An iterative evolutionary evaluation method employing an automated and efficient test generator algorithm

Abstract

This invention discloses an iterative evolutionary evaluation method using an automated and efficient test generator algorithm. The method generates test cases based on this algorithm, with the following specific steps: First, based on the test metrics of a multi-agent system, all relevant test parameters are selected from the system environment model and behavior model. Second, the value range of each test parameter is defined. Then, the relationships between parameters are divided, and parameter constraints are confirmed to resolve value conflicts. Finally, based on the determined parameters, values, and constraints, parameter grouping is completed, and test cases are generated using the AETG algorithm.

Description

Technical Field

[0001] This invention relates to the field of simulation testing technology, and in particular to an iterative evolutionary evaluation method using an Automatic Efficient Test Generator (AETG) algorithm. Background Technology

[0002] Multi-agent systems are swarm systems composed of multiple autonomous intelligent agents. Their core objective is to break down large, complex systems into smaller, manageable subsystems that can communicate and interact with each other. In recent years, multi-agent systems have attracted great attention from scholars in related fields and have become an important research hotspot in control science and artificial intelligence.

[0003] Software testing is a crucial step in ensuring product quality. It not only drives business optimization and reduces later maintenance costs, but also improves system reliability and user experience. In fact, creating fully functional, stable, and error-free software requires continuous and comprehensive iterative optimization. Delivering a high-quality product is itself a process of continuous verification and validation; each stage of software testing directly impacts the overall performance and efficiency of the system and application. Therefore, standardized and reliable testing ensures that software applications meet various technical specifications and business requirements.

[0004] Common software testing methods include equivalence partitioning, scenario-based testing, boundary value analysis, decision tables, orthogonal permutation, and test outline methods. Among these, equivalence partitioning, boundary value analysis, and decision tables are all black-box testing methods, verifying only whether the software functionality meets requirements from the outside, and are mainly used in the software acceptance testing phase. White-box testing, also known as structural testing or logic-driven testing, focuses on code, path, and condition coverage, emphasizing the detection of the software's internal structure and runtime performance, verifying whether the program path executes as expected, rather than directly verifying functionality, and is mostly used in the software verification phase. Overall, traditional testing methods require a large number of test cases, making them difficult to directly adapt to the testing requirements of multi-agent systems.

[0005] Multi-agent systems require collaboration and competition to accomplish complex tasks that a single agent cannot achieve independently. This significantly increases the complexity of interactions between agents and demands real-time responses to environmental changes. Agent systems must be deeply integrated with their external environment; each agent views other agents as both collaborators and part of the environment, mutually incentivizing each other during collaboration to ensure the system's state continuously evolves dynamically. Therefore, testing multi-agent systems presents considerable challenges.

[0006] Currently, efficient testing methods for multi-agent systems are still lacking. Existing testing and evaluation methods for multi-agent systems focused on swarm intelligence mostly rely on simulation platforms, verifying system functionality by executing a complete simulation process. However, in real-world applications, multi-agent systems exhibit various types of uncertainty, leading to a significant increase in the execution cost and time overhead of test cases.

[0007] In testing multi-agent systems, a large number of test parameters are typically involved, and each parameter contains multiple discrete values. Directly performing full combination testing on all parameters would lead to a dramatic increase in the size of test cases, resulting in a combinatorial explosion problem, which in turn significantly increases the testing cost and difficulty of multi-agent systems. Summary of the Invention

[0008] In view of this, the purpose of this invention is to overcome the shortcomings of the prior art and propose an iterative evolutionary evaluation method using an automatic and efficient test generator algorithm. This method realizes the iterative evolutionary generation of test cases based on the automatic and efficient test generator algorithm, and the specific steps are as follows:

[0009] Step 1: Select test parameters;

[0010] Based on the test metrics of multi-agent systems, all test parameters associated with the test metrics are selected from the system environment model and behavior model. The selected test parameters and their values ​​are divided into four categories, and the value rules for each category are as follows:

[0011] Finite class parameters: The values ​​take the form of a continuous interval, constrained by natural conditions or task scenarios, with clearly defined upper and lower limits;

[0012] Percentage parameter: The value ranges from [0,1], representing the probability of a specific event occurring;

[0013] Natural class parameters: values ​​take the value of a continuous interval, but without explicit upper or lower limits;

[0014] Discrete parameters: take on a finite number of discrete values;

[0015] This step includes the following sub-steps:

[0016] Step 1.1: For the finite class parameters and percentage parameters, select several discrete typical values ​​within their value range as test values;

[0017] Step 1.2: Use the parameter iteration method to obtain the values ​​of the natural class parameters;

[0018] Step 1.3: Determine the values ​​of discrete parameters using an item-by-item enumeration method;

[0019] The parameter iteration method for obtaining the values ​​of natural class parameters is as follows:

[0020] Step 1.21: Select a typical value for this parameter and fix it as a constant, then test other parameters;

[0021] Step 1.22: After completing most of the tests, select the parameter combination with the best overall performance as the preferred group, and select the combination with the worst performance as the control group.

[0022] Step 1.23: Fix all parameters in the preferred group and the control group except for the target natural class parameter, adjust only the natural class parameter, and observe its effect on the tested system;

[0023] Step 1.24: Repeat step 1.23 to obtain the optimal test result corresponding to the natural class parameter;

[0024] Step 2: Determine the values ​​of the test parameters;

[0025] Clarify the specific value range and optional values ​​of all the test parameters selected in step 1;

[0026] Step 3: Determine the correlation between test parameters;

[0027] The relationships between parameters are determined through the following sub-steps:

[0028] Step 3.1: Based on the classification of test indicators, group the parameters that affect the same test indicator together;

[0029] Step 3.2: Support parameters to exist across groups, that is, the same parameter can be classified into multiple related groups;

[0030] Step 3.3: If the impact of parameters on different test indicators is unclear, group semantically related parameters into one group based on semantic relevance.

[0031] Step 4: Confirmation of parameter constraints and conflict resolution;

[0032] Step 4.1: Based on the parameters, their values, and their ranges, determine whether there are any value conflicts among the parameters;

[0033] Step 4.2: For conflicting parameters, adjust the constraints to eliminate value conflicts and ensure the effectiveness of parameter combinations;

[0034] Step 5: Test case generation based on an automated and efficient test generator algorithm;

[0035] Based on steps 1-4, the parameters are grouped, and test cases are generated using an automated and efficient test generator algorithm. The specific sub-steps are as follows:

[0036] Step 5.1: Select parameter X in the current parameter group, and its value Y must satisfy the following condition: Y has the highest frequency of occurrence in the uncovered parameter combinations;

[0037] Step 5.2: Denote the selected parameter X as X1, and randomly sort the remaining parameters to obtain the parameter sequence X1, X2, X3, ..., Xn;

[0038] Step 5.3: Determine the corresponding value sequence for the parameter sequence X1, X2, ..., Xn, where Ym is the value of parameter Xm. The specific process is as follows:

[0039] Step 5.31: Determine the next parameter X to be assigned a value. n+1 ;

[0040] Step 5.32: For X n+1 For each selectable value Y, calculate the set {(X)} n+1 The number of new parameter combinations that are not covered in { =Y, Xm=Ym), 1≤m≤n};

[0041] Step 5.33: Select the value Y that produces the most new parameter combinations as X. n+1 The final value;

[0042] Step 5.4: Repeat steps 5.1-5.3 until complete parameter values ​​for all test cases are generated.

[0043] The beneficial effects of this invention are as follows: Given limited testing time and costs, a process for generating test cases using an automated and efficient test generator algorithm is proposed for complex intelligent systems. Using the same test cases covers as many test scenarios as possible, making the evaluation results more comprehensive and reliable. Simultaneously, to fully utilize the information provided by each running test case, statistical methods are used to analyze the test data, discover hidden problems, and generate new test cases based on these problems for verification. This ensures the reliability of the evaluation even with a limited number of test cases. Attached Figure Description

[0044] Figure 1 This is a flowchart of the technical solution of the present invention. Detailed Implementation

[0045] The technical problem solved by this application;

[0046] 1. A framework for test case generation using an automated and efficient test generator algorithm for testing multi-agent systems.

[0047] 2. The goal is to address how to analyze historical test results of multi-agent systems and optimize test case generation based on this analysis.

[0048] The present invention will be further described below with reference to the accompanying drawings and specific embodiments. The process of the present invention is as follows: Figure 1 As shown:

[0049] Test case generation;

[0050] Using the environment model, behavioral model, and test metrics of the system under test as input, this method first generates test cases for the system under test based on the fundamental principles of combinatorial testing. Then, it performs statistical analysis on the test results and continuously adjusts the test case generation strategy based on the statistical analysis results, thereby iteratively generating new test cases. This approach aims to comprehensively evaluate the system under test within a limited test case cost.

[0051] Step 1. Select test parameters;

[0052] Based on the established test metrics and expert experience, select all test parameters related to the test metrics from the environmental and behavioral models of the system under test. Parameters unrelated to the test metrics should also be selected if there is a clear need to test them.

[0053] Step 2. Determine the parameter values;

[0054] The selected parameters are generally divided into the following four categories:

[0055] 1) Finite-class parameters – These parameters take values ​​within a continuous range and are constrained by natural conditions or task scenarios, with clearly defined upper and lower limits for the range. For example, initial position offset and formation spacing are both parameters that are constrained by natural conditions and task scenarios, and thus have clearly defined upper and lower limits.

[0056] 2) Percentage parameters—These parameters range from 0 to 1. They often represent the probability of a type of event occurring. For example, the perception rate and packet loss rate are both probabilities.

[0057] 3) Natural parameters – These parameters take values ​​within a continuous range, but the natural conditions or task environment impose relatively weak restrictions on them, with no clear upper or lower bounds. For example, initial quantities. Natural conditions do not restrict initial quantities, and system developers often find it difficult to determine the upper limit of the initial quantities the system can calculate.

[0058] 4) Discrete parameters – These parameters have a limited number of possible values. For example, communication mode and array configuration only have a few possible choices.

[0059] For finite-class parameters and percentage parameters, several dispersed typical values ​​are selected within the value range as test values ​​for the parameters. Specifically, the value range is evenly divided into several small intervals, and then a random value is selected for each interval. For natural-class parameters, a typical value for one parameter is first selected and used as a constant. Other parameters are then tested first. After most tests are completed, a parameter combination with good overall performance and a parameter combination with poor overall performance are selected as a control group. Other parameters within these two combinations are kept constant, and only the natural-class parameters in these two combinations are adjusted to observe their impact on the tested system, thus obtaining the test results for the natural-class parameters. The number of values ​​for discrete-class parameters is limited and can be listed individually.

[0060] Step 3. Determine the parameter relationships;

[0061] Based on the test metrics, parameters affecting a single test metric are grouped together. The same parameter may exist in multiple groups. For parameters whose impact on different test metrics cannot be determined, they are grouped according to their semantics, with semantically related parameters grouped together.

[0062] Step 4. Confirm parameter constraints;

[0063] Based on the parameters and their values ​​and ranges, confirm whether there are any conflicts in parameter values. If so, adjust the parameter constraints to eliminate the conflicts.

[0064] Step 5. Test case generation;

[0065] After determining the test parameters, their values, and the constraints between them, and grouping the parameters, test cases are generated using a classic automated and efficient test generator algorithm used in combinatorial testing. The specific process is as follows:

[0066] 1) Select parameter X and parameter value Y within the parameters, where Y appears the most frequently in the uncovered parameter combinations;

[0067] 2) After determining the parameter X, let X = X1, and then randomly sort the remaining parameters to obtain a sequence of parameters — X1, X2, X3, ...;

[0068] 3) Assume the sequence X1, X2, ..., X n The value of Y has been determined. m For X m The value of X is determined by following these steps. n+1 The value of X. n+1 For each value Y, calculate the set {(X)} n+1 =Y,X m =Y mThe number of new parameter combinations in the set {1 ≤ m ≤ n}. The value of Y with the most new parameter combinations is X. n+1 The value of .

[0069] 4) Repeat steps 1)-3) to generate all test cases;

[0070] Data analysis;

[0071] The test cases generated using Scheme 2.2.1 drive the system under test and collect operational data to conduct test data analysis. This step first requires specifying the expected impact of relevant parameters for each metric on that metric. The impact is categorized into three types: positive impact, no impact, and negative impact.

[0072] The impact of discrete parameters on test metrics needs to be discussed in groups, without providing specific expected impacts. Secondly, the expected range for each test metric needs to be given. Using the test metric as the dependent variable and all test parameter values ​​as independent variables, statistical analysis should be conducted using significance tests and linear regression analysis. Check whether the parameters and their values ​​used to generate test cases fall into the following two categories:

[0073] Question 1: The impact of test parameters on test metrics does not meet expectations;

[0074] Question 2: The results of some test indicators are not within the expected range;

[0075] Example

[0076] If neither of the above two types of problems exists, then the selected test parameters and their value ranges remain unchanged; only the value ranges are further subdivided. Then, steps 2-5 above are performed again.

[0077] If either of the above two types of problems is found, the problematic test metric is first selected as the input for the next round of multi-test case generation. Then, the test parameters related to the metric are identified as a new set of candidate parameters. Further, parameters related to the test metric that meet expectations after statistical analysis and parameters that have no impact on the metric output value are removed, and the remaining parameters are selected as test parameters.

[0078] The range of test parameter values ​​remains unchanged, but the values ​​are further subdivided. The values ​​of other parameters are set to constants based on the results of the previous round of testing. The division and determination of parameter relationships remain the same, and the specific implementation process is the same as step 3.

[0079] To address problem 1, the Cartesian product is applied directly to all test parameters during test case generation, ensuring that test cases cover all combinations of the obtained test parameters. For problem 2, an automated and efficient test generator algorithm for combinatorial testing is still used to generate test cases.

[0080] During testing, test cases should be generated primarily for the identified issues. Regardless of whether the issue is real or caused by testing errors, a conclusion for this round of testing must be provided. After resolving the issue, the initially selected test parameters and their value ranges should be used, with further subdivision of the value ranges, and then steps 2-5 should be performed again.

[0081] The iteratively generated test cases are used to redrive the system under test and recalculate all test metrics. It's important to note that the agents exhibit a high degree of autonomy, leading to uncertainty in the behavior of multi-agent systems. Therefore, the iterative evolutionary evaluation method proposed in this paper is needed to conduct multiple iterations to expose problems as comprehensively as possible. The final condition for ending the test iteration is the exhaustion of test resources.

[0082] (1) System Introduction

[0083] This section uses a drone swarm formation cruise system as an example to explain the processing flow of each step. Based on the water demand and indicator system, this system has two test indicators: average centroid distance, which mainly describes the average distance between the centroid of the drone swarm and the predetermined flight path; and total collision count, which mainly describes the total number of non-collisions during the operation of the drone swarm.

[0084] (2) Implementation process.

[0085] Step 1. Select test parameters;

[0086] Based on the determined test metrics, relevant parameters were found from the system under test. In addition, based on expert opinions and the developer's test support capabilities, the final selected test parameters are shown in Table 1 below.

[0087] Table 1 Test Parameter Table;

[0088]

[0089] There are three test parameters related to the average distance to the center of mass, and four test parameters related to the total number of collisions. The initial number of drone swarms is theoretically unrelated to the two test metrics, but since it is a parameter that needs to be tested, it is still chosen as a test parameter.

[0090] Step 2: Determine the parameter values;

[0091] Based on the limitations and testing capabilities of the system under test regarding test parameters, the types and value ranges of test parameters are shown in Table 2 below.

[0092] Table 2. Parameter Value Details;

[0093]

[0094] For natural and percentage parameters, the value range is divided into N equal intervals, and a random value is taken within each interval. For example, the perception rate is restricted to the range [0,1] in the task scenario. The value range can be divided into four intervals between 0 and 1 with a distance of 0.25, namely [0,0.25], (0.25,0.5), (0.5,0.75), and [0.75,1]. Then, a random number is taken from each of the four intervals. Assuming the result is 0.18, 0, 44, 0.62, and 0.83, these four numbers are the initial values ​​for the perception rate test parameter. In this example, all finite-class parameters and percentage parameters are taken with N=4. For natural parameters such as the initial quantity, a typical value of 20 is taken. The final detailed value table is shown in Table 3 below:

[0095] Table of parameters for Wheat 3;

[0096]

[0097] Step 3. Determine the parameter relationships;

[0098] Based on the test metrics, parameters affecting a single metric are grouped together. The same parameter may exist in multiple groups. For parameters whose impact on different test metrics cannot be determined, they are grouped according to their semantics, with semantically related parameters grouped together. In this example, initial position offset, array spacing, and array configuration form one group; sensing rate, packet loss rate, array spacing, and communication mode form another; and the initial quantity forms a separate group.

[0099] Step 4. Confirm parameter constraints;

[0100] Based on the parameters and their values ​​and ranges, determine if there are any parameter value conflicts. If so, adjust the parameter constraints to eliminate the conflicts. In this example, there is no such explicit constraint relationship.

[0101] Step 5. Test case generation;

[0102] After determining the test parameters, their values, and the constraints between them, and grouping the parameters, test cases are generated using a classic automated and efficient test generator algorithm used in combinatorial testing.

[0103] Step 6: Data Analysis;

[0104] First, confirm the impact of the test parameters on the test indicators. The relationship between the indicators and parameters is shown in Table 4. At the same time, confirm the expected values ​​for each indicator: the expected value for the average centroid distance is [0, 200], and the expected value for the total number of collisions is [0, 2].

[0105] Table 4. Impact of Test Parameters;

[0106]

[0107] First, Spearman correlation analysis and descriptive statistical analysis were performed on the test parameters and indicators based on the overall test data. Then, Spearman correlation analysis and descriptive statistical analysis were performed again after classifying the test data according to discrete parameters. The following issues were discovered: the average total number of collisions was higher than expected when the perception rate was 0.83, and the packet loss rate had no correlation with the total number of collisions.

[0108] Therefore, the only test metric for the next round is the total number of collisions, and the only related test parameters are the perception rate and packet loss rate. All other test parameters are set to fixed values. The test parameters and values ​​for this round are shown in Table 5 below.

[0109] Table 5. Parameters and values ​​for the second round of testing;

[0110]

[0111] The parameter relationships and constraints in this section are the same as in the first round. Test cases are generated using the Cartesian product of control variables. Final data analysis confirmed the conclusion that the packet loss rate indeed has no impact on the total number of collisions; the higher-than-expected total number of collisions when the perception rate is 0.83 is an isolated incident. After confirming these two issues, a third round of testing was conducted. Since no issues were found in this round, the selected test parameters and their value ranges remained unchanged, only the value ranges were further subdivided. This round selected to subdivide the ranges into 8 regions. The specific test parameters and their values ​​are shown in Table 6 below:

[0112] Table 6. Test parameters and values ​​for the third round;

[0113]

[0114] The parameter relationships and constraints in this section are the same as in the first round. Test cases are generated using an automated and efficient test generator algorithm. No problems were found in the final data analysis. After the third round, due to limited testing resources, it was necessary to supplement the testing of the natural class parameter, i.e., the initial quantity. In the fourth round, the task environment's constraint on the initial quantity was a minimum of 4 aircraft, so the initial value was 4. Then, the quantity was increased exponentially by 9, 16, 32, 64… The performance of the control group was observed to determine when to stop the increase. This was used to obtain the test results for the natural class parameter. The specific test parameters and parameter values ​​are shown in Table 7 below.

[0115] Table 7. Test parameters and values ​​for the fourth round;

[0116]

[0117] The control group was selected as shown in Table 8 below:

[0118] Table 8. Values ​​of the control group in the fourth round;

[0119]

[0120] The test case generation time was calculated by performing a Cartesian product of the initial number and the control group values. Test resources were exhausted when the number of test cases reached 1024, at which point the test ended. Data analysis showed that the number of tests did not significantly affect the test metrics; however, after the number exceeded 256, the model's running speed was found to be significantly impacted.

[0121] The above examples 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 preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. An iterative evolutionary evaluation method employing an automated and efficient test generator algorithm, wherein the method utilizes an automated and efficient test generator algorithm to iteratively evolve and generate test cases, characterized in that: The specific steps are as follows: Step 1: Select test parameters; Based on the test metrics of multi-agent systems, all test parameters associated with the test metrics are selected from the system environment model and behavior model. The selected test parameters and their values ​​are divided into four categories, and the value rules for each category are as follows: Finite class parameters: The values ​​take the form of a continuous interval, constrained by natural conditions or task scenarios, with clearly defined upper and lower limits; Percentage parameter: The value ranges from [0,1], representing the probability of a specific event occurring; Natural class parameters: values ​​take the value of a continuous interval, but without explicit upper or lower limits; Discrete parameters: take on a finite number of discrete values; This step includes the following sub-steps: Step 1.1: For the finite class parameters and percentage parameters, select several discrete typical values ​​within their value range as test values; Step 1.2: Use the parameter iteration method to obtain the values ​​of the natural class parameters; Step 1.3: Determine the values ​​of discrete parameters using an item-by-item enumeration method; The parameter iteration method for obtaining the values ​​of natural class parameters is as follows: Step 1.21: Select a typical value for this parameter and fix it as a constant, then test other parameters; Step 1.22: After completing most of the tests, select the parameter combination with the best overall performance as the preferred group, and select the combination with the worst performance as the control group. Step 1.23: Fix all parameters in the preferred group and the control group except for the target natural class parameter, adjust only the natural class parameter, and observe its effect on the tested system; Step 1.24: Repeat step 1.23 to obtain the optimal test result corresponding to the natural class parameter; Step 2: Determine the values ​​of the test parameters; Clarify the specific value range and optional values ​​of all the test parameters selected in step 1; Step 3: Determine the correlation between test parameters; The relationships between parameters are determined through the following sub-steps: Step 3.1: Based on the classification of test indicators, group the parameters that affect the same test indicator together; Step 3.2: Support parameters to exist across groups, that is, the same parameter can be classified into multiple related groups; Step 3.3: If the impact of parameters on different test indicators is unclear, group semantically related parameters into one group based on semantic relevance. Step 4: Confirmation of parameter constraints and conflict resolution; Step 4.1: Based on the parameters, their values, and their ranges, determine whether there are any value conflicts among the parameters; Step 4.2: For conflicting parameters, adjust the constraints to eliminate value conflicts and ensure the effectiveness of parameter combinations; Step 5: Test case generation based on an automated and efficient test generator algorithm; Based on steps 1-4, the parameters are grouped, and test cases are generated using an automated and efficient test generator algorithm. The specific sub-steps are as follows: Step 5.1: Select parameter X in the current parameter group, and its value Y must satisfy the following condition: Y has the highest frequency of occurrence in the uncovered parameter combinations; Step 5.2: Denote the selected parameter X as X1, and randomly sort the remaining parameters to obtain the parameter sequence X1, X2, X3, ..., Xn; Step 5.3: Determine the corresponding value sequence for the parameter sequence X1, X2, ..., Xn, where Ym is the value of parameter Xm. The specific process is as follows: Step 5.31: Determine the next parameter X to be assigned a value. n+1 ; Step 5.32: For X n+1 For each selectable value Y, calculate the set {(X)} n+1 The number of new parameter combinations that are not covered in { =Y, Xm=Ym), 1≤m≤n}; Step 5.33: Select the value Y that produces the most new parameter combinations as X. n+1 The final value; Step 5.4: Repeat steps 5.1-5.3 until complete parameter values ​​for all test cases are generated.

2. The iterative evolutionary evaluation method using an automatic and efficient test generator algorithm as described in claim 1, characterized in that: The method for determining the discrete typical value in step 1.1 includes: uniformly dividing the value interval into several smaller intervals, and taking a random value in each smaller interval.

3. The iterative evolutionary evaluation method using an automatic and efficient test generator algorithm as described in claim 1, characterized in that: The percentage-type parameter mentioned in step 1 represents the probability of a class of events occurring.

4. The iterative evolutionary evaluation method using an automatic and efficient test generator algorithm as described in claim 1, characterized in that: The finite class parameters mentioned in step 1 are the initial position offset and the array spacing.

5. The iterative evolutionary evaluation method using an automatic and efficient test generator algorithm according to claim 1, characterized in that: The percentage parameters mentioned in step 1 are the sensing rate and the packet loss rate.

6. The iterative evolutionary evaluation method using an automatic and efficient test generator algorithm according to claim 1, characterized in that: The natural conditions or task scenarios described in step 1 impose relatively weak restrictions on natural parameters, which are the initial quantities.

7. The iterative evolutionary evaluation method using an automatic and efficient test generator algorithm according to claim 1, characterized in that: The test values ​​for the natural class parameters are obtained by first taking a typical value of one parameter and using it as a constant. Then, the other parameters are tested. After most of the tests are completed, a parameter combination with good overall performance and a parameter combination with poor overall performance are selected as a control group. The other parameters in the two combinations are fixed as constants, and only the natural class parameters in the two combinations are adjusted to observe the impact of the natural class parameters on the tested system, thereby obtaining the test results of the natural class parameters.

8. The iterative evolutionary evaluation method using an automatic and efficient test generator algorithm according to claim 1, characterized in that: The discrete parameters mentioned in step 1 are the communication mode and array configuration.

9. The iterative evolutionary evaluation method using an automatic and efficient test generator algorithm according to claim 1, characterized in that: The multi-agent system is a group system composed of multiple autonomous individuals. It transforms a large and complex system into a small, easily managed system that can communicate and interact with each other.

10. The iterative evolutionary evaluation method using an automatic and efficient test generator algorithm according to claim 1, characterized in that: The multi-agent system is a drone swarm formation cruise system.

Images