variation of key parameters is calculated, as shown in the following expression: CV≤2, Values ​​range from [0, 100].

[0020] Furthermore, the method for generating high-value simulation test scenarios based on the improved genetic algorithm includes using data from the simulation test scenario's complexity, coverage, danger, realism, non-redundancy, real-time performance, and reproducibility as decision variables. A hybrid encoding method is used, where discrete enumeration-type index data is encoded in binary and constrained within a given value range, continuous numerical indexes are encoded in real numbers, gene values ​​are directly taken from the index's physical quantity and constrained within a given value range, and integer continuous indexes are encoded in real numbers and then rounded down, preserving integer characteristics during evolution. Each chromosome corresponds to a complete simulation test scenario. Chromosome genes are all secondary scenario elements arranged sequentially. Assume there are m secondary indicators in the scenario. Individual chromosomes are represented as follows:

[0021] The gene value of the i-th secondary indicator corresponds to the values ​​of each indicator in Tables 1 to 5; Then perform population initialization, assuming the population size is... Randomly generated The initial population consists of chromosomes that satisfy the feature constraints. , Indicates the first Generation of population; determine the population size N, the number of chromosome genes m, and the upper and lower limits of each gene value, and randomly generate gene segments; among them, binary enumeration gene segments are uniformly and randomly generated in the coding space as 0 / 1 strings, which are mapped to the corresponding scene element values; real number gene segments are uniformly distributed and randomly sampled in the value range. Then, legality verification and repair are performed, including truncating genes that exceed the boundaries to the upper and lower limits; rounding integer indicators to the nearest integer and constraining them to the integer domain; eliminating invalid conflict scenarios and regenerating replacement individuals.

[0022] Furthermore, a fitness function is designed based on complexity, coverage, risk, realism, non-redundancy, scenario simulation accuracy, scenario real-time performance, and scenario reproducibility. This includes combined weight calculation. To balance objectivity and test requirements, a combined weight method is used to allocate feature weights. First, the objective weights are calculated using the entropy weight method. First, calculate the information entropy and objective weights, then calculate the subjective weights using the analytic hierarchy process (AHP). ; Further refine the selection process by combining dual-elite retention with Roulette-Tournament selection. The aforementioned dual-elite retention includes the first type of elites, retaining the second type. Fitness in the population decreases from high to low across generations. ( The chromosomes of the individual directly enter the next generation's elite pool; The second type of elite calculates the diversity index of each chromosome in the population. The higher the value, the greater the diversity; retain forward ( Chromosomes from two groups are added to the elite pool to ensure population diversity; the two elite groups together... One, directly replicated to the next generation population. ; The Roulette-Tournament hybrid selection includes the remaining One spot will be allocated using a mixed selection method: First, calculate the selection probability for each chromosome. (Basic choices for roulette) The population is then sorted in descending order of selection probability. The top 50% of individuals are selected using roulette wheel selection (to preserve superior genes), while the bottom 50% are selected using tournament selection to improve diversity. Repeat the selection until the page is full. One spot is allocated to form a post-selection population. ; Then perform a dynamic adaptive crossover operation, including generating a A uniformly random number is generated, and then a trigger judgment is performed: like If the chromosome undergoes crossover, it will enter the population awaiting crossover. ; like If the chromosome does not trigger crossing over, it will be directly retained as a offspring. After triggering the judgment, a population to be crossed is formed. Then perform dynamic mutation judgment to generate a Uniform random numbers; Trigger judgment, if If the chromosome mutates, it will enter the population to be mutated. ; like If the chromosome does not trigger a mutation, it will be directly retained as an offspring. After the above triggering judgment, a population to be mutated is formed. ,exist Chromosomes undergo mutations. The methods of chromosome mutation include multi-point mutation of the chromosome to be mutated, where 2-5 mutation points can be randomly selected and mutated according to different situations, as detailed below: This includes discrete enumeration-type gene mutations. For discrete enumeration-type genes such as road type, weather type, and driving intention, the gene values ​​are randomly selected from the gene value set, but it must be ensured that the values ​​are different after mutation. Then perform continuous numerical gene mutations, iterate until termination and output of results. The iteration termination condition is: the iteration stops when any of the following conditions are met: Condition 1: The number of iterations reaches the preset maximum number of generations. ; Condition 2: The change in optimal fitness value over 20 consecutive generations of the population; Then, scenario selection and evaluation are carried out, including outputting the top 30% of individuals in the final population with the highest fitness values, which correspond to high-value simulation test scenarios. Individuals with fitness values ​​between 30% and 70% in the final population correspond to medium-value simulation test scenarios; the rest correspond to low-value simulation test scenarios.

[0023] Furthermore, the optimization of high-value simulation test scenarios includes, based on the scenario's operating status, test indicator achievement data, and actual feedback data on the exposure effect of system defects, reverse-engineering the ineffective causes of pseudo-high-value scenarios that have no actual testing value after actual testing and verification, and implementing optimization measures. The optimization measures include scene optimization and dynamic fine-tuning of core parameters; the scene optimization involves adjusting the element parameters of pseudo-high-value scenes, including increasing the number of interfering vehicles, reducing the road surface adhesion coefficient, setting extreme weather conditions, or increasing the scene interaction complexity. The dynamic fine-tuning of the core parameters includes dynamically adjusting and improving the core operating parameters of the genetic algorithm; and simultaneously fine-tuning the combined weights of the scene value evaluation indicators. The optimized scenario is used as a new initial sample and fed back into the step of generating high-value simulation test scenarios based on the improved genetic algorithm, and then substituted back into the improved genetic algorithm for iterative optimization.

[0024] Furthermore, the present invention provides a computer storage medium storing a computer program that can run on a processor, wherein the processor executes the computer program to implement the method for generating a high-value simulation test scenario for intelligent connected vehicles as described above.

[0025] As can be seen from the above technical solution, the beneficial technical effects of the present invention are as follows: 1. This invention establishes a hierarchical simulation test scenario element framework covering road environment, traffic participants, weather and lighting, traffic control and communication, and driving behavior conditions. It refines these elements to secondary quantitative indicators and uniformly sets symbolic expressions and reasonable value ranges, overcoming the shortcomings of existing technologies such as fragmented scenario dimensions, ambiguous parameter definitions, and lack of quantitative constraints. It achieves full-dimensional coverage of the static road base, dynamic traffic interaction, natural environmental interference, network information transmission, and vehicle driving behavior for intelligent connected vehicles, accurately reflecting the characteristics of real and complex road traffic, and significantly improving the realism, completeness, and standardization of the simulation test scenario.

[0026] 2. This invention designs dedicated chromosome hybrid coding rules and compliant population initialization methods for different types of scene elements. It employs a hybrid selection process of dual elite retention + Roulette-Tournament, dynamic adaptive crossover, and dynamic adaptive mutation, solving the problems of traditional manual scene construction, such as high subjectivity, large workload, uneven distribution of randomly sampled scenes, and potential coverage blind spots. It can autonomously traverse element combinations within the parameter constraint space, efficiently mining extreme conditions, critical boundaries, and high-risk test scenarios, reducing manual design costs and upgrading simulation test scenarios from manual customization to intelligent automatic generation.

[0027] 3. This invention integrates parameter boundary constraints and scenario logic validity verification mechanisms throughout the entire algorithm process, avoiding the generation of invalid scenarios with out-of-bounds parameters and traffic logic conflicts. The generated scenarios are adaptable to various intelligent connected vehicle simulation test platforms. It can cover both common scenarios and rare dangerous conditions in batches, effectively making up for the shortcomings of traditional test scenario coverage. It fully verifies the robustness of the perception, decision-making, and planning control systems of intelligent connected vehicles in complex environments, reducing the risks and costs of real-vehicle road testing. At the same time, the modular and expandable element system is applicable to multiple scenarios, multiple vehicle models, and multiple testing and certification requirements, demonstrating outstanding versatility and engineering application value. Attached Figure Description

[0028] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0029] Figure 1 This is a flowchart of the method of the present invention; Figure 2 This is a structural diagram of the storage medium according to Embodiment 2 of the present invention. Detailed Implementation

[0030] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.

[0031] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application should have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0032] Example 1 like Figure 1 As shown, this embodiment provides a method for generating high-value simulation test scenarios for intelligent connected vehicles; Construct simulation test scenario elements including road environment, traffic participants, weather and lighting, traffic control and communication, and driving behavior conditions, forming a standardized scenario element framework covering all dimensions of static road base, dynamic traffic participation, natural environmental conditions, network communication control, and vehicle behavior decision-making; Construct a simulation test scenario feature system, including three types of simulation scenario features: scenario simulation accuracy, scenario real-time performance, and scenario reproducibility; Based on a standardized scenario element framework, the feature elements in the simulation test scenario feature system are used as decision variables, and a high-value simulation test scenario is generated based on an improved genetic algorithm. High-value simulation test scenarios are applied to closed-loop simulation testing of intelligent connected vehicles, and real-time data on scenario operation status, test indicator achievement, and actual feedback data on system defect exposure are collected; the high-value simulation test scenarios are then optimized.

[0033] In this embodiment, the road environment elements include road type, number of lanes, road longitudinal slope, road surface adhesion coefficient, lane width, and curve radius of curvature data; The traffic participant elements include data on the vehicle's initial speed, vehicle's initial position, number of background vehicles, background vehicle speed, background vehicle initial position, number of pedestrians, pedestrian crossing speed, and pedestrian initial position. The meteorological lighting elements mentioned include weather type, visibility, light intensity, rainfall intensity, snowfall intensity, wind speed, wind direction, and road surface water thickness data; Traffic control and communication elements include traffic sign type, traffic light timing, communication delay, communication packet loss rate, and communication distance data; Driving behavior and operating condition elements include the vehicle's driving intention, longitudinal acceleration, lateral acceleration, lane change lateral deviation rate, following safety distance, and following time distance data.

[0034] In this embodiment, the construction of the simulation test scenario feature system includes scenario complexity, scenario coverage, scenario danger, scenario realism, scenario non-redundancy, scenario real-time performance, and scenario reproducibility. The aforementioned scene complexity includes decomposing scene complexity into multiple secondary evaluation indicators, for ;For example Indicates the number of traffic participants, This represents the topological complexity.

[0035] In this embodiment, the secondary indicators of road environment element R are shown in Table 1; the secondary indicators of traffic participant element V are shown in Table 1. As shown in Table 2, the secondary indicators of meteorological illumination element W are shown in Table 3; the secondary indicators of traffic control and communication element C are shown in Table 4; and the secondary indicators of driving behavior and operating condition element B are shown in Table 5. Table 1. Secondary Indicators of Road Environmental Element R

[0036] Traffic participant element V includes the vehicle's initial speed, initial position, number of background vehicles, speed of background vehicles, initial position of background vehicles, number of pedestrians, pedestrian crossing speed, and initial position of pedestrians, as detailed in the table below.

[0037] Table 2 Secondary Indicators of Traffic Participant Element V

[0038] Meteorological illumination elements W include weather type, visibility, illumination intensity, rainfall intensity, snowfall intensity, wind speed, wind direction, and road surface water thickness, as detailed in the table below.

[0039] Table 3 Secondary Indices of Meteorological Light Element W

[0040] Traffic control and communication elements C Traffic control and communication elements include traffic sign type, traffic light timing, communication delay, communication packet loss rate, and communication distance, as detailed in the table below.

[0041] Table 4. Secondary Indicators of Traffic Control and Communication Element C

[0042] Driving behavior and operating condition element B includes the driver's driving intention, longitudinal acceleration, lateral acceleration, lane change lateral deviation rate, following safety distance, following time distance, etc., as detailed in the table below.

[0043] Table 5. Secondary Indicators of Driving Behavior and Operating Condition Element B

[0044] Specifically, each secondary evaluation indicator is scored by experts on a scale of 1 to 5 based on the scenario, resulting in the original scoring vector for each scenario. Then perform subjective weight calculation: Construct a judgment matrix by having experts score the pairwise importance of each indicator (e.g., "number of participants" is slightly more important than "road topology," so it's scored 3 points). Calculate the largest eigenvalue of the judgment matrix. The subjective weights are obtained by normalizing the corresponding feature vectors. ; Perform a consistency check; ,like If so, then the weights are valid; Next, the objective weights are calculated, including the standardization of the evaluation index scores for all scenarios. Let the total number of scenarios be m, and the total number of secondary evaluation indicators for scenario complexity C be n, to obtain the standardization matrix. ;

[0045] Calculate the information entropy of each secondary evaluation indicator.

[0046] in , ( (Total number of scenes).

[0047] Calculate objective weights .

[0048] Then calculate the combined weights and the final score. The expression for calculating the combined weights is as follows: ( This represents the subjective weighting, typically set at 0.5.

[0049] Specifically, for a single scenario, a weighted score is calculated. For all scenarios Perform linear normalization and map to The range is used to obtain the value of the complexity C for each scenario, as shown in the following expression:

[0050] The scenario coverage rate R refers to the degree to which the simulation scenario covers the functional requirements of intelligent connected vehicles (such as adaptive cruise control (ACC), automatic emergency braking (AEB), and navigation assist (NOA)). Based on the functional requirements document, the matching rate between the scenario and the requirement points is calculated, with a value range of [0, 100]. The steps are as follows: Specifically, a functional requirements list is compiled, outlining all functional requirements of the system under test, such as ACC (Adaptive Cruise Control), AEB (Automatic Emergency Braking), and NOA (Navigation-Assisted Driving). Each function is further subdivided into sub-requirements (e.g., AEB's recognition of pedestrians, bicycles, and stationary vehicles), resulting in a requirement set. ,common One demand point.

[0051] The scenario and requirement matching determination is performed. For each simulation scenario, it is determined whether it can cover each requirement (0 = no coverage, 1 = coverage), and a matching vector is obtained.

[0052] in This indicates that the scenario can cover the requirements. .

[0053] Specifically, the coverage rate is then calculated, and the coverage rate value is mapped to... The coverage value for each scene is obtained from the interval, as shown in the following expression:

[0054] The scenario hazard refers to the probability of intelligent connected vehicles facing safety risks such as collisions and loss of control in the simulated scenario, which is quantified by combining virtual TTC (time headway) and THW (distance between vehicles); the expression is as follows:

[0055] The value range is [0, 100] (the larger the value, the higher the risk); the specific steps are as follows: first, data collection is performed: Specifically, during the simulation, the minimum values ​​(or the values ​​at the peak risk moment) of the time-to-column distance (TTC) and the time-to-headway distance (THW) between the target vehicle and the intelligent vehicle can be collected in real time and recorded as follows: and .

[0056] Where TTC is the distance between the two vehicles / relative speed (unit: s), the smaller the value, the higher the risk. THW is the distance between the two vehicles / vehicle speed (unit: s), the higher the value, the higher the risk. The calculation formula is as follows:

[0057] The coefficient 0.4 in the formula is for illustrative purposes only and can be adjusted according to testing standards. The larger the D value, the higher the risk.

[0058] The range constraint truncates the result if it exceeds the range [0, 100]. The expression is as follows:

[0059] In this embodiment, the scene realism refers to the similarity between the simulated scene and real road data (such as the NGSIM / HighD dataset). The Dynamic Time Warping (DTW) algorithm is used to calculate the feature vector similarity, with values ​​ranging from [0, 100]. The specific steps are as follows: First, a feature vector is constructed. In the real-world scenario, trajectory features (position, velocity, acceleration) of the target vehicle are extracted from public datasets such as NGSIM / HighD to construct a time series vector. Run the scenario in the simulation platform and extract time series vectors with the same features. .

[0060] Then perform DTW distance calculation and construct... The distance matrix, where (Euclidean distance).

[0061] Dynamic programming is used to find the shortest regular path to obtain the DTW distance. The smaller the distance, the more similar the trajectories.

[0062] Then, the DTW distance for all scenes is linearly normalized and mapped to the [0, 100] interval, as shown in the following expression:

[0063] Among them, the smaller the distance, the higher the T value, and the better the authenticity.

[0064] Specifically, the scene non-redundancy refers to the degree to which the simulated scene does not overlap with other high-quality scenes, thus avoiding waste of testing resources. This includes calculating the cosine similarity of scene feature vectors, expressed as follows:

[0065] Values ​​range from [0, 100]. The larger the value, the higher the non-redundancy. Specific steps include constructing scene feature vectors: Extract key features of the scene (such as the number of traffic participants, road type, driving intention, speed range, etc.), and represent each scene as a standardized vector. .

[0066] Specifically, similarity calculations are then performed with other high-quality scenes, using vectors of high-quality scenes already included in the test library. Using the baseline as a reference, calculate the cosine similarity between the current scene and the baseline:

[0067] The non-redundancy score is calculated using the following expression: The value range is [0, 100]. A larger value indicates a greater difference between the scenario and existing scenarios, and a higher degree of non-redundancy; a smaller value indicates a higher degree of scenario repetition and redundancy.

[0068] Specifically, the scene simulation accuracy refers to the degree of deviation between the physical model (such as vehicle dynamics, tire model) and environmental model (such as weather, lighting) in the simulation scene and the real scene, including the root mean square error (RMSE) of comparing simulation data with real test data, A=100-10×RMSE, with a value of [0,100] (RMSE≤10). First, data acquisition is performed. In actual vehicle or bench testing, real data sequences of target physical quantities (such as vehicle speed, acceleration, steering angle, tire force, etc.) are collected. .

[0069] Under the same input conditions, the corresponding data sequence output by the simulation platform is collected. Next, the root mean square error (RMSE) is calculated, as shown in the following expression:

[0070] The precision score is then calculated using the following expression:

[0071] This expression applies to cases where RMSE < 10. If RMSE > 10, the score is truncated to 0. The value range is [0, 100]. The smaller the RMSE, the higher the A value and the better the precision.

[0072] Specifically, the real-time performance of the scenario refers to the real-time response capability during simulation, reflecting the processing efficiency of the simulation platform. This is converted into a calculation of the frame rate compliance rate, and the specific steps are as follows: First, run the simulation scenario and calculate the actual average frame rate of the simulation platform. (Unit: frames per second), and simultaneously determine the target frame rate required for the test. (e.g., 30fps or 60fps).

[0073] Calculate the frame rate compliance rate:

[0074] Further range constraints are applied, but the actual frame rate cannot be infinitely high, so truncation is performed:

[0075] The value range is [0, 100]. When the actual frame rate is greater than the target frame rate, L = 100, which means that the real-time requirements are fully met.

[0076] The reproducibility of the scenario refers to the degree of consistency of the simulation scenario when it is repeatedly run under the same initial conditions. By repeatedly running the scenario, the coefficient of variation of key parameters (such as vehicle speed and position) is calculated, as shown in the following expression: CV≤2, Values ​​[0, 100] Specifically, the steps are as follows: Under the same initial conditions (initial position, vehicle speed, environmental parameters), run the scenario n times (e.g., n=5 times) and collect the result sequence of key parameters (such as vehicle position, speed, acceleration). The mean of the sequence is calculated using the following expression:

[0077] The standard deviation is calculated using the following expression:

[0078] The coefficient of variation is calculated using the following expression:

[0079] The reproducibility score is calculated using the following expression:

[0080] This formula applies to In the case where CV > 2, the score is directly truncated to 0. The value range is [0, 100]. The smaller the CV, the higher the score. The higher the value, the better the reproducibility.

[0081] In this embodiment, the generation of high-value simulation test scenarios based on the improved genetic algorithm includes using data from the simulation test scenario's complexity, coverage, hazard, realism, non-redundancy, real-time performance, and reproducibility as decision variables, and employing hybrid encoding. Discrete enumeration indicators (such as road type, weather type, driving intention, etc.): use binary encoding and are constrained to a given value range.

[0082] For continuous numerical indicators (vehicle speed, gradient, adhesion coefficient, time delay, etc.): real number encoding is used, and the gene value is directly taken from the physical quantity of the indicator, constrained within a given value range. For integer continuous indicators (such as the number of road lanes, the number of background vehicles), real number encoding is used followed by rounding constraints, and integer characteristics are preserved during evolution.

[0083] Each chromosome corresponds to a complete simulation test scenario. Chromosome genes = all secondary scenario elements arranged in sequence. Assume there are m secondary indicators in the scenario. Individual chromosomes are represented as follows:

[0084] The gene value of the i-th secondary indicator corresponds to the values ​​of each indicator in Tables 1 to 5.

[0085] Then perform population initialization, assuming the population size is... Randomly generated The initial population consists of chromosomes that satisfy the feature constraints. , Indicates the first Generation population. Determine the population size N (based on testing requirements), the number of genes on the chromosome m, and the upper and lower limits of each gene value, and generate the population randomly by gene segment; Specifically, binary enumeration gene segments generate 0 / 1 strings uniformly and randomly in the encoding space, which are mapped to the values ​​of the corresponding scene elements. Real number gene segments are randomly sampled in a uniform distribution within the value range.

[0086] Then, legality verification and repair are performed, including truncating genes that exceed the boundaries to the upper and lower limits; rounding integer indicators and constraining them to the integer domain; and eliminating invalid conflict scenarios (such as unreasonable scenarios like too many pedestrians on the highway or traffic light timings on the highway) and regenerating replacement individuals.

[0087] Specifically, a fitness function is designed based on complexity, coverage, risk, realism, non-redundancy, scenario simulation accuracy, scenario real-time performance, and scenario reproducibility. This includes a combined weight calculation (entropy weight method + analytic hierarchy process). To balance objectivity and test requirements, a combined weight method is used to allocate feature weights. First, the objective weights are calculated using the entropy weight method. (Based on the degree of feature dispersion): Feature normalization (positive index), the expression is as follows:

[0088] in For the first The first scenario Original values ​​of features (e.g.) Indicates the first The first feature of each scene (i.e., the scene complexity C value). , The first The minimum and maximum values ​​of each feature; The information entropy is calculated using the following expression: ; The objective weights are calculated using the following expression: ( (Corresponding to 8 features).

[0089] Subjective weights are then calculated using the Analytic Hierarchy Process (AHP). (Based on testing requirements, such as the focus of security testing) Real-time testing focuses on ),in,

[0090] Combined weights:

[0091] in Weighting coefficients (can be taken as follows) (While considering both customer perspectives and subjective needs), and satisfying: .

[0092] fitness function The fitness value comprehensively reflects the value of the simulation scenario. The higher the fitness value, the better the scenario and the higher its value. The expression is as follows:

[0093] in, : No. The fitness value for each simulation scenario ranges from [0, 100]. : No. The first scenario Normalized values ​​of each feature (values ​​[0,1]); : No. The combined weight coefficients of each feature; Diversity reward coefficient ( This is used to reward individuals in the population that have low similarity to other scenarios, thus avoiding local optima; : No. The population diversity index for each scenario is expressed as follows:

[0094]

[0095] The cosine similarity of the feature vectors.

[0096] Specifically, further refinement and selection operations will be performed (retaining dual elites + Roulette-Tournament hybrid selection). The aforementioned dual-elite retention includes the first type of elites, retaining the second type. Fitness in the population decreases from high to low across generations. ( The chromosomes of the individual directly enter the next generation's elite pool; The second type of elite calculates the diversity index of each chromosome in the population. The higher the value, the greater the diversity. (Retain) forward ( Chromosomes from two elite groups are added to the elite pool to ensure population diversity. One, directly replicated to the next generation population. .in The function expression is as follows:

[0097]

[0098] Where P represents the population. In the population , The Euclidean distance between the eigenvectors of two chromosomes Chromosomes The j-th eigenvalue, Chromosomes The j-th eigenvalue.

[0099] Roulette-Tournament blending selection, for the remaining One spot will be allocated using a mixed selection method: First, calculate the selection probability for each chromosome. (Basic choices for roulette) The population is then sorted in descending order of selection probability. The top 50% of individuals are selected using roulette wheel selection (to preserve superior genes), while the bottom 50% are selected using tournament selection to improve diversity. Specifically, repeat the selection until the page is filled. One spot is allocated to form a post-selection population. .

[0100] Then, a dynamic adaptive crossover operation is performed, where the dynamic adjustment expression for the crossover probability includes the early stage of the iteration ( Linear Decreasing Phase

[0101]

[0102] in, Maximum crossover probability (value 0.95). Minimum crossover probability (value 0.6); : Current iteration number : No. Chromosomes in generational time (scenario) The probability of crossover. Maximum number of iterations; , : No. The minimum and maximum fitness values ​​of chromosomes in the population after generation selection. Individuals with higher chromosome fitness have a higher crossover probability.

[0103] Later stages of iteration The adaptive phase based on fitness distribution is expressed as follows:

[0104] in, : No. Era Scene The probability of crossover. Maximum number of iterations; , , : No. The average, minimum, and maximum fitness values ​​of the population. Higher fitness results in a lower crossover probability, preventing the destruction of high-quality chromosomes. Individuals with fitness below the population average have a higher crossover probability as fitness decreases, increasing their evolutionary opportunities.

[0105] Then perform dynamic cross-validation, including: Generate a A uniformly random number is generated, and then a trigger judgment is performed: like If the chromosome undergoes crossover, it will enter the population awaiting crossover. .

[0106] like If the chromosome does not trigger crossing over, it will be directly retained as a offspring.

[0107] After triggering the judgment, a population to be crossed is formed. ,exist Randomly pair up and cross over.

[0108] Specifically, multiple crossovers are performed on two chromosomes. The number of crossover points can be 2-5, and the crossover points (genes) can be randomly selected. Assuming there are m genes (secondary indicators) on the chromosomes, two crossover points (genes) are selected as... , The specific crossover process is as follows: Assume the two paternal chromosomes are respectively

[0109]

[0110] The new chromosomes obtained after crossover are as follows:

[0111]

[0112] Then, a dynamic adaptive mutation operation is performed, where the mutation probability is dynamically adjusted as follows:

[0113] in, Maximum mutation probability (value 0.015). Minimum mutation probability (value 0.001); : Fitness value of the chromosome to be mutated; Individuals with fitness below average are given the maximum mutation probability to increase the chance of generating superior genes; individuals with fitness above average are given a lower mutation probability as fitness increases to avoid destroying superior genes.

[0114] Then perform dynamic mutation judgment to generate a Uniform random numbers.

[0115] Trigger judgment, if If the chromosome mutates, it will enter the population to be mutated. .

[0116] like If the chromosome does not trigger a mutation, it will be directly retained as an offspring.

[0117] After the above triggering judgment, a population to be mutated is formed. ,exist Chromosomes undergo mutations.

[0118] The mutation methods include multi-point mutation of the chromosome to be mutated, randomly selecting 2-5 mutation points (genes) and performing mutations according to different situations, as detailed below: This includes discrete enumeration-type gene mutations. For discrete enumeration-type genes such as road type, weather type, and driving intention, the gene values ​​are randomly selected from the gene value set, but it must be ensured that the values ​​are different after mutation.

[0119] Specifically, continuous numerical gene mutations are then performed. For continuous numerical genes such as vehicle speed, gradient, adhesion coefficient, and time delay, a polynomial mutation method is used.

[0120]

[0121]

[0122] in For the values ​​of the mutated genes, Gene values ​​before mutation The variation perturbation factor. A random number in the range [0,1]. This is the variation distribution index, which controls the magnitude of variation; it is generally set to a value of 6. This represents the upper limit of the gene's possible values. This represents the lower limit of the gene's value.

[0123] Then proceed with the iteration termination and result output. The iteration termination condition is as follows: iteration stops when any of the following conditions are met: Condition 1: The number of iterations reaches the preset maximum number of generations. (like: ); Condition 2: The change in optimal fitness value of the population over 20 consecutive generations. ( ) And the change in average fitness value ( ).

[0124] in: The fitness value of the chromosome with the highest fitness in the t-th generation population; Counting back 20 generations, the optimal fitness value in the population at that time; : The average fitness value of all individuals in the t-th generation population; Counting back 20 generations, the average fitness value of the population at that time.

[0125] Specifically, the next step is to screen and evaluate the scenarios, including outputting the top 30% of individuals in the final population based on their fitness values, which correspond to high-value simulation test scenarios. Individuals with a fitness value between 30% and 70% in the final population correspond to the mid-value simulation test scenario. The rest are low-value simulation test scenarios.

[0126] In this embodiment, the optimization of high-value simulation test scenarios includes, based on the scenario's running status, test indicator achievement data, and actual feedback data on the exposure effect of system defects, reverse-engineering the ineffective causes of pseudo-high-value scenarios that have no actual test value after actual testing and verification, and implementing optimization measures. The optimization measures include scene optimization and dynamic fine-tuning of core parameters; the scene optimization involves adjusting the element parameters of pseudo-high-value scenes, including increasing the number of interfering vehicles, reducing the road surface adhesion coefficient, setting extreme weather conditions, or increasing the scene interaction complexity; enhancing the danger level and testing relevance of the scene. The dynamic fine-tuning of the core parameters includes dynamically adjusting and improving the core operating parameters of the genetic algorithm; for example, optimizing the population size, adjusting the core operating parameters in crossover probability and mutation probability, and fine-tuning the combined weights of the scenario value evaluation indicators. The optimized scenario is used as a new initial sample and fed back into the step of generating high-value simulation test scenarios based on the improved genetic algorithm, and then substituted back into the improved genetic algorithm for iterative optimization.

[0127] Example 2 The present invention provides a computer storage medium storing a computer program that can run on a processor. When the processor executes the computer program, it implements a method for generating a high-value simulation test scenario for intelligent connected vehicles as described above.

[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A method for generating high-value simulation test scenarios for intelligent connected vehicles, characterized in that, include: Construct simulation test scenario elements including road environment, traffic participants, weather and lighting, traffic control and communication, and driving behavior conditions, forming a standardized scenario element framework covering all dimensions of static road base, dynamic traffic participation, natural environmental conditions, network communication control, and vehicle behavior decision-making; Construct a simulation test scenario feature system, including three types of simulation scenario features: scenario simulation accuracy, scenario real-time performance, and scenario reproducibility; Based on a standardized scenario element framework, the feature elements in the simulation test scenario feature system are used as decision variables, and a high-value simulation test scenario is generated based on an improved genetic algorithm. High-value simulation test scenarios are applied to closed-loop simulation testing of intelligent connected vehicles, and real-time data on scenario operation status, test indicator achievement data, and actual feedback data on the exposure effect of system defects are collected. Optimize high-value simulation test scenarios.

2. The method for generating high-value simulation test scenarios for intelligent connected vehicles according to claim 1, characterized in that, The road environment elements mentioned include road type, number of lanes, road longitudinal slope, road surface adhesion coefficient, lane width, and curve radius of curvature data; The traffic participant elements include data on the vehicle's initial speed, vehicle's initial position, number of background vehicles, background vehicle speed, background vehicle initial position, number of pedestrians, pedestrian crossing speed, and pedestrian initial position. The meteorological lighting elements mentioned include weather type, visibility, light intensity, rainfall intensity, snowfall intensity, wind speed, wind direction, and road surface water thickness data; Traffic control and communication elements include traffic sign type, traffic light timing, communication delay, communication packet loss rate, and communication distance data; Driving behavior and operating condition elements include the vehicle's driving intention, longitudinal acceleration, lateral acceleration, lane change lateral deviation rate, following safety distance, and following time distance data.

3. The method for generating high-value simulation test scenarios for intelligent connected vehicles according to claim 1, characterized in that, The simulation test scenario feature system is constructed as follows: scenario complexity, scenario coverage, scenario danger, scenario realism, scenario non-redundancy, scenario real-time performance, and scenario reproducibility. The aforementioned scene complexity includes decomposing scene complexity into multiple secondary evaluation indicators, for ; For each secondary evaluation indicator, experts scored it from 1 to 5 points according to the scenario, resulting in the original score vector for each scenario. Then perform subjective weight calculation: Reconstruct the judgment matrix and have experts score the pairwise importance of each indicator; Calculate the largest eigenvalue of the judgment matrix The subjective weights are obtained by normalizing the corresponding feature vectors. ; Perform a consistency check; ,like If so, then the weights are valid; Next, the objective weights are calculated, including the standardization of the evaluation index scores for all scenarios. Let the total number of scenarios be m, and the total number of secondary evaluation indicators for scenario complexity C be n, to obtain the standardization matrix. ; Calculate the information entropy of each secondary evaluation indicator. in , ; Calculate objective weights ; Then calculate the combined weights and the final score. The expression for calculating the combined weights is as follows: 。 4. The method for generating high-value simulation test scenarios for intelligent connected vehicles according to claim 3, characterized in that, For a single scene, the weighted score is calculated as follows: For all scenarios Perform linear normalization and map to The range is used to obtain the value of the complexity C for each scenario, as shown in the following expression: The scenario coverage rate R is the degree to which the simulation scenario covers the functional requirements of intelligent connected vehicles. Based on the functional requirements document, the matching rate between the scenario and the requirement points is calculated, with a value range of [0, 100].

5. The method for generating high-value simulation test scenarios for intelligent connected vehicles according to claim 4, characterized in that, The steps for calculating the matching rate between the scenario and the requirement are as follows: First, sort out the functional requirement list and list all the functional requirement points of the system under test. Then, subdivide each function into sub-requirements to obtain the requirement set. ,common One demand point; Scenario and requirement matching determination: For each simulation scenario, determine whether it can cover each requirement point to obtain a matching vector; in This indicates that the scenario can cover the requirements. ; Then calculate the coverage rate and map the coverage rate values ​​to... The coverage value for each scene is obtained from the interval, as shown in the following expression: The scenario hazard refers to the probability of intelligent connected vehicles facing collision and loss of control safety risks in the simulated scenario. The probability of intelligent connected vehicles facing collision and loss of control safety risks is quantified by combining virtual TTC and THW.

6. The method for generating high-value simulation test scenarios for intelligent connected vehicles according to claim 5, characterized in that, The probability of collision and loss of control safety risks faced by intelligent connected vehicles is quantified by combining virtual TTC and THW. The specific steps are as follows: First, data acquisition is performed. During the simulation, the minimum values ​​of the time headway (TTC) and headway (THW) between the target vehicle and the intelligent vehicle are collected in real time and denoted as... and ; Where TTC is the distance between the two vehicles / relative speed, the smaller the value, the higher the risk; THW is the distance between the two vehicles / vehicle speed, the higher the value, the higher the risk. The expression is as follows: The higher the D value, the higher the risk; Next, apply range constraints. If the calculated result exceeds the [0, 100] interval, truncation is performed, as shown in the following expression: The scene realism refers to the similarity between the simulated scene and real road data. The feature vector similarity is calculated using a dynamic time warping algorithm, with values ​​ranging from [0, 100]. The steps are as follows: First, feature vectors are constructed, then DTW distance is calculated. Given the distance matrix, we use dynamic programming to find the shortest regular path and obtain the DTW distance. The smaller the distance, the more similar the trajectories; Then, the DTW distance for all scenarios is linearly normalized and mapped to the [0,100] interval; The non-redundancy of the scenario refers to the degree to which the simulation scenario does not overlap with other high-quality scenarios, thus avoiding waste of test resources; This includes calculating the cosine similarity of scene feature vectors, extracting key features of scenes, and representing each scene as a standardized vector; Then, similarity calculations are performed with other high-quality scenes, using vectors of high-quality scenes already included in the test library. Using the baseline as a reference, calculate the cosine similarity between the current scene and the baseline; The scene simulation accuracy refers to the degree of deviation between the physical model and the environmental model in the simulation scene and the real scene, including the root mean square error of the simulation data and the real test data. The real-time performance of the scenario refers to the real-time response capability of the simulation scenario during runtime, and the reproducibility of the scenario refers to the degree of consistency of the simulation scenario when it is repeatedly run under the same initial conditions. By repeatedly running the scenario, the coefficient of variation of key parameters is calculated.

7. The method for generating high-value simulation test scenarios for intelligent connected vehicles according to claim 1, characterized in that, The method for generating high-value simulation test scenarios based on improved genetic algorithms includes using data from scenario complexity, scenario coverage, scenario danger, scenario realism, scenario non-redundancy, scenario real-time performance, and scenario reproducibility as decision variables. A hybrid encoding method is used, in which discrete enumeration index data is encoded in binary and constrained within a given value range, continuous numerical indexes are encoded in real numbers, gene values ​​are directly taken from the physical quantities of the indexes and constrained within a given value range, and integer continuous indexes are encoded in real numbers and then rounded down to the nearest integer. Integer characteristics are preserved during evolution. Each chromosome corresponds to a complete simulation test scenario. Chromosome genes are all secondary scenario elements arranged sequentially. Assume there are m secondary indicators in the scenario. Individual chromosomes are represented as follows: in, Let be the gene value of the i-th secondary indicator; Then perform population initialization, assuming the population size is... Randomly generated The initial population consists of chromosomes that satisfy the feature constraints. ; Then, legality verification and repair are performed, including truncating genes that exceed the boundaries to the upper and lower limits; rounding integer indicators to the nearest integer and constraining them to the integer domain; eliminating invalid conflict scenarios and regenerating replacement individuals.

8. The method for generating high-value simulation test scenarios for intelligent connected vehicles according to claim 7, characterized in that, The fitness function is designed based on complexity, coverage, risk, realism, non-redundancy, scene simulation accuracy, scene real-time performance, and scene reproducibility. First, the objective weight is calculated using the entropy weight method, the information entropy is calculated, and the objective weight is calculated again. Then, the subjective weight is calculated using the analytic hierarchy process. Further refine the selection process by combining dual-elite retention with Roulette-Tournament selection. The aforementioned dual-elite retention includes the first type of elites, retaining the second type. Fitness in the population decreases from high to low across generations. The chromosomes of these individuals are directly transferred to the next generation's elite pool. The second type of elite calculates the diversity index of each chromosome in the population. ; reserve forward The chromosomes are added to the elite pool and directly replicated into the next generation of the population. ; The Roulette-Tournament hybrid selection includes the remaining One spot will be allocated using a mixed selection method: First, the selection probability of each chromosome is calculated; then the population is sorted in descending order of selection probability, the top 50% of individuals are selected using roulette wheel selection, and the bottom 50% of individuals are selected using tournament selection, thus improving diversity. Repeat the selection until the page is full. One spot is allocated to form a post-selection population. ; Then perform a dynamic adaptive crossover operation, including generating a A uniformly random number is generated, and then a trigger judgment is performed: like If the chromosome undergoes crossover, it will enter the population awaiting crossover. ; like If the chromosome does not trigger crossing over, it will be directly retained as a offspring. After triggering the judgment, a population to be crossed is formed. ; Then perform dynamic mutation judgment to generate a Uniform random numbers; Perform a trigger check if If the chromosome mutates, it will enter the population to be mutated. ; like If the chromosome does not trigger a mutation, it will be directly retained as an offspring. After triggering the judgment, a population to be mutated is formed. ,exist Chromosomes undergo mutations.

9. The method for generating high-value simulation test scenarios for intelligent connected vehicles according to claim 1, characterized in that, The optimization of high-value simulation test scenarios includes, based on the scenario's operating status, test indicator achievement data, and actual feedback data on the exposure effect of system defects, reverse-engineering the ineffective causes of pseudo-high-value scenarios that have no actual testing value after actual testing and verification, and implementing optimization measures. The optimization measures include scene optimization and dynamic fine-tuning of core parameters; the scene optimization involves adjusting the element parameters of pseudo-high-value scenes, including increasing the number of interfering vehicles, reducing the road surface adhesion coefficient, setting extreme weather conditions, or increasing the scene interaction complexity. The dynamic fine-tuning of the core parameters includes dynamically adjusting and improving the core operating parameters of the genetic algorithm; and simultaneously fine-tuning the combined weights of the scene value evaluation indicators. The optimized scenario is used as a new initial sample and fed back into the step of generating high-value simulation test scenarios based on the improved genetic algorithm, and then substituted back into the improved genetic algorithm for iterative optimization.

10. A computer storage medium storing a computer program executable on a processor, characterized in that, When the processor executes the computer program, it implements a method for generating a high-value simulation test scenario for intelligent connected vehicles according to any one of claims 1-9.

Images