Multi-Platform Linkage Testing Method for Autonomous Driving Simulation Based on Offline Task Allocation

Abstract

The present invention belongs to the technical field of autonomous driving testing, and specifically relates to a multi-platform linkage testing method for autonomous driving simulation based on offline task allocation. It includes: First, mathematically express the offline task allocation of multi-platform linkage testing, and clarify the definitions of key elements, including the definition of test duration elements and the definition of test load elements; Second, establish test task allocation constraints, including test accuracy constraints, test priority constraints, and test load constraints; Third, use the particle swarm algorithm to solve the task allocation results, so as to obtain the specific tasks of different test platforms. The present invention realizes the linkage testing of multiple simulation test platforms through test task allocation. This method can solve the linkage testing method when there are multiple test platforms in the autonomous driving system by setting test process allocation constraints and solving; since the test platform accuracy is considered simultaneously during the task allocation process, and the testing capabilities of multiple platforms can be utilized, the testing efficiency can be improved while ensuring the testing accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving testing, and specifically relates to a multi-platform linkage testing method for autonomous driving simulation based on offline task allocation. Background Art

[0002] With the continuous maturity of autonomous driving vehicle technology, its mass production and implementation require a scientific and perfect test evaluation system as a support. Currently, scenario-based simulation testing has become an important support for solving the performance verification problems of autonomous driving vehicles. However, the simulation testing of autonomous driving vehicles involves different types of test platforms. Considering that there are numerous scenario element dimensions and it is extremely easy to generate the problem of dimensional explosion, how to establish a multi-simulation platform linkage testing method by coordinating the advantages of simulation testing platforms is an important means to improve testing efficiency and ensure testing accuracy. Although the research on multi-platform linkage testing has begun, the test task allocation mostly focuses on the simple allocation of scenario elements or directly integrating multiple platforms. The former cannot make full use of the accuracy advantages of each test platform, while the latter needs to test all test items and is prone to the problem of dimensional explosion. Therefore, analyzing the technical advantages of each simulation testing platform and establishing a multi-platform linkage testing method have become important contents of autonomous driving vehicle simulation testing. Summary of the Invention

[0003] To solve the above technical problems, the present invention provides a multi-platform linkage testing method for autonomous driving simulation based on offline task allocation, which realizes the linkage testing of multiple simulation testing platforms through test task allocation. This method can solve the linkage testing method when there are multiple test platforms in the autonomous driving system by setting test process allocation constraints and solving them. Since the test platform accuracy is considered during the task allocation process and the multi-platform testing capabilities can be utilized, the testing efficiency can be improved while ensuring the testing accuracy.

[0004] The technical solution of the present invention is described in conjunction with the accompanying drawings as follows:

[0005] A multi-platform linkage testing method for autonomous driving simulation based on offline task allocation includes the following steps:

[0006] Step 1: Mathematically express the offline task allocation of multi-platform linkage testing, and clarify the definition of key elements, including the definition of test duration elements and the definition of test load elements;

[0007] Step 2: Establish test task allocation constraints, including test accuracy constraints, test priority constraints, and test load constraints;

[0008] Step 3: Use the particle swarm algorithm to solve the task allocation results to obtain the specific tasks of different test platforms.

[0009] Furthermore, the specific method of Step 1 is as follows:

[0010] 11) Describe the multi - platform test task allocation for autonomous vehicles;

[0011] An autonomous vehicle has m specific scenario parameter combinations to be tested U , expressed as { U 1, U 2, …, U m}. For multiple simulation test platforms, the test task allocation process is to allocate these m test tasks to the simulation test platforms that need to be linked for testing P , expressed as { P s , P c , P m , ……}; For each test task, it includes a specific scenario parameter combination U i , an allocated simulation test platform P j , and the corresponding test time consumption t ij ; The sum of test tasks is m or less than m . On this basis, consider the time for different simulation test platforms to execute test tasks and establish a test load estimate value for the test tasks;

[0012] 12) Define the test task allocation matrix;

[0013] The defined test task allocation matrix T is shown in formula (1) as follows:

[0014] (1)

[0015] In the formula, t s , t c , t m are the times consumed by different types of simulation test platforms to execute a test task respectively, and the subscripts 1, 2, 3, …, q are the test task numbers executed by different test platforms; q is the number of test tasks of the test platform with the largest allocation quantity; When the test platform i has no actual test task occupancy in the j th test process in the matrix, the corresponding element in the matrix ti,j = 0;

[0016] Sample a logical scenario described by a parameter space to obtain a large number of specific test processes for testing. The final test time consumption is as shown in formula (2):

[0017] (2)

[0018] Thus, establish the final objective function for the entire test process:

[0019] (3)

[0020] 13) Define the test duration;

[0021] The test duration constraint refers to the time required to test a single scenario on different test platforms, that is t s , t c , t m , …… ; For different simulation test platforms, when performing the same test task, there is a certain deviation in the test time consumption. When the running duration of the test task is set to t set seconds, during the automated test, the digital simulation test platform and the hardware-in-the-loop test platform will cause the actual test duration to be longer than the set scenario duration due to scenario invocation and parameter reset. The duration delays caused by the platform automation performance of different platforms are respectively t s_y , t c_y , t m_y , …… , and the set test task time consumption of different simulation platforms is as shown in formula (4):

[0022] (4)

[0023] In the formula, is the expected time consumed for the i-th simulation test platform to execute a test task; i is the duration defined for the test task when the i-th simulation test platform executes the test task; is the delay in the test process caused by platform performance limitations; is the delay in the test process caused by platform performance limitations;

[0024] 14) Define the test load;

[0025] The test load is the probability that the task will be actually tested by the platform; the test load λIt is divided into two cases for consideration: continuous type parameters and enumeration type parameters; the digital representation is as follows:

[0026] (5)

[0027] In the formula, The test load defined for continuous type parameters; The test load defined for discrete type parameters.

[0028] Furthermore, the specific method of the said step 14) is as follows:

[0029] a. Define the load of continuous type parameters;

[0030] Define the i test load coefficient of the λ c_i th combination of continuous parameters as:

[0031] (6)

[0032] (7)

[0033] In the formula, δ is the adjustment coefficient that continuously shrinks with the progress of the test iteration; n c is the dimension number of continuous type scenario parameters; d i_min is the i th shortest normalized distance between the combination of continuous parameters and the combination of continuous parameters whose test result prediction is opposite; q is the maximum value of the single-platform test task; k t is the number of tests that have been carried out on this platform currently;

[0034] d i_min is calculated by the method of normalizing the maximum and minimum values, as shown below:

[0035] (8)

[0036] In the formula, p j is the value of the calculated combination of continuous parameters in the j dimension; p min_j is the value of the point whose test result is opposite to the calculated combination of parameters and is closest in the j dimension; and are the maximum and minimum values of the logical scenario parameter space in the j dimension;

[0037] b. Define an enumeration class parameter test load;

[0038] Define the i class of enumeration class parameter test loads λ d_i The calculation method is as follows:

[0039] (9)

[0040] In the formula, w i is the difficulty coefficient of the i class of enumeration class parameters; w near is the difficulty coefficient of the parameter combination that is closest to the test difficulty of the difficulty coefficient and the i class of enumeration class parameters and has been tested; w worst is the maximum difficulty coefficient of the enumeration class parameter combination; if w i is equal to w near and w near the test result of is safe, define λ d_i at this time as 0.5.

[0041] Furthermore, the specific method of the second step is as follows:

[0042] 21) Define test precision constraints;

[0043] Precision threshold constraints mean that a test task can only be assigned to a test platform that meets the test precision requirements. If there is a test task for which all test platforms do not meet the test precision requirements, then the test task does not participate in the platform task allocation process, as shown in formula (10):

[0044] (10)

[0045] In the formula, is the precision result of the th test platform for the th specific test scenario; is the probability that the th specific test scenario is assigned to the th test platform;

[0046] 22) Define test priority constraints;

[0047] The test priority constraint means that when the test platform that meets the accuracy requirements of the test task includes different test platforms at the same time, the test task is preferentially assigned to the test platform with higher test accuracy; this part of the constraint is determined by defining the duration shortening coefficient; after increasing the duration shortening coefficient, the multi-platform test task allocation process matrix is shown in formula (11):

[0048] (11)

[0049] Among them,

[0050] (12)

[0051] In the formula, is the duration shortening coefficient of the th test platform for the specific scenario parameters;

[0052] 23) Define the test load balancing constraint;

[0053] The test load balancing constraint is expressed as:

[0054] (13)

[0055] In the formula, is the test load; is the number of test tasks with a test load of assigned to the test platform ; is the average value of all test tasks with a test load of assigned to all platforms, and the subscript represents different simulation test platforms.

[0056] Furthermore, the specific method of step three is as follows:

[0057] 31) The ultimate goal of test task allocation is to find the optimal solution of the ultimate objective function while satisfying multiple constraint conditions, and the ultimate goal is equivalent to a multi-objective optimization process;

[0058] 32) Use the improved particle swarm optimization algorithm to accelerate the solution of the multi-objective optimization process; when using the particle swarm optimization algorithm to solve the task allocation, the optimal solution is searched through the movement of particles in the solution space; the update of speed and position in the particle swarm optimization algorithm is shown in formulas (14)-(15), and through continuous iteration of speed and position, the optimal solution is finally obtained;

[0059] (14)

[0060] (15)

[0061] In the formula, In the formula, is the position of the particle in the -th iteration in dimensions; is the position of the particle in the -th iteration in dimensions; is the velocity of the particle in the -th iteration in dimensions; is the velocity of the particle in the -th iteration in dimensions; is the inertia weight; and are the learning factors; and are random numbers with values between 0 and 1; is the optimal position of the -th particle in dimensions; is the optimal position of all particles in the -th iteration in dimensions.

[0062] The beneficial effects of the present invention are as follows:

[0063] The present invention realizes the linkage test of multiple simulation test platforms through test task allocation; this method can solve the linkage test method for an autonomous driving system with multiple test platforms by setting test process allocation constraints and solving. Since the test platform accuracy is considered during the task allocation process and the test capabilities of multiple platforms can be utilized, the test efficiency can be improved while ensuring the test accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] 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 thus 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.

[0065] Figure 1 is the flow diagram of the present invention;

[0066] Figure 2 is the schematic diagram of the convergence result during the test task allocation for 300 test tasks;

[0067] Figure 3Schematic diagram of the convergence result when allocating test tasks for 500 test tasks;

[0068] Figure 4 Schematic diagram of the convergence result when allocating test tasks for 800 test tasks;

[0069] Figure 5 Schematic diagram of the convergence result when allocating test tasks for 1000 test tasks. Specific implementation manners

[0070] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It can be understood that the specific embodiments described herein are only used to explain the present invention, rather than limiting the present invention. Additionally, it should be noted that for the convenience of description, only parts related to the present invention rather than all structures are shown in the drawings.

[0071] Embodiment 1: Refer to Figure 1 , this embodiment provides a multi-simulation platform linkage test method for autonomous driving vehicle simulation tests, including the following steps:

[0072] Step 1. Mathematically express the offline task allocation of the multi-platform linkage test, and clarify the definition of key elements, including the definition of test duration elements and test load elements, as follows:

[0073] 11) Describe the test task allocation of autonomous driving vehicle multi-platforms;

[0074] For the test task allocation of autonomous driving vehicle multi-platforms, it can be described as the following process: An autonomous driving vehicle has m specific scenario parameter combinations to be tested in each logical scenario U , expressed as { U 1, U 2,..., U m}, for multiple simulation test platforms, the test task allocation process is to allocate these m test tasks to the simulation test platforms that need to be linked for testing P , expressed as { P s , P c , P m , ……}; for each test task, it includes a specific scenario parameter combination U i , an allocated simulation test platform P j , and the corresponding test duration t ij ; the sum of the test tasks ism or less than m , on this basis, consider the time taken by different simulation test platforms to execute test tasks, and establish an estimated test load value for the test tasks;

[0075] 12) Define the test task allocation matrix;

[0076] The defined test task allocation matrix T is shown in formula (1):

[0077] (1)

[0078] In the formula, t s , t c , t m are the times consumed by different types of simulation test platforms to execute a test task respectively, and the subscripts 1, 2, 3,..., q are the test task numbers executed by different test platforms; q is the number of test tasks of the test platform with the largest allocation quantity; when the test platform i in the j th test process in the matrix has no actual test task occupied, the corresponding element t i,j = 0 (there may be platforms without test tasks after matrix 1, and at this time its test time consumption is 0);

[0079] Sample a logical scenario described by a parameter space to obtain a large number of test processes for specific tests. The final test time consumption is shown in formula (2):

[0080] (2)

[0081] Thus, establish the final objective function for the entire test process:

[0082] (3)

[0083] 13) Define the test duration;

[0084] The test duration constraint refers to the time required for different test platforms to test a single scenario, that is, t s , t c , t m , ……; For different simulation test platforms, there are certain deviations in the test time consumption when performing the same test task. When the running duration of the test task (the duration from the set start to the end of a single specific scenario) is set to t set seconds, during the automated test, the digital simulation test platform and the hardware-in-the-loop test platform will cause the actual test duration to be longer than the set scenario duration due to scenario calls, parameter resets, etc. The duration delays caused by the platform automation performance of different platforms are respectively t s_y 、 t c_y 、 t m_y , …… , the test time consumption of different simulation platform test tasks is set as shown in formula (4):

[0085] (4)

[0086] In the formula, is the time expected to be consumed by the i th simulation test platform to execute a test task; is the duration defined by the test task when the th simulation test platform executes the test task;

[0087] 14) Define the test load;

[0088] For different test tasks, there are certain differences in their test loads. The test load can be understood as the probability that the task will be actually tested by the platform. Since the current simulation test platforms mostly adopt accelerated test methods, not all test tasks will be tested. These tasks that will not be tested can be understood as having a lower test load. To avoid task allocation errors caused by experience, the test load of different specific test scenarios is not considered when initially allocating test tasks. When reallocating test tasks with certain test results in the subsequent stage, this part of the content needs to be considered. It should be noted that when a specific test is performed on a single test platform, the test scenario elements with a high test load are not necessarily tested first. This part of the content is only to ensure that the time consumption of each platform is approximately equal at the end of the final test. The test load λIt is divided into two consideration cases: continuous class parameters and enumeration class parameters. For calculating the specific scenario parameters of a group to be tested, first judge the test value of the continuous scenario parameter part in this parameter. If the continuous scenario parameter part in this group of parameters has not been tested, directly use the test load of this part of the continuous scenario parameter to represent the test load of the entire specific scenario parameter combination. If the continuous scenario parameter part in this group of parameters has been tested, then consider using the enumeration class parameters to calculate the test load of this group of specific scenario parameters. The above analysis process is transformed into digital representation as follows:

[0089] (5)

[0090] In the formula, The test load defined for continuous class parameters; The test load defined for discrete class parameters.

[0091] Specifically, a. Define the continuous class parameter load;

[0092] Define the i th test load coefficient of the continuous parameter combination λ c_i as:

[0093] (6)

[0094] (7)

[0095] In the formula, δ Is the adjustment coefficient that continuously shrinks with the test iteration; n c Is the dimension number of the continuous class scenario parameters; d i_min For the i th continuous parameter combination, the shortest normalized distance between it and the continuous parameter combination with the opposite predicted test result; q Is the maximum value of the single-platform test task; k t Is the number of tests that have been carried out on this platform currently; The reason for taking the constant value 3 in formula (7) is that the existing accelerated test algorithms generally need to consume about 30% of the computing power to determine the performance boundary of the system under test. Therefore, one-third of the total number of tests is used as the basis for parameter adjustment in this formula;

[0096] d i_min The calculation method of adopts the maximum-minimum normalization method, as shown below:

[0097] (8)

[0098] In the formula, pj The value of the calculated continuous parameter combination in j dimension; p min_j The value of the point with the opposite test result closest to the calculated parameter combination in j dimension; and are the maximum and minimum values of the logical scenario parameter space in j dimension; It should be noted that p min_j is not the point closest to the measured point in j dimension with the closest test result, but the point with the shortest comprehensive distance to the opposite test result.

[0099] b. Define the enumerated class parameter test load;

[0100] For the enumerated class parameter test load, the difficulty of the untested parameter combination is mainly considered, which mainly includes two situations: If the currently tested continuous parameter combination is a dangerous situation, the higher the difficulty of the untested enumerated class parameter, the lower the test load, because if the tested algorithm has failed in a simple scenario, it is very likely to have an accident in a complex scenario; If the currently tested continuous parameter combination is a safe situation, the higher the difficulty of the untested enumerated class parameter, the higher the test load, because although the tested algorithm has not had an accident in a simple scenario, it may have an accident in a complex enumerated class parameter combination. According to the above analysis process, the i type of enumerated class parameter test load λ d_i is calculated as follows:

[0101] (9)

[0102] In the formula, w i is the i type of enumerated class parameter difficulty coefficient; w near is the difficulty coefficient closest to the test difficulty of the i type of enumerated class parameter among the tested parameter combinations; w worst is the maximum difficulty coefficient of the enumerated class parameter combination; If w i is equal to w near and w near the test result of is safe, define λ d_i at this time as 0.5.

[0103] Step 2. Establish test task assignment constraints, including test precision constraints, test priority constraints, and test load constraints, as follows:

[0104] 21) Define test precision constraints;

[0105] The precision threshold constraint means that a test task can only be assigned to a test platform that meets the required test precision. If there are test tasks for which all test platforms do not meet the test precision requirements, then the test tasks do not participate in the platform task assignment process, as shown in formula (10):

[0106] (10)

[0107] In the formula, F r_i_j is the precision result of the i th test platform for the j th specific test scenario; P i_j is the probability that the j th specific test scenario is assigned to the i th test platform;

[0108] 22) Define test priority constraints;

[0109] The test priority constraint means that when test platforms that meet the precision requirements of a test task include different test platforms, the test task is preferentially assigned to the test platform with higher test precision; this part of the constraint is determined by defining a duration shortening coefficient; after adding the duration shortening coefficient, the multi-platform test task assignment process matrix is as shown in formula (11):

[0110] (11)

[0111] Among them,

[0112] (12)

[0113] In the formula, is the duration shortening coefficient of the th test platform for the specific scenario parameters;

[0114] 23) Define test load balancing constraints;

[0115] For test task assignment, in order to avoid allocating all test tasks with high test loads to a single test platform under the major objective of obtaining the optimal test time, it is necessary to add test load balancing constraints to ensure the balance of different test platforms during the test load process. The test load balancing constraint can be expressed as:

[0116] (13)

[0117] In the formula, is the test load; is the number of test tasks with test load assigned to the test platform ; is the average value of the tasks with all test loads assigned to all platforms, and the subscript represents different simulation test platforms.

[0118] Step 3: Use the particle swarm algorithm to solve the task assignment result, so as to obtain the specific tasks of different test platforms. The specific method is as follows:

[0119] 31) The ultimate goal of test task assignment is to find the optimal solution of formula (3) while satisfying multiple constraints. The above process can be equivalent to a multi-objective optimization process;

[0120] 32) Use the improved particle swarm algorithm to accelerate the solution of this process. When using the particle swarm algorithm to solve task assignment, each particle is equivalent to completing a test task assignment, and each dimension in the particle can be considered as the assignment result of the corresponding task. By adjusting different assignment strategies, the optimal test time is achieved. The particle swarm algorithm is an optimization algorithm based on swarm intelligence. In this algorithm, each solution in the solution space is called a "particle", and the optimal solution is searched through the movement of the particles in the solution space. In this process, the most important changes include the position x ij and velocity v ij . Among them, the position represents the current solution of the particle and represents the assignment result of a group of test tasks, and the velocity represents the moving rate of the particle in the solution space and represents the adjustment status of the test tasks. The update of velocity and position in the traditional particle swarm algorithm is shown in formulas (14)-(15). Through the continuous iteration of velocity and position, the optimal solution is finally obtained. (14)

[0121] (15)

[0122] In the formula, is the position of the particle in the th iteration in the th dimension; is the position of the particle in the th iteration in the th dimension; is the velocity of the particle in the th iteration in the th dimension; is the velocity of the particles in the in the -dimensional space; is the inertia weight; and are the learning factors; and are random numbers, taking values between 0 and 1; is the optimal position of the particles in the in the -dimensional space; is the optimal position of all particles in the -dimensional space in the

[0123] Example 2: In this example, 4 groups of specific test tasks with 300, 500, 800, and 1000 tasks respectively and the corresponding platform test accuracy results are randomly generated. The proposed test task allocation method is used to allocate test tasks among multiple simulation platforms, and the total time consumed in the corresponding overall test process is calculated. All test task loads and the platform types that meet the accuracy requirements are randomly generated, and the basic duration of each test task is set to 30s. The requirements of some generated test tasks and the corresponding platform simulation fidelity results are shown in Table 1.

[0124] Table 1 Test task and platform simulation fidelity data

[0125]

[0126] The best test duration results of each round of iteration of the particle swarm algorithm in the four groups of test task allocation processes are as Figures 2 - 5 shown. It can be seen from the Figures 2 - 5 results that when the algorithm is applied to test task allocation for different numbers of test tasks, it shows good convergence results and can effectively reduce the total task time consumed during the collaborative testing of different test platforms.

[0127] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-platform linkage testing method for autonomous driving simulation based on offline task allocation, characterized in that, It includes the following steps: Step 1: Mathematically express the offline task allocation of multi-platform linkage testing, and clarify the definition of key elements, including the definition of test duration elements and the definition of test load elements; Step 2: Establish test task allocation constraints, including test accuracy constraints, test priority constraints, and test load constraints; Step 3: Use the particle swarm algorithm to solve the task allocation results, so as to obtain the specific tasks of different test platforms; The specific method of the above Step 1 is as follows: 11) Describe the multi-platform test task allocation of autonomous vehicles; An autonomous vehicle has m specific scenario parameter combinations U to be measured in each logical scenario, expressed as {U1, U2, …, U m}, for multiple simulation test platforms, the test task allocation process is to allocate these m test tasks to the simulation test platforms P that need to be jointly tested, expressed as {P s , P c , P m , ……}; for each test task, it includes a specific scenario parameter combination U i , an allocated simulation test platform P j , and the corresponding test time consumption t ij ; the sum of test tasks is m or less than m. On this basis, consider the time for different simulation test platforms to execute test tasks and establish a test load estimate for the test tasks; 12) Define the test task allocation matrix; The defined test task allocation matrix T is shown in formula (1): where t s , t c , t m are respectively the time consumed for a different type of simulation test platform to execute a test task, and the subscripts 1, 2, 3, …, q are the test task numbers executed by different test platforms; q is the number of test tasks of the test platform with the largest allocation quantity; When there is no actual test task occupancy in the j-th test process of the test platform i in the matrix, the corresponding element t i,j = 0; Sample a logical scenario described by the parameter space to obtain a large number of specific test processes for testing, and the final test time consumption is shown in formula (2): Thus, establish the final objective function of the entire test process: F = min{T test} (3) 13) Define the test duration; The test duration constraint refers to the time required to test a single scenario on different test platforms, i.e., t s , t c , t m , ……; For different simulation test platforms, there is a certain deviation in the test time when performing the same test task. When the running duration of the test task is set to t set seconds, during the automated test, the digital simulation test platform and the hardware-in-the-loop test platform will cause the actual test duration to be longer than the set scenario duration due to scenario invocation and parameter reset. The duration delays caused by the platform automation performance of different platforms are t s_y , t c_y , t m_y , ……, The set test task durations of different simulation platforms are shown in formula (4): t i ′ = t i + t i_y (4) where t i ' is the time expected to be consumed for the i-th simulation test platform to execute a test task; t i is the duration defined for the test task when the test task is executed on the i-th simulation test platform; t i_y is the delay during the test process caused by platform performance limitations; 14) Define the test load; The test load is the probability that the task will be actually tested by the platform; the test load λ is considered in two cases: continuous class parameters and enumeration class parameters; the digital representation is as follows: where λ c is the test load defined for continuous class parameters; λ d is the test load defined for discrete class parameters.

2. The method for joint testing of multiple platforms in an autonomous driving simulation based on offline task allocation according to claim 1, wherein The specific method of the above Step 14) is as follows: a. Define the continuous class parameter load; Define the test load factor λ for the i-th consecutive parameter combination c_i as follows: where δ is an adjustment coefficient that continuously shrinks as the test iteration progresses; n c is the number of dimensions of the continuous class scenario parameters; d i_min is the shortest normalized distance between the i-th continuous parameter combination and the continuous parameter combination whose test result prediction is opposite; q is the maximum value of the single-platform test task; k t is the number of tests that have been conducted on this platform currently; d i_min The calculation method adopts the method of normalizing the maximum and minimum values, as shown below: where p j is the value of the calculated continuous parameter combination in the j-th dimension; p min_j is the value of the point with the opposite test result closest to the calculated parameter combination in the j-th dimension; l j_max and l j_min are the maximum and minimum values of the logical scenario parameter space in the j-th dimension; b. Define the enumeration class parameter test load; Define the test load λ of the i-th enumerated class parameter d_i The calculation method is as follows: where w i is the difficulty coefficient of the i-th enumerated parameter; w near is the difficulty coefficient of the tested parameter combination closest to the test difficulty of the i-th enumerated parameter; w worst is the maximum difficulty coefficient of the enumerated parameter combination; if w i is equal to w near and the test result of w near is safe, define λ d_i at this time as 0.

5.

3. A method for joint testing of multiple platforms in an autonomous driving simulation based on offline task allocation according to claim 1, characterized in that, The specific method of the above Step 2 is as follows: 21) Define the test accuracy constraint; The accuracy threshold constraint means that the test task can only be assigned to the test platform whose test accuracy meets the requirements. If there is a test task for which all test platforms do not meet the test accuracy requirements, the test task does not participate in the platform task allocation process, as shown in formula (10): if F r_i_j <precision threshold, P i_j = 0(10) where F r_i_j is the accuracy result of the i-th test platform for the j-th specific test scenario; P i_j is the probability assigned to the i-th test platform for the j-th specific test scenario; 22) Define the test priority constraint; The test priority constraint means that when the test platforms that meet the accuracy requirements of the test task include different test platforms at the same time, the test task is preferentially assigned to the test platform with higher test accuracy; this part of the constraint is determined by defining the duration shortening coefficient; after adding the duration shortening coefficient, the multi-platform test task allocation process matrix is shown in formula (11): Among them, where γ h_i is the duration shortening coefficient of the i-th test platform for specific scenario parameters; 23) Define the test load balancing constraint; The test load balancing constraint is expressed as: where λ k is the test load; n λk_i is the number of test tasks with test load λ k assigned to test platform i; is the average value of all tasks with test load λ k assigned to all platforms, and the subscript i represents different simulation test platforms.

4. The multi-platform linkage test method for autonomous driving simulation based on offline task allocation according to claim 1, wherein The specific method of the above Step 3 is as follows: 31) The ultimate goal of test task allocation is to find the optimal solution of the final objective function while meeting multiple constraint conditions, and equivalently transform the final objective into a multi-objective optimization process; 32) Use the improved particle swarm algorithm to accelerate the solution of the multi-objective optimization process; when using the particle swarm algorithm to solve the task allocation, the optimal solution is searched through the movement of particles in the solution space; the update of velocity and position in the particle swarm algorithm is shown in formulas (14)-(15), and through the continuous iteration of velocity and position, the optimal solution is finally obtained; Wherein, is the position of particle i in the j-th dimension at the t-th iteration; is the position of particle i in the j-th dimension at the (t + 1)-th iteration; is the velocity of particle i in the j-th dimension at the t-th iteration; is the velocity of particle i in the j-th dimension at the (t + 1)-th iteration; h is the inertia weight; o1 and o2 are learning factors; l1 and l2 are random numbers with values between 0 and 1; is the optimal position of particle i in the j-th dimension at the t-th iteration; is the optimal position of all particles in the j-th dimension at the t-th iteration.

Images