An Accelerated Testing Method for Autopilot Based on Improved Genetic Algorithm

Abstract

The present invention is an accelerated test method for autonomous driving based on an improved genetic algorithm. It includes: First, determining the key scenario parameter types, ranges, distributions, and fitness functions of the system to be tested; Second, establishing the sorting logic of each scenario element, performing floating-point encoding, and generating an initial population; Third, decoding individuals to generate phenotypes and simulating the individuals within the population; Fourth, performing encoding and screening out inferior individuals within the population according to the simulation results; Fifth, using an improved elitist retention selection operator, heuristic crossover, and non-uniform mutation operator to perform chromosome crossover and mutation operations to generate offspring; Sixth, calculating the population redundancy according to the individual information within the population and performing redundancy screening; Seventh, setting termination conditions and interrupting the test process in a timely manner. The present invention has the ability to quickly optimize in the initial stage of search and continue to explore after finding some optimal solutions, and can accelerate the search for a large number of dangerous boundary scenarios that are beneficial to the iterative optimization of autonomous driving algorithms.

Description

Technical Field

[0001] The present invention belongs to the technical field of automobiles, and specifically relates to an autonomous driving acceleration test method based on an improved genetic algorithm. Background Art

[0002] With the continuous maturity of autonomous driving technology, how to verify the safety of autonomous driving technology has become the main obstacle restricting the mass production of autonomous driving vehicles on the road. The scenario-based virtual simulation technology is an effective means to test the performance of autonomous driving vehicles. The core of virtual simulation testing is the test case generation technology that combines scenario generation with test requirements. However, simulation testing faces challenges such as a large dimension of scenario parameters and a large proportion of invalid test cases, which severely restricts the testing process of autonomous driving vehicles.

[0003] The autonomous driving vehicle acceleration test method based on optimization search takes the dangerous boundary scenario with higher test value as the search target, combines intelligent optimization algorithms, and performs intensive search in the scenario parameter space, thereby greatly improving the test efficiency and reducing the simulation resource occupancy. At present, the optimization search algorithm for dangerous scenarios is slow at the initial stage of execution, severely restricting the ability to quickly find the optimal solution for key scenarios. And each optimization algorithm mainly focuses on the search speed. After finding a certain optimal solution, the population is prone to fall into local optimality and lacks the ability to continuously explore a large number of optimal solutions. The dangerous boundary scenario of autonomous driving vehicles is a parameter space with many solutions, and the current search algorithms are difficult to meet the demand for finding a large number of optimal solutions. There is an urgent need for an optimization search strategy that simultaneously has the abilities of global fast optimization and continuous exploration to improve the test efficiency. Summary of the Invention

[0004] The present invention provides an autonomous driving acceleration test method based on an improved genetic algorithm, which improves the selection, crossover, mutation and other operators of the traditional genetic method, and newly adds a repetition degree screening and inferior individual screening module, with the ability to quickly find the optimal solution at the initial stage of search and still continuously explore after finding some optimal solutions, so as to accelerate the search for a large number of dangerous boundary scenarios beneficial to the iterative optimization of the autonomous driving algorithm, accelerate the test process of the autonomous driving algorithm under test in the logical scenario, and solve the above problems existing in the existing acceleration test methods.

[0005] The technical solution of the present invention is described in conjunction with the accompanying drawings as follows:

[0006] An autonomous driving acceleration test method based on an improved genetic algorithm includes the following steps:

[0007] Step 1: Determine the key scenario parameter types, ranges and distribution conditions of the system under test as the parameter space optimized by the improved genetic algorithm; and determine the fitness function to guide the evolution process of the genetic algorithm;

[0008] Step 2: Establish the sorting logic of each scenario element, perform floating-point encoding, and generate the initial population;

[0009] Step 3: Decode the individuals to generate phenotypes, import them into the simulation system, and introduce a parallel acceleration module to perform rapid simulation of the individuals within the population;

[0010] Step 4: Perform encoding and screen out inferior individuals within the population according to the simulation results in Step 3;

[0011] Step 5: Use the improved elitist retention selection operator, heuristic crossover, and non-uniform mutation operator to perform chromosome crossover and mutation operations to generate offspring;

[0012] Step 6: Calculate the population redundancy based on the individual information within the population and perform redundancy screening;

[0013] Step 7: Repeat Steps 3 - 6 and set termination conditions to interrupt the test process in a timely manner.

[0014] Furthermore, the specific method of Step 1 is as follows:

[0015] 11) Generate specific test cases for autonomous driving tests based on scenarios in the order of functional scenarios - logical scenarios - specific scenarios, thereby establishing the types, ranges, and distribution of key scenario parameters of the system under test;

[0016] 12) Establish four stages of fitness functions according to the characteristics of the dangerous boundary scenario, dangerous scenario, general safety scenario, and unreasonable scenario. Among them, the fitness function stage value of the dangerous boundary scenario is the highest, the fitness function stage value of the dangerous scenario is moderate, the fitness function stage value of the general safety scenario is lower, and the fitness function stage value of the unreasonable scenario is the lowest.

[0017] Furthermore, the specific method of Step 2 is as follows:

[0018] 21) Based on the types, ranges, and distribution of key scenario parameters of the system under test obtained in Step 1, establish the arrangement order of each element and use floating-point encoding;

[0019] 22) Determine the population size of the genetic process and generate the initial population encoding by means of random values.

[0020] Furthermore, the specific method of Step 3 is as follows:

[0021] 31) On the basis of the initial population encoding, decode the individuals to generate specific scenario parameters, import them into the simulation system, and perform automated testing;

[0022] 32) Introduce a parallel acceleration module, use multi-device cloud data for synchronous simulation, and save the simulation scene parameters and results in the scene library.

[0023] Furthermore, the specific method of step 4 is as follows:

[0024] The superior and inferior characteristics are established according to the operating results of the individuals in the population, and the inferior individuals, i.e. the individuals whose fitness values ​​are less than the threshold, are deleted.

[0025] Furthermore, the specific method of step five is as follows:

[0026] 51) At the selection operator, roulette wheel and elite retention selection methods are used. The roulette wheel selection method proportional to the fitness is used for individuals, and the current optimal individuals are retained in proportion to continuously guide the population evolution; the selection probability formula of the xith individual in the roulette wheel selection method is as follows:

[0027]

[0028] In the formula, f(xi) is the fitness value of the xith individual; N is the number of individuals in the population after removing the elite individuals;

[0029] 52) At the crossover operator, a heuristic crossover method is used to perform a heuristic transformation of the offspring individuals according to the fitness value of the parents; two parents are randomly selected from the population and compared to move the parameters of the worse individual toward the better individual;

[0030] 53) At the mutation operator, a non-uniform mutation method is used to establish the repeatability according to the distribution of individual parameters of the population. The standard is that both variables are within a discrete step length interval, which is considered to be repeated; the population repeatability is used as an indicator for selecting the mutation probability; the population repeatability is inversely proportional to the mutation probability; after the repeatability reaches the threshold, the population screening module will add new random scenarios.

[0031] Furthermore, the specific method of step six is ​​as follows:

[0032] The repetition degree is calculated again. If the repetition degree within the population reaches the set threshold, that is, the proportion of repeated individuals in the whole population is greater than 50%, the repeated individuals exceeding the threshold are deleted and an equal number of random individuals are introduced to replace them so that the overall repetition degree is less than or equal to 50%.

[0033] Furthermore, the specific method of step seven is as follows:

[0034] Set termination conditions, 50 iterations, or no new key scenarios are found for three consecutive rounds. If the conditions are met, interrupt the test process in time.

[0035] The beneficial effects of the present invention are:

[0036] 1) The present invention improves operators such as selection, crossover, and mutation of traditional genetic methods, and newly adds a duplication degree screening module and an inferior individual screening module. Compared with traditional algorithms, it has the ability to quickly find an optimal solution at the initial stage of search and continue to explore even after finding some optimal solutions, so as to accelerate the search for a large number of dangerous boundary scenarios beneficial to the iterative optimization of the autonomous driving algorithm and accelerate the test process of the autonomous driving algorithm under test in logical scenarios;

[0037] 2) The present invention can meet the requirements of the autonomous driving test field for accelerating tests and scenario generation, can more effectively guide the test process, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the embodiments. It should be understood that the following drawings only show some embodiments of the present invention, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0039] Figure 1 is a schematic diagram of the process of the present invention;

[0040] Figure 2 is a schematic diagram of the process of the improved genetic algorithm;

[0041] Figure 3 is a schematic diagram of the scenario description in Step 1;

[0042] Figure 4 is a schematic diagram of the parallel simulation architecture. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present invention.

[0044] Refer to Figure 1 and Figure 2 , an autonomous driving acceleration test method based on an improved genetic algorithm, comprising the following steps:

[0045] Step 1: Determine the types, ranges, and distribution of the key scenario parameters of the system under test as the parameter space optimized by the improved genetic algorithm; and determine the fitness function to guide the evolution process of the genetic algorithm;

[0046] The specific method is as follows:

[0047] 11) The scenario-based autonomous driving test generates specific test cases according to the steps of functional scenario - logical scenario - specific scenario, so as to determine the types, ranges, and distributions of key scenario parameters of the system under test.

[0048] The premise of genetic algorithm encoding is to understand the input values to be optimized, and the key scenario elements are important inputs for scenario-based autonomous driving tests. It is necessary to combine the analysis of natural driving datasets to determine the distributions of corresponding functional scenarios, key scenario element types, and value ranges to meet the requirements of the authenticity and coverage of the generated scenarios. Taking the AEB system test as an example, the key scenario elements in the following-following functional scenario are the host vehicle speed, the leading vehicle speed, the leading vehicle's braking deceleration, and the distance between the two vehicles. Combining the sensor detection accuracy, their value ranges and discrete step sizes are determined. The key scenario elements are as Figure 3 shown.

[0049] 12) Combining the research of existing scenarios, the invention divides the autonomous driving simulation scenarios into the following four types: ① extremely dangerous, a dangerous scenario where a collision cannot be avoided; ② the collision edge, a dangerous boundary scenario that can be optimized and improved; ③ ensuring no collision, a general safety scenario with little value for algorithm improvement; ④ generating an unreasonable scenario that cannot be simulated, where the movement does not conform to physical laws or the simulation result is absolutely safe. Four stages of fitness functions are determined according to the characteristics of the dangerous boundary scenario, dangerous scenario, general safety scenario, and unreasonable scenario. Among them, the fitness functions of the dangerous boundary scenario, dangerous scenario, general safety scenario, and unreasonable scenario decrease in turn. The following is an illustration with examples:

[0050] The fitness function is the key to guiding the iterative evolution of the genetic algorithm and needs to be determined in combination with the specific scenario characteristics. Taking the AEB system test as an example, the time to collision (TTC) is selected as the key scenario index. TTC in the range of 0.6 - 1 s represents the dangerous boundary scenario, TTC in the range of 0 - 0.6 s represents the dangerous scenario, TTC greater than 1 s represents the general safety scenario, and TTC being negative represents the unreasonable scenario. Four stages of fitness functions are determined according to the characteristics of the four search scenarios, and different numerical settings are used to guide the scenarios to converge to the dangerous boundary in different stages. The TTC calculation formula and the fitness function formula are as follows:

[0051] TTC = s / (v ego - v front )

[0052]

[0053] In the formula, v ego is the host vehicle speed; v front is the leading vehicle speed; s is the initial relative position between the host vehicle and the leading vehicle; TTC is the calculated time to collision; ffit is the calculated fitness value, which is used to guide the search process of the genetic algorithm.

[0054] Step 2: Establish the sorting logic of each scenario element, perform floating-point encoding, and generate the initial population;

[0055] The specific method is as follows:

[0056] 21) According to the key scenario parameter types, ranges, and distributions of the system under test obtained in Step 1, establish the arrangement order of each element for encoding, which is convenient for decoding in this order later. The initial population is the key to the execution of the genetic algorithm. Since binary encoding has problems such as expanding the search space and discontinuous search, floating-point encoding is adopted to directly sort the key scenario parameters. For example, use the main vehicle speed, the speed of the vehicle in front, the braking acceleration of the vehicle in front, and the distance between the two vehicles when the vehicle in front brakes as the sequence.

[0057] 22) Determine the population size of the genetic process, and generate the initial population encoding by means of random values.

[0058] For example, each scenario element can be randomly sampled 30 times in the scenario space to generate an initial population containing 30 individuals.

[0059] Step 3: Decode the individual to generate the phenotype, import it into the simulation system, and introduce a parallel acceleration module to perform fast simulation of the individuals within the population;

[0060] The specific method is as follows:

[0061] 31) Based on the initial population encoding, decode the individual to generate specific scenario parameters, import them into the simulation system, and perform automated testing;

[0062] 32) Introduce a parallel acceleration module, use multi-device cloud data synchronization simulation, and save the simulation scenario parameters and results to the scenario library.

[0063] Since there is no influence between individuals within a single population of the genetic algorithm, in order to further improve the simulation efficiency, a multi-device parallel acceleration simulation scheme is designed, and the simulation results are transmitted to the genetic algorithm execution host in real time through the network. The simulation test process and the parallel acceleration architecture are as Figure 4 shown. A total of 3 computers are set. If the population is large, the number of slave machines can be considered increased.

[0064] Build a basic scenario in the PreScan / MATLAB simulation platform of three computers, set up an automated test method, calculate the real-time TTC value using sensor signals, and take the minimum TTC during one simulation process as the scenario selection index. The host first decodes the 30 individuals obtained by iterating the improved genetic algorithm to obtain the input values of various scenario parameters required for a single scenario, and compares them with the already simulated database in the cloud. If it has been simulated, there is no need to simulate, and the TTC operation result can be directly obtained; the un-simulated scenario data is evenly divided into three parts as much as possible, and two of them are stored in the task assignment database in the cloud server, and one part of the data is simulated by the host. The two slave computers obtain their respective tasks in the cloud server and perform automated simulations. During the automated simulation process of the three computers, the already simulated data and the obtained dangerous boundary scenario data are uploaded to the already simulated database and the excellent individual database in the cloud respectively. After the simulation is over, the host will obtain the simulation results of the current population to perform subsequent genetic operations.

[0065] Step 4: Perform encoding and screen out inferior individuals in the population according to the simulation results in Step 3;

[0066] Determine the superior and inferior characteristics according to the running results of the population individuals, and delete the inferior individuals, that is, the individuals with too low fitness values; to avoid the influence of inferior individuals on the population evolution process and interfere with subsequent genetic operator operations, design an inferior individual screening module, and define the scenarios with too low fitness (below 0.05) as inferior individuals, and directly eliminate them from the scenario library to avoid the search operator searching for this value during subsequent iterations, which increases the local optimization ability and fast optimization ability of the algorithm.

[0067] Step 5: Use the improved elitist retention selection operator, heuristic crossover, and non-uniform mutation operator to perform chromosome crossover and mutation operations to generate offspring;

[0068] To further improve the search speed of dangerous scenarios, improve the design of the basic genetic operator.

[0069] The specific method is as follows:

[0070] 51) At the selection operator, use the roulette wheel and elitist retention selection method. For individuals, use the roulette wheel selection method proportional to the fitness, and retain the current optimal individual at a ratio of 5%, continuously guiding the population evolution; the selection probability formula for the xi-th individual in the roulette wheel selection method is as follows:

[0071]

[0072] Wherein, f(xi) is the fitness value of the xi-th individual; N is the number of individuals in this population after removing the elite individuals; according to this formula, individuals can still be selected according to the fitness value in addition to elite retention, reflecting the heuristic nature of the genetic algorithm.

[0073] 52) At the crossover operator, a heuristic crossover method is adopted to perform heuristic transformation of the offspring individuals according to the fitness values of the parents; randomly select two parents in the population and compare them, and move the parameters of the inferior individual in the direction of the superior individual, so as to reflect the heuristic nature at the crossover level and accelerate the convergence of the population individuals to the optimal value. The principle is shown in the following formula:

[0074] x′ low =x low +k cross ×(x high -x low )

[0075] x′ high =x high

[0076] Wherein, x low is the element value of the individual with the lower fitness among the two values of the parent; x high is the element value of the individual with the higher fitness; k cross is the crossover heuristic coefficient; x′ low and x′ hight are the element values of the offspring generated by the heuristic crossover operator.

[0077] 53) At the mutation operator, a non-uniform mutation method is adopted, and the population repetition degree is used as the mutation probability selection index; the repetition degree is determined according to the distribution of the population individual parameters, and the standard is that both variables are within a certain discrete step length interval, that is, they are considered repeated. The mutation probability of the genetic algorithm is inversely proportional to the population repetition degree;. The principle is shown in the following formula:

[0078]

[0079] Wherein, x is the value of a certain element in the parent; x’ is the value of this element in the offspring, x max is the maximum value of the value space of this element; x min is the minimum value of the value space of this element; r is a random number between 0 and 1; b is a non-uniform coefficient; t and T are the current repetition degree and the repetition degree threshold; the mutation effect is related to the repetition degree, and the greater the repetition degree, the smaller the change value, so as to dynamically adjust the search step size and promote local optimization exploration.

[0080] Define the population redundancy as the number of individuals with the same values for all scenario elements. First, construct a judgment matrix for the population in the genetic algorithm, set the equal individuals to 1 and the unequal individuals to 0. Taking a matrix example with 7 individuals, based on the characteristics of the comparison matrix (the equality between elements is transitive), design the following algorithm: Sum the elements of each row in turn. If it is greater than or equal to 1, return the column numbers of all elements that are 1, keep the column of the first element that is 1, and clear the columns where the subsequent elements that are 1 are located. As shown in the following matrix, finally, it is determined that individuals No. 1, No. 2, No. 4, and No. 5 are equal, and individuals No. 3, No. 6, and No. 7 are equal. The population redundancies are 4 and 3 respectively.

[0081]

[0082] Step 6: Recalculate the population redundancy based on the individual information within the population and perform redundancy screening;

[0083] Recalculate the redundancy again. If the redundancy within the population reaches the set threshold (the proportion of duplicate individuals is greater than 50%), then delete the duplicate individuals exceeding the threshold and introduce an equal number of random individuals to replace them, so that the overall redundancy is less than or equal to 50%.

[0084] The traditional genetic algorithm will fall into a local optimum after executing a certain number of generations. Most of the individuals within the population have the same values, and it is impossible to find enough optimal solutions through genetic operations. For this reason, the present invention introduces a redundancy screening module to screen the individuals with too high redundancy in the population. If the redundancy is greater than the threshold of 50% (which is 4 in this example), then re-explore the individuals outside the threshold and replace the original data in the population.

[0085] Step 7: Repeat Steps 3 - 6 and set a termination condition to interrupt the test process in a timely manner.

[0086] Set a termination condition, with the number of iterations being 50 times, or no new key scenarios are found in three consecutive rounds, and interrupt the test process in a timely manner after meeting the conditions.

[0087] Repeating the above Steps 3 to 6 is a complete iteration process. To ensure that the algorithm can stop iterating in a timely manner to save computational consumption, this method sets a termination condition as follows: When the number of iterations reaches the set threshold (50 times), the genetic process is terminated. Or each time an iteration is performed, obtain the number of newly added dangerous boundary scenarios from the cloud database. After a certain number of iterations, if no new dangerous boundary scenarios are found in three consecutive rounds of iterations, then the iteration is terminated.

[0088] Result evaluation method: To accurately evaluate the acceleration test performance of the designed algorithm, design the following evaluation indicators in combination with specific acceleration test requirements, algorithm improvement purposes, and improvement processes:

[0089] (1) Coverage effect: Evaluate the coverage of the evaluation algorithm for dangerous boundary scenarios.

[0090] Test the designed acceleration algorithm, standard genetic algorithm, and traditional traversal algorithm for the same logical scenario space, calculate the proportion of dangerous boundary scenarios found by the acceleration algorithm in this paper, and evaluate whether the algorithm can find enough dangerous boundary scenarios to achieve the test purpose.

[0091] (2) Search efficiency: Evaluate the acceleration effect of the algorithm.

[0092] Define the search efficiency η = number of excellent individuals / number of simulations. Similarly, test the designed acceleration algorithm, standard genetic algorithm, and traditional traversal algorithm, calculate the ratio of the found dangerous boundary scenarios to the number of simulation test executions, and evaluate the acceleration effect of the algorithm.

[0093] (3) Exploration ability: Evaluate the continuous search ability of the algorithm.

[0094] Define the average number of newly found excellent individuals in every 10 iterations as the exploration ability, that is, the exploration ability ω = number of excellent individuals / 10. Whether a sufficient number of new dangerous boundary scenarios are generated after multiple iterations reflects the application effect of the repetition filtering strategy and evaluates the continuous exploration ability of the algorithm.

[0095] In addition, it should be noted that, in the various specific technical features described in the above specific embodiments, without conflict, they can be combined in any suitable way. To avoid unnecessary repetition, the present invention will not separately describe various possible combination methods.

[0096] In addition, any combination can be made between various different embodiments of the present invention, as long as it does not violate the idea of the present invention, it should also be regarded as the content disclosed by the present invention.

Claims

1. An accelerated test method for autonomous driving based on an improved genetic algorithm, characterized in that, It includes the following steps: Step 1: Determine the types, ranges, and distributions of the key scenario parameters of the system to be tested, which serve as the parameter space for optimizing the improved genetic algorithm; and determine the fitness function to guide the evolution process of the genetic algorithm; Step 2: Determine the sorting logic of each scenario element, perform floating-point encoding, and generate the initial population; Step 3: Decode the individuals to generate phenotypes, import them into the simulation system, and introduce a parallel acceleration module to perform rapid simulation of the individuals within the population; Step 4: Perform encoding and screen out inferior individuals within the population according to the simulation results in Step 3; Step 5: Use the improved elitist retention selection operator, heuristic crossover, and non-uniform mutation operator to perform chromosome crossover and mutation operations to generate offspring; Step 6: Calculate the population redundancy according to the individual information within the population and perform redundancy screening; Step 7: Repeat Steps 3 - 6 and set termination conditions to interrupt the test process in a timely manner; The specific method of Step 4 is as follows: Determine the superior and inferior characteristics according to the operation results of the population individuals, and delete the inferior individuals, that is, the individuals with fitness values less than the threshold; The specific method of Step 5 is as follows: 51) At the selection operator, use the roulette wheel and elitist retention selection method. For individuals, use the roulette wheel selection method proportional to the fitness, and retain the current optimal individual proportionally to continuously guide the population evolution; the selection probability formula for the xi-th individual in the roulette wheel selection method is as follows: In the formula, f(xi) is the fitness value of the xi-th individual; N is the number of individuals in this population after removing the elite individuals; 52) At the crossover operator, use the heuristic crossover method to perform heuristic transformation of the offspring individuals according to the fitness values of the parents; randomly select two parents in the population and compare them, and move the parameters of the inferior individual towards the direction of the better individual; 53) At the mutation operator, use the non-uniform mutation method. Determine the redundancy according to the distribution of the population individual parameters. The criterion is that if both variables are within a certain discrete step size interval, it is considered a repeat; use the population redundancy as the mutation probability selection index; the population redundancy is inversely proportional to the mutation probability; after the redundancy reaches the threshold, the population screening module will supplement new random scenarios; The specific method of Step 6 is as follows: Calculate the redundancy again. If the redundancy within the population reaches the set threshold, that is, the proportion of repeated individuals in the whole is greater than 50%, then delete the repeated individuals exceeding the threshold and introduce an equal number of random individuals to replace them, so that the overall redundancy is less than or equal to 50%.

2. The automatic driving acceleration test method based on an improved genetic algorithm according to claim 1, wherein The specific method of Step 1 is as follows: 11) Generate specific test cases for scenario-based autonomous driving tests according to the steps of functional scenario - logical scenario - specific scenario, so as to determine the types, ranges, and distributions of the key scenario parameters of the system to be tested; 12) Determine the four stages of the fitness function according to the characteristics of the dangerous boundary scenario, dangerous scenario, general safety scenario, and unreasonable scenario. Among them, the fitness function stage value of the dangerous boundary scenario is the highest, the fitness function stage value of the dangerous scenario is moderate, the fitness function stage value of the general safety scenario is lower, and the fitness function stage value of the unreasonable scenario is the lowest.

3. The automatic driving acceleration test method based on an improved genetic algorithm according to claim 1, characterized in that The specific method of the second step is as follows: 21) According to the key scenario parameter types, ranges, and distribution obtained in the first step, determine the arrangement order of each element and use floating-point encoding; 22) Determine the population size of the genetic process and generate the initial population encoding by means of random values.

4. The automatic driving acceleration test method based on an improved genetic algorithm according to claim 1, wherein, The specific method of the third step is as follows: 31) Based on the initial population encoding, decode the individuals to generate specific scenario parameters, import them into the simulation system, and perform automated testing; 32) Introduce a parallel acceleration module, use multi-device cloud data synchronous simulation, and save the simulation scenario parameters and results to the scenario library.

5. The automatic driving acceleration test method based on an improved genetic algorithm according to claim 1, characterized in that The specific method of the seventh step is as follows: Set termination conditions, with 50 iterations or no new key scenarios found in three consecutive rounds. Interrupt the test process in a timely manner after meeting the conditions.

Images