Artificial intelligence application software anomaly detection device and method

Abstract

The invention relates to the technical field of application software anomaly detection, in particular to an artificial intelligence application software anomaly detection device and method, and the method is realized based on an artificial intelligence application software anomaly detection system. The system comprises a data collection module, a preliminary data analysis module, a three-number middle-taking method partition detection module, a recursion detection module, a double-shaft rapid sorting anomaly detection module and an anomaly diagnosis classification module. The method sequentially comprises the following steps of data acquisition, benchmark test, partition anomaly detection, recursion anomaly detection, double-shaft rapid sorting anomaly detection and anomaly diagnosis and repair, and can comprehensively detect whether the logic of an algorithm is correct or not, including a partition process and a recursion process, so that potential logic errors are found, and the accuracy of the algorithm is improved. And algorithm logic anomaly detection is quickly carried out on the artificial intelligence application software.

Description

Technical Field

[0001] The present invention relates to the technical field of application software anomaly detection, and in particular to an artificial intelligence application software anomaly detection device and method. Background Art

[0002] Artificial intelligence software refers to technical products that can simulate and extend human intelligence. It can process and solve complex problems through learning, reasoning, perception, planning and other methods. It is widely used in various fields in modern society, such as risk control systems and recommendation systems. It not only provides risk identification, real-time monitoring and decision support for risk control systems, but also provides personalized recommendations, precision marketing and content optimization for recommendation systems.

[0003] Therefore, the operation and maintenance of artificial intelligence software is also very important. Generally, traditional methods of anomaly detection for artificial intelligence software mainly include code review, unit testing, and integration testing. However, these methods are still unable to detect algorithm logic. They mainly focus on the correctness of the code and the implementation of functions, but it is difficult to conduct in-depth detection of the algorithm logic, resulting in some potential algorithm problems, such as logical errors and performance problems, which may not be discovered during the testing phase. Due to the complexity and scale of artificial intelligence software, traditional detection methods require a lot of time and manpower for testing. At the same time, it is difficult to cope with dynamic changes, such as changes in data distribution and changes in user behavior.

[0004] In summary, it is necessary to propose an artificial intelligence application software anomaly detection device and method to solve the above problems. Summary of the invention

[0005] The purpose of the present invention is to provide an artificial intelligence application software anomaly detection device and method to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] An artificial intelligence application software anomaly detection method, the method is implemented based on an artificial intelligence application software anomaly detection system, the system includes a data collection module, a preliminary data analysis module, a three-number middle method partition detection module, a recursive detection module, a dual-axis quick sort anomaly detection module and an anomaly diagnosis classification module;

[0008] The method comprises the following steps:

[0009] S1. Collecting the operation data of the artificial intelligence application software by obtaining log files, calling API interfaces or system monitoring software. This step is implemented by the data collection module;

[0010] S2. Use the random selection benchmark method to perform a preliminary analysis of the data and evaluate the distribution of the data. Let the array be A 1 , with a length of n 1 , generate a random number r (0 <= r <n 1 ), select A 1 [r] as a benchmark pivot 1 , benchmark the collected data, simulate the partitioning process of quick sort by randomly selecting benchmark elements, evaluate the distribution of the data, and detect whether the data has abnormal distribution, thereby indirectly reflecting the possible logical errors in the algorithm. This step is implemented by the preliminary data analysis module;

[0011] S3. Use the three-number median method to select the benchmark method, assuming that the array is A 2 , with a length of n 2 , calculate the middle index mid = floor (n 2 / 2), take A 2 [0], A 2 [n 2 -1] and A 2 The median value of [mid] is used as the benchmark pivot 2 , partition the data so that the elements smaller than the benchmark are on the left side of the benchmark and the elements larger than the benchmark are on the right side of the benchmark, and check whether the partition result meets expectations, which is used to detect whether there are abnormalities in the partition process. Partition abnormalities include element misalignment and improper handling of repeated elements. This step is implemented by the three-number middle method partition detection module;

[0012] S4. Recursively detect the partitioned sub-arrays by simulating a recursive process, and iteratively process smaller sub-arrays to reduce stack space usage, to detect whether there are stack overflows and abnormal problems such as excessive recursive depth during the recursive process, and this step is implemented by the recursive detection module;

[0013] S5. Select the benchmark pivot in step S2 1 and the benchmark pivot in step S3 2 , the array is rearranged into the following three parts:

[0014] Smaller than pivot 1 part of;

[0015] In Pivot 1 and pivot 2 The part between;

[0016] Greater than pivot 2 part of;

[0017] Then, the data is anomaly detected by using the dual-axis quick sort, with special attention paid to the selection and partitioning results of the two benchmarks, to detect whether there are abnormal problems such as incorrect partitioning and unreasonable benchmark selection in the case of dual benchmarks. This step is implemented by the dual-axis quick sort anomaly detection module;

[0018] S6. Based on the detection results, the anomalies are diagnosed and classified by using a machine learning algorithm for the post-repair process of the artificial intelligence application software. This step is implemented by the anomaly diagnosis and classification module.

[0019] Preferably, the data collection module is composed of a log file reading component, an API interface calling component and monitoring software;

[0020] The log file reading component is used for log parsing and file scanning;

[0021] The API interface calling component is used to call the external API interface to obtain the operation data related to the artificial intelligence application software and interact and integrate with the data of the third-party service or system;

[0022] The monitoring software is used to receive monitoring data sent by the monitoring software in real time or periodically, providing support for performance analysis and fault diagnosis of artificial intelligence application software.

[0023] Preferably, the preliminary data analysis module consists of a random number generator, a benchmark executor and a data distribution evaluator:

[0024] The random number generator is used to generate a random index r, which is used to select an element from the data set as the benchmark pivot 1 ;

[0025] The benchmark executor selects the benchmark pivot based on a random number generator. 1 , used to benchmark the collected data;

[0026] The data distribution evaluator is used to analyze the results of the benchmark test and evaluate the distribution of the data.

[0027] Preferably, the three-number median method partition detection module is composed of an index calculator, a median selector and a partition executor;

[0028] The index calculator is used to calculate the array A in step S3. 2 The middle index mid;

[0029] The median selector is used to select the median value from array A according to the index provided by the index calculator. 2 Select three elements (A 2 [0], A 2 [mid]、A2 [n 2 -1]), and calculate the median of these three elements as the benchmark pivot 2 ;

[0030] The partition executor is used to pivot according to the benchmark 2 For array A 2 Partitioning.

[0031] Preferably, the recursive detection module is composed of a recursive simulator, an iterative processor and an anomaly detector:

[0032] The recursive simulator is used to perform recursive detection on the partitioned subarrays during the simulated recursive process, and track the level and order of the recursive calls;

[0033] The iterative processor is used to process smaller sub-arrays, reducing the use of stack space;

[0034] The anomaly detector is used to monitor and record abnormal behaviors such as excessive recursion depth and stack overflow during simulation of recursive processes and iterative processing.

[0035] Preferably, the dual-axis quick sorting anomaly detection module is composed of a dual-reference selector, an array rearranger and a dual-axis quick sorting executor:

[0036] The dual reference selector is used to select the reference pivot from step S2 and step S3. 1 and pivot 2 ;

[0037] The array rearranger is used to rearrange the array according to two references provided by the dual reference selector;

[0038] The dual-axis quick sort executor is used to sort the rearranged array according to the dual-axis quick sort method.

[0039] Preferably, the three-number median partition detection module and the dual-axis quick sort anomaly detection module both include a partition result checker;

[0040] The partition result checker is used to verify whether there is element misalignment and improper handling of repeated elements in the partition process of the three-number centering method in the detection step S3;

[0041] The partition result checker is also used to detect whether the partition result of the dual-axis quick sort in step S5 meets expectations.

[0042] Preferably, the abnormal diagnosis and classification module is composed of a machine learning algorithm integration module, an abnormal diagnosis device and a classifier;

[0043] The machine learning algorithm integration module is used to integrate support vector machines, decision trees and neural networks;

[0044] The anomaly diagnostic device uses an integrated machine learning algorithm to analyze and diagnose the anomaly data and identify the root cause and type of the anomaly;

[0045] The classifier is used to classify abnormal data into different types according to the result of the abnormality diagnostic device.

[0046] Based on the above system and method, the present invention also proposes an artificial intelligence application software anomaly detection device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the artificial intelligence application software anomaly detection method when executing the computer program.

[0047] Based on the above device, the present invention also proposes a storage medium, on which a computer program is stored. When the computer program is executed by a processor, an artificial intelligence application software anomaly detection method is implemented.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] The present invention provides a comprehensive detection of the algorithm logic of artificial intelligence software. By simulating the execution process of the algorithm and analyzing the data, it comprehensively detects whether the logic of the algorithm is correct, including the partition process and the recursive process, so as to discover potential logical errors and quickly perform algorithm logic anomaly detection on the artificial intelligence application software. At the same time, it integrates and optimizes traditional detection methods to more accurately discover and handle software abnormalities, monitor the running status and data changes of the artificial intelligence application software in real time, promptly discover and handle abnormal situations, and analyze and diagnose abnormal data through machine learning algorithms, so as to more accurately identify the root cause and type of the anomaly, and provide strong support for later repairs.

[0050] In summary, the artificial intelligence application software anomaly detection device and method provided by the present invention can comprehensively and accurately detect algorithmic logic anomalies in artificial intelligence application software, thereby improving the stability and reliability of the software. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 The flowchart of the artificial intelligence application software anomaly detection method of the present invention is shown;

[0052] Figure 2 The application flow chart of the present invention in the recommendation system is shown;

[0053] Figure 3 The flowchart of the application of the present invention in the wind control system is shown. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0055] See also Figure 1 , the present invention proposes an artificial intelligence application software anomaly detection system, which is stored in an artificial intelligence application software anomaly detection device through a storage medium and is run through a computer program running on a processor of the artificial intelligence application software anomaly detection device;

[0056] Among them, the artificial intelligence application software anomaly detection system includes a data collection module, a preliminary data analysis module, a three-number middle method partition detection module, a recursive detection module, a dual-axis quick sort anomaly detection module and an anomaly diagnosis classification module;

[0057] Specifically, the data collection module consists of a log file reading component, an API interface calling component, and monitoring software;

[0058] The log file reading component is used for log parsing and file scanning;

[0059] The API interface calling component is used to call external API interfaces to obtain operation data related to artificial intelligence application software and interact and integrate data with third-party services or systems;

[0060] The monitoring software is used to receive monitoring data sent by the monitoring software in real time or periodically, and to provide support for performance analysis and fault diagnosis of the artificial intelligence application software;

[0061] This module is used to implement the following steps:

[0062] S1. Collect the operation data of artificial intelligence application software through log file acquisition, API interface call or system monitoring software;

[0063] In step S1, the operation data of the artificial intelligence application software is collected by acquiring log files, calling API interfaces or using system monitoring software, so as to provide a comprehensive and accurate data basis for subsequent anomaly detection, and the data reflects the operation status of the artificial intelligence software and the execution of algorithm logic, which serves as a direct basis for diagnosis;

[0064] Specifically, the preliminary data analysis module consists of a random number generator, a benchmark executor, and a data distribution evaluator:

[0065] The random number generator is used to generate a random index r, which is used to select an element from the dataset as the pivot.1 ;

[0066] The benchmark executor selects the benchmark pivot based on a random number generator. 1 , used to benchmark the collected data;

[0067] The data distribution evaluator is used to analyze the results of the benchmark and evaluate the distribution of the data;

[0068] This module is used to implement the following steps:

[0069] S2. Use the random selection benchmark method to perform a preliminary analysis of the data and evaluate the distribution of the data. Let the array be A 1 , with a length of n 1 , generate a random number r (0 <= r <n 1 ), select A 1 [r] as a benchmark pivot 1 ,Benchmark the collected data, simulate the partitioning process of quick sort by randomly selecting benchmark elements, and evaluate the distribution of data, which is used to detect whether the data has abnormal distribution, thereby indirectly reflecting possible logical errors in the algorithm;

[0070] In step S2, the data is preliminarily analyzed using a random selection benchmark test method to evaluate the distribution of the data and detect whether the data has abnormal distribution, thereby indirectly reflecting the role of possible logical errors in the algorithm, quickly locating data anomalies, and providing clues for in-depth analysis;

[0071] Specifically, the partition detection module of the three-number median method consists of an index calculator, a median selector, and a partition executor;

[0072] Used by the index calculator to calculate the array A in step S3 2 The middle index mid;

[0073] The median selector is used to select the median value from array A according to the index provided by the index calculator. 2 Select three elements (A 2 [0], A 2 [mid]、A 2 [n 2 -1]), and calculate the median of these three elements as the benchmark pivot 2 ;

[0074] The partition executor is used to pivot according to the benchmark 2 For array A 2 Partitioning;

[0075] This module is used to implement the following steps:

[0076] S3. Use the three-number median method to select the benchmark method, assuming that the array is A 2 , with a length of n 2 , calculate the intermediate index:

[0077] mid=floor(n 2 / 2);

[0078] At that time, take A 2 [0], A 2 [n 2 -1] and A 2 The median value of [mid] is used as the benchmark pivot 2 , when partitioning, the data is partitioned with elements smaller than the benchmark on the left side of the benchmark and elements larger than the benchmark on the right side of the benchmark, and the partitioning result is checked to see if it meets expectations. This is used to detect whether there are anomalies in the partitioning process. Partitioning anomalies include element misalignment and improper handling of duplicate elements;

[0079] In step S3, a benchmark method is selected by using the three-number median method to perform partition detection on the array, and detect whether there are abnormalities in the partitioning process, such as element misalignment and improper handling of duplicate elements, to ensure the correctness of the partitioning operation and verify the algorithm logic;

[0080] Specifically, the recursive detection module consists of a recursive simulator, an iterative processor, and an anomaly detector:

[0081] The recursive simulator is used to perform recursive detection on the partitioned subarrays during the simulated recursive process and track the level and order of recursive calls;

[0082] Iteration processor is used to process smaller sub-arrays, reducing the use of stack space;

[0083] The anomaly detector is used to monitor and record abnormal behaviors such as excessive recursion depth and stack overflow during simulated recursive and iterative processing;

[0084] This module is used to implement the following steps:

[0085] S4. recursively detect the partitioned sub-arrays by simulating the recursive process, and use an iterative method to process smaller sub-arrays to reduce the use of stack space, so as to detect whether there are stack overflow and excessive recursive depth abnormal problems in the recursive process;

[0086] The recursive process is simulated in step S4 to recursively detect the partitioned sub-arrays, and the smaller sub-arrays are processed in an iterative manner to reduce the use of stack space, and to detect whether there are abnormal problems such as stack overflow and excessive recursive depth in the recursive process, so as to ensure the stability and efficiency of the recursive algorithm;

[0087] Specifically, the dual-axis quick sort anomaly detection module consists of a dual-reference selector, an array rearranger, and a dual-axis quick sort executor:

[0088] The dual pivot selector is used to select the pivot from step S2 and step S3. 1 and pivot 2 ;

[0089] The array rearranger is used for rearranging the array according to two references provided by the dual reference selector;

[0090] The dual-axis quick sort executor is used to sort the rearranged array according to the dual-axis quick sort method;

[0091] This module is used to implement the following steps:

[0092] S5. Select the benchmark pivot in step S2 1 and the benchmark pivot in step S3 2 , the array is rearranged into the following three parts:

[0093] Smaller than pivot 1 part of;

[0094] In Pivot 1 and pivot 2 The part between;

[0095] Greater than pivot 2 part of;

[0096] Then, we use dual-axis quick sorting to detect anomalies in the data, paying special attention to the selection and partitioning results of the two benchmarks, and detecting whether there are abnormal problems such as incorrect partitioning and unreasonable benchmark selection in the case of dual benchmarks.

[0097] In step S5, two benchmarks are selected, the array is rearranged, and anomaly detection is performed using a dual-axis quick sort method, with particular attention paid to the selection and partitioning results of the two benchmarks, and the function of detecting whether there are abnormal problems such as incorrect partitioning and unreasonable benchmark selection in the case of dual benchmarks can be used to more comprehensively verify the correctness and efficiency of the sorting algorithm;

[0098] Among them, in step S3 and step S5, it should also be noted that the three-number middle method partition detection module and the dual-axis quick sort anomaly detection module both include a partition result checker;

[0099] The partition result checker is used to verify whether there is element misalignment and improper handling of repeated elements in the partitioning process of the three-number center method in the detection step S3;

[0100] The partition result checker is also used to detect whether the partition result of the dual-axis quick sort in step S5 meets expectations;

[0101] Specifically, the anomaly diagnosis and classification module is composed of a machine learning algorithm integration module, an anomaly diagnostic device, and a classifier;

[0102] The machine learning algorithm integration module is used to integrate support vector machines, decision trees, and neural networks;

[0103] The anomaly diagnostic tool uses integrated machine learning algorithms to analyze and diagnose abnormal data and identify the root cause and type of anomalies;

[0104] The classifier is used to classify abnormal data into different types according to the results of the anomaly diagnostic device;

[0105] This module is used to implement the following steps:

[0106] S6. Based on the detection results, the abnormalities are diagnosed and classified by using machine learning algorithms for the post-repair process of artificial intelligence application software;

[0107] Through step S6, the machine learning algorithm is used to diagnose and classify the anomalies according to the detection results, providing accurate and detailed anomaly information for the later repair process of the artificial intelligence application software, helping developers to quickly locate problems, analyze causes and take corresponding repair measures.

[0108] Example 1

[0109] Clearly refer to Figure 2 Based on the above device, system and method, in the recommendation system, the artificial intelligence software algorithm is responsible for recommending products based on the user's browsing history and purchasing behavior. The application of the system, device and method of the present invention in the abnormal detection of the recommendation system includes the following steps:

[0110] The log file reading component is used to collect the operation logs of the recommendation system, including user behavior data and recommendation results. The API interface calling component is used to obtain relevant data from other services of the e-commerce platform, including product information and user portraits.

[0111] Randomly select benchmark elements and benchmark the recommendation result data to evaluate the distribution of the data and detect whether there is abnormal distribution, including some products being over-recommended or never recommended;

[0112] Select the median of the three elements (first, last, and middle) in the recommendation result array as the benchmark, perform partition detection, and check whether the partition result meets expectations, that is, products with values ​​smaller than the benchmark are on the left, and products with values ​​larger than the benchmark are on the right;

[0113] Simulate the recursive process to perform recursive detection on the partitioned sub-arrays to ensure that there is no stack overflow or excessive recursive depth during the recursive process;

[0114] Select two benchmark elements (the benchmark in the preliminary data analysis and the benchmark in the three-number median method), rearrange the recommendation result array into three parts, use the dual-axis quick sort to detect anomalies in the data, and pay special attention to whether the partition results are correct;

[0115] Based on the detection results, machine learning algorithms are used to diagnose and classify anomalies, and abnormal data is classified into different types of logical errors, including duplicate recommendation results and unreasonable recommendations.

[0116] Through the above steps, various logical error anomalies in the recommendation system are detected and classified. Based on the classification results, the development team can quickly locate the problem and fix it to solve the algorithmic logic error problem of artificial intelligence software in the recommendation system.

[0117] Example 3

[0118] See also Figure 3 Based on the above-mentioned device, system and method, in the risk control system, the artificial intelligence software algorithm is responsible for assessing the risk level according to the user's transaction behavior and credit record. The application of the system, device and method of the present invention in the abnormal detection of the recommendation system includes the following steps:

[0119] The monitoring software collects the operating data of the risk control system in real time, including user transaction records, credit scores, etc., and uses the API interface to call components to obtain the user's credit report from an external credit assessment agency;

[0120] Benchmark the collected data to assess its distribution and detect any abnormal distributions, such as unusually high or low credit scores for some users;

[0121] Select the median of the three elements (first, last, and middle) in the user credit score array as the benchmark, perform partition detection, and check whether the partition result meets expectations, that is, users with credit scores lower than the benchmark are on the left, and users with credit scores higher than the benchmark are on the right;

[0122] Perform recursive checks on the partitioned subarrays to ensure that there is no stack overflow or excessive recursion depth during the recursive process;

[0123] Select two benchmark elements (the benchmark in the preliminary data analysis and the benchmark in the three-number median method), rearrange the user credit score array into three parts, and use dual-axis quick sort to detect anomalies in the data, paying special attention to whether the partition results are correct and whether there are any missed reviews;

[0124] Based on the detection results, machine learning algorithms are used to diagnose and classify anomalies, and abnormal data is classified into different types of logical errors, such as inaccurate credit score assessment and missed assessment.

[0125] Through this method, logical errors and anomalies in the artificial intelligence software in the risk control system are detected and classified. Based on the classification results, the risk control team can quickly respond and fix the problems, improving the accuracy and reliability of the risk control system.

[0126] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for detecting anomalies in artificial intelligence application software, which is implemented based on an artificial intelligence application software anomaly detection system, and is characterized in that: The system includes a data collection module, a preliminary data analysis module, a median-of-three partitioning detection module, a recursive detection module, a dual-pivot quicksort anomaly detection module, and an anomaly diagnosis and classification module; It is characterized in that the method includes the following steps: S1. Collect the operation data of the artificial intelligence application software by means of log file acquisition, API interface call or system monitoring software, and this step is implemented by the data collection module; S2. Use the random selection benchmark test method to conduct a preliminary analysis of the data, evaluate the data distribution. Let the array be A1 with a length of n1, generate a random number r (0 <= r < n1), select A1[r] as the benchmark pivot1, and conduct a benchmark test on the collected data. By randomly selecting the benchmark element to simulate the partitioning process of quicksort, evaluate the data distribution, and use it to detect whether there is an abnormal distribution in the data, so as to indirectly reflect the possible logical errors in the algorithm. This step is implemented by the preliminary data analysis module; S3. Use the median-of-three method to select the benchmark method. Let the array be A2 with a length of n2, calculate the middle index: mid = floor(n2 / 2), and take the median of A2[0], A2[n2 - 1] and A2[mid] as the benchmark pivot2. When partitioning, partition the data with elements less than the benchmark on the left side of the benchmark and elements greater than the benchmark on the right side of the benchmark, and check whether the partitioning result meets the expectations, and use it to detect whether there are any anomalies in the partitioning process. Partitioning anomalies include element misalignment and improper handling of duplicate elements. This step is implemented by the median-of-three partitioning detection module; S4. Conduct a recursive detection on the partitioned sub-arrays by simulating the recursive process, and at the same time use an iterative method to process smaller sub-arrays to reduce the use of stack space, and use it to detect whether there are abnormal problems such as stack overflow and excessive recursive depth in the recursive process. This step is implemented by the recursive detection module; S5. Select the benchmark pivot1 in step S2 and the benchmark pivot2 in step S3, and rearrange the array into the following three parts: The part less than pivot1; The part between pivot1 and pivot2; The part greater than pivot2; Then use dual-pivot quicksort to conduct anomaly detection on the data, especially pay attention to the selection of the two benchmarks and the partitioning results, and detect whether there are abnormal problems such as incorrect partitioning and unreasonable benchmark selection in the case of dual benchmarks. This step is implemented by the dual-pivot quicksort anomaly detection module; S6. According to the detection results, use machine learning algorithms to diagnose and classify the anomalies for the later repair process of the artificial intelligence application software. This step is implemented by the anomaly diagnosis and classification module.

2. The artificial intelligence application software anomaly detection method according to claim 1, characterized in that: The data collection module consists of a log file reading component, an API interface call component, and a monitoring software; The log file reading component is used for log parsing and file scanning; The API interface call component is used to call external API interfaces, obtain the operation data related to the artificial intelligence application software, and interact and integrate with the data of third-party services or systems; The monitoring software is used to receive monitoring data sent by the monitoring software in real time or periodically, providing support for performance analysis and fault diagnosis of artificial intelligence application software.

3. The artificial intelligence application software anomaly detection method according to claim 2, characterized in that: The preliminary data analysis module consists of a random number generator, a benchmark executor, and a data distribution evaluator: The random number generator is used to generate a random index r, which is used to select an element from the data set as a benchmark pivot1; The benchmark test executor is used to perform a benchmark test on the collected data according to the benchmark pivot1 selected by the random number generator; The data distribution evaluator is used to analyze the results of the benchmark test and evaluate the distribution of the data.

4. The artificial intelligence application software anomaly detection method according to claim 3, characterized in that: The three-number median partition detection module is composed of an index calculator, a median selector and a partition executor; The index calculator is used to calculate the middle index mid of the array A2 in step S3; The median selector is used to select three elements (A2[0], A2[mid], A2[n2-1]) from the array A2 according to the index provided by the index calculator, and calculate the median of the three elements as the benchmark pivot2; The partition executor is used to partition the array A2 according to the reference pivot2.

5. The artificial intelligence application software anomaly detection method according to claim 4, characterized in that: The recursive detection module consists of a recursive simulator, an iterative processor and an anomaly detector: The recursive simulator is used to perform recursive detection on the partitioned subarrays during the simulated recursive process, and track the level and order of the recursive calls; The iterative processor is used to process smaller sub-arrays, reducing the use of stack space; The anomaly detector is used to monitor and record abnormal behaviors such as excessive recursion depth and stack overflow during simulation of recursive processes and iterative processing.

6. The artificial intelligence application software anomaly detection method according to claim 5, characterized in that: The dual-axis quick sort anomaly detection module is composed of a dual-reference selector, an array rearranger, and a dual-axis quick sort executor: The dual reference selector is used to select the references pivot1 and pivot2 from step S2 and step S3; The array rearranger is used to rearrange the array according to two references provided by the dual reference selector; The dual-axis quick sort executor is used to sort the rearranged array according to the dual-axis quick sort method.

7. The artificial intelligence application software anomaly detection method according to claim 6, characterized in that: The three-number median partition detection module and the dual-axis quick sort anomaly detection module both include a partition result checker; The partition result checker is used to verify whether there is element misalignment and improper handling of repeated elements in the partition process of the three-number centering method in the detection step S3; The partition result checker is also used to detect whether the partition result of the dual-axis quick sort in step S5 meets expectations.

8. The artificial intelligence application software anomaly detection method according to claim 7, characterized in that: The abnormal diagnosis and classification module is composed of a machine learning algorithm integration module, an abnormal diagnosis device and a classifier; The machine learning algorithm integration module is used to integrate support vector machines, decision trees and neural networks; The anomaly diagnostic device uses an integrated machine learning algorithm to analyze and diagnose the anomaly data and identify the root cause and type of the anomaly; The classifier is used to classify abnormal data into different types according to the result of the abnormality diagnostic device.

9. An artificial intelligence application software anomaly detection device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the artificial intelligence application software anomaly detection method as described in any one of claims 1-8 are implemented.

10. A storage medium, characterized in that: The storage medium stores a computer program, which, when executed by a processor, implements the steps of the artificial intelligence application software anomaly detection method as described in any one of claims 1 to 8.

Images