An artificial intelligence-based system interface verification method and system

Abstract

This invention provides a system and method for verifying system interfaces based on artificial intelligence, relating to the field of interface verification technology. The method includes: acquiring historical concurrent user counts, historical system performance data, and historical load impact data; acquiring system interface architecture data and historical system interface architecture data; acquiring real-time load impact data; determining a concurrent user count relationship function based on historical concurrent user counts and historical load impact data; determining the predicted concurrent user count for a prediction period based on the real-time load impact data and the concurrent user count relationship function; obtaining a trained system performance prediction model; processing the predicted concurrent user counts and system interface architecture data according to the trained system performance prediction model to obtain predicted system performance coefficients; and generating a system interface verification report based on the predicted system performance coefficients. According to this invention, the accuracy of system interface verification can be improved.

Description

Technical Field

[0001] This invention relates to the field of interface verification technology, and in particular to a system interface verification method and system based on artificial intelligence. Background Technology

[0002] In related technologies, system interface verification can be performed based on artificial intelligence. However, these technologies struggle to predict the load on system interfaces and perform predictive performance testing based on the predicted load. In other words, they are unable to predict the interface performance bottleneck under future loads.

[0003] The information disclosed in the background section of this application is intended only to enhance the understanding of the general background of this application and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0004] This invention provides a system interface verification method and system based on artificial intelligence, which can solve the technical problem that related technologies have difficulty predicting the interface performance bottleneck under future loads.

[0005] According to a first aspect of the present invention, an artificial intelligence-based system interface verification method is provided, comprising:

[0006] Obtain historical concurrent user counts, historical system performance data, and historical load impact data across multiple historical periods;

[0007] Obtain system interface architecture data and historical system interface architecture data;

[0008] Obtain real-time load impact data;

[0009] Based on the historical concurrent user count and the historical load impact data, determine the concurrent user count relationship function;

[0010] Based on the real-time load impact data and the concurrent user number relationship function, determine the predicted concurrent user number for the prediction period;

[0011] Based on the historical system interface architecture data, the historical concurrent user count, and the historical system performance data, the system performance prediction model is trained to obtain the trained system performance prediction model.

[0012] The predicted concurrent user count and the system interface architecture data are processed based on the trained system performance prediction model to obtain the predicted system performance coefficients.

[0013] Based on the predicted system performance coefficients, a system interface verification report is generated.

[0014] According to the present invention, a function relating concurrent users is determined based on the historical concurrent user count and the historical load impact data, including:

[0015] Obtain the historical baseline number of monthly active users for each historical period;

[0016] Based on the historical load impact data, determine the historical advertising volume, historical weather temperature, and historical weather rainfall.

[0017] The relationship function for the number of concurrent users is determined based on the historical concurrent user count, the historical monthly active baseline user count, the historical advertising volume, the historical weather temperature, and the historical weather rainfall.

[0018] According to the present invention, a function relating concurrent users is determined based on the historical concurrent user count, the historical monthly active baseline user count, the historical advertising volume, the historical weather temperature, and the historical weather rainfall, including: according to the formula:

[0019] ,

[0020] Determine the equation of undetermined coefficients for the function relating concurrent users, where, Let be the historical concurrent users in the i-th historical period. To preset the threshold for the number of concurrent users, Let be the historical monthly active baseline number of users corresponding to the i-th historical period. This represents the historical advertising volume for the i-th historical period. To preset the threshold for ad delivery volume, Let be the historical weather temperature for the i-th historical period. To preset the weather temperature threshold, Let be the historical rainfall for the i-th historical period. To preset the weather rainfall threshold, , , , , , , , and The first undetermined coefficient of the equation of undetermined coefficients;

[0021] Based on the historical concurrent user count, the historical monthly active baseline user count, the historical advertising volume, the historical weather temperature, and the historical weather rainfall, the first undetermined coefficient is solved to obtain the solution value of the first undetermined coefficient;

[0022] Based on the solved values ​​of the first undetermined coefficients and the equation of the undetermined coefficients, the function relating the number of concurrent users is determined.

[0023] According to the present invention, determining the predicted concurrent users for a prediction period based on the real-time load impact data and the concurrent user count relationship function includes:

[0024] Obtain the current active baseline number of users corresponding to the prediction period;

[0025] Based on the real-time load impact data, determine the planned advertising volume;

[0026] Based on the real-time load impact data, weather forecast data is determined;

[0027] Based on the aforementioned weather forecast data, the predicted weather temperature and predicted weather rainfall are determined;

[0028] The predicted concurrent user count for the prediction period is determined based on the relationship function between the current active baseline user count, the planned advertising volume, the predicted weather temperature, the predicted weather rainfall, and the concurrent user count.

[0029] According to the present invention, a system performance prediction model is trained based on the historical system interface architecture data, the historical concurrent user count, and the historical system performance data to obtain a trained system performance prediction model, including:

[0030] Based on the historical system performance data, determine the historical system performance coefficients;

[0031] The historical concurrent user count and the system interface architecture data are processed according to the historical system performance prediction model to obtain the sample system performance coefficient. The historical system interface architecture data includes: the historical maximum number of threads in the thread pool, the historical maximum number of connections in the database connection pool, and the historical number of containerized replicas.

[0032] Based on the historical system performance coefficients, the sample system performance coefficients, the historical concurrent user counts, and the historical system interface architecture data, determine the training loss function for the system performance prediction model;

[0033] The system performance prediction model is trained using the training loss function of the system performance prediction model to obtain the trained system performance prediction model.

[0034] According to the present invention, determining the historical system performance coefficient based on the historical system performance data includes:

[0035] Based on the historical system performance data, determine the historical response time, historical application instance metrics, historical operating system CPU utilization, and historical database connection count;

[0036] Based on the historical response time, determine the historical user perception layer performance coefficient;

[0037] Based on the historical application instance metrics, determine the historical application layer performance coefficients;

[0038] Based on the historical operating system CPU utilization, determine the historical system resource layer performance coefficient;

[0039] Based on the historical database connection count, determine the historical downstream dependency layer performance coefficient;

[0040] The historical system performance coefficient is determined based on the historical user perception layer performance coefficient, the historical application layer performance coefficient, the historical system resource layer performance coefficient, and the historical downstream dependency layer performance coefficient.

[0041] According to the present invention, the training loss function of the system performance prediction model is determined based on the historical system performance coefficients, the sample system performance coefficients, the historical concurrent user counts, and the historical system interface architecture data, including: according to the formula:

[0042] ,

[0043] Determine the training loss function of the system performance prediction model ,in, Let be the historical system performance coefficient for the i-th historical period. Let be the sample system performance coefficients for the i-th historical period. Let be the historical concurrent users in the i-th historical period. To preset the threshold for the number of concurrent users, The maximum number of threads in the historical thread pool for the i-th historical period. To preset the thread count threshold, The maximum number of connections in the historical database connection pool for the i-th historical period. To preset the maximum number of connections threshold, Let be the number of containerized replicas of the history in the i-th historical period. The preset replica count threshold is denoted as n, where n is the number of historical periods, i ≤ n, and both i and n are positive integers.

[0044] According to a second aspect of the present invention, an artificial intelligence-based system interface verification system is provided, comprising:

[0045] The historical data module is used to obtain historical concurrent user counts, historical system performance data, and historical load impact data for multiple historical periods.

[0046] The architecture data module is used to obtain system interface architecture data and historical system interface architecture data;

[0047] The impact data module is used to obtain real-time load impact data;

[0048] The relational function module is used to determine the relational function for the number of concurrent users based on the historical concurrent user count and the historical load impact data;

[0049] The predictive user module is used to determine the predicted concurrent user count for the prediction period based on the real-time load impact data and the concurrent user count relationship function.

[0050] The model training module is used to train the system performance prediction model based on the historical system interface architecture data, the historical concurrent user count, and the historical system performance data, so as to obtain the trained system performance prediction model.

[0051] The prediction coefficient module is used to process the predicted concurrent user count and the system interface architecture data according to the trained system performance prediction model to obtain the predicted system performance coefficients.

[0052] The verification report module is used to generate a system interface verification report based on the predicted system performance coefficients.

[0053] Technical Effects: According to this invention, the relationship between load impact data and concurrent users can be accurately analyzed. Based on this relationship, the number of concurrent users in the prediction period can be predicted. Furthermore, when predicting the performance of system interfaces during the prediction period, the system performance prediction model is calibrated and trained based on system architecture data and the impact of historical concurrent users on system performance, determining the predicted system performance coefficients. Further, based on the predicted system performance coefficients, a system interface verification report is generated, improving the accuracy of system interface verification. When determining the concurrent user number relationship function, it can be determined based on historical concurrent users, historical monthly active baseline users, historical advertising volume, historical weather temperature, and historical rainfall. This accurately describes the impact of historical advertising volume, historical weather temperature, and historical rainfall on historical concurrent users, improving the accuracy and objectivity of the concurrent user number relationship function. When determining the training loss function of a system performance prediction model, it can be based on historical system performance coefficients, sample system performance coefficients, historical concurrent user counts, and historical system interface architecture data. During the calculation process, the impact of the system architecture data and historical concurrent user counts on system performance can be used to determine the influence of the above data on the error of the sample system performance coefficients. Based on this influence and the relative error of the sample system performance coefficients, the training loss function can be set to reduce the training loss function during the training process, thereby more effectively improving the accuracy of the system performance prediction model.

[0054] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only, and are not intended to limit the invention. Other features and aspects of the invention will become clearer from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0055] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other embodiments can be obtained based on these drawings without creative effort.

[0056] Figure 1 An exemplary flowchart of an artificial intelligence-based system interface verification method according to an embodiment of the present invention is shown.

[0057] Figure 2 A schematic diagram illustrating the function for determining the number of concurrent users according to an embodiment of the present invention is shown;

[0058] Figure 3 An exemplary schematic diagram illustrates the prediction of the number of concurrent users for determining the prediction period according to an embodiment of the present invention;

[0059] Figure 4 An exemplary schematic diagram illustrating the process of obtaining a trained system performance prediction model according to an embodiment of the present invention is shown.

[0060] Figure 5 A block diagram of an artificial intelligence-based system interface verification system according to an embodiment of the present invention is shown as an example. Detailed Implementation

[0061] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0062] The technical solution of the present invention will be described in detail below with reference to specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0063] Figure 1An exemplary flowchart illustrates a system interface verification method based on artificial intelligence according to an embodiment of the present invention, the method comprising:

[0064] Step S1: Obtain historical concurrent user counts, historical system performance data, and historical load impact data from multiple historical periods;

[0065] Step S2: Obtain system interface architecture data and historical system interface architecture data;

[0066] Step S3: Obtain real-time load impact data;

[0067] Step S4: Determine the concurrent user count relationship function based on the historical concurrent user count and the historical load impact data;

[0068] Step S5: Determine the predicted concurrent users for the prediction period based on the real-time load impact data and the concurrent user number relationship function;

[0069] Step S6: Based on the historical system interface architecture data, the historical concurrent user count, and the historical system performance data, train the system performance prediction model to obtain the trained system performance prediction model.

[0070] Step S7: Process the predicted concurrent user count and the system interface architecture data according to the trained system performance prediction model to obtain the predicted system performance coefficients.

[0071] Step S8: Generate a system interface verification report based on the predicted system performance coefficients.

[0072] According to an embodiment of the present invention, the system interface verification method based on artificial intelligence can accurately analyze the relationship between load impact data and the number of concurrent users, and then predict the number of concurrent users in the prediction period based on this relationship. When predicting the performance status of the system interface in the prediction period, the system performance prediction model is corrected and trained based on the impact of system architecture data and historical concurrent user count on system performance to determine the predicted system performance coefficients. Furthermore, a system interface verification report is generated based on the predicted system performance coefficients to improve the accuracy of system interface verification.

[0073] According to an embodiment of the present invention, in step S1, the historical concurrent user count, historical system performance data, and historical load impact data in multiple historical periods are obtained.

[0074] For example, if the verification goal of a system interface is to conduct predictive performance testing on the system interface for the next upcoming event day or holiday (such as "Double Eleven"), and predict the interface performance bottleneck under future load, then the peak periods (such as 7 pm to 10 pm) of several past event days or holidays (such as the first day of "Double Eleven" last year and the first day of "Double Eleven" the year before) are used as historical periods. The historical concurrent user count, historical system performance data (such as historical operating system CPU utilization), and historical load impact data (data on external factors that may affect the load during the historical period, such as advertising volume, weather temperature, and rainfall) of the same or similar interfaces to the system interface to be verified are obtained from multiple historical periods.

[0075] According to one embodiment of the present invention, in step S2, system interface architecture data and historical system interface architecture data are obtained.

[0076] For example, during the application of the system, the system architecture will be continuously upgraded. Historical system interface architecture data (such as the historical maximum number of threads in the thread pool, the historical maximum number of connections in the database connection pool, and the historical number of containerized replicas) of the interfaces that are the same or similar to the system interfaces that need to be verified in the historical period, as well as the system interface architecture data of the system interfaces that need to be verified in the current period (such as the maximum number of threads in the thread pool, the maximum number of connections in the database connection pool, and the number of containerized replicas).

[0077] According to one embodiment of the present invention, in step S3, real-time load impact data is acquired.

[0078] For example, obtain load impact data (such as planned advertising volume for Double Eleven, weather forecast information) for a forecast period (e.g., 7 pm to 10 pm on the first day of the next "Double Eleven" event), i.e., real-time load impact data.

[0079] According to an embodiment of the present invention, in step S4, a function relating the number of concurrent users is determined based on the historical concurrent user count and the historical load impact data.

[0080] Figure 2 A schematic diagram illustrating the relationship function for determining the number of concurrent users according to an embodiment of the present invention is shown.

[0081] According to an embodiment of the present invention, step S4 includes:

[0082] Step S41: Obtain the historical monthly active baseline user count for each historical period;

[0083] Step S42: Based on the historical load impact data, determine the historical advertising volume, historical weather temperature, and historical weather rainfall.

[0084] Step S43: Determine the concurrent user count relationship function based on the historical concurrent user count, the historical monthly active baseline user count, the historical advertising volume, the historical weather temperature, and the historical weather rainfall.

[0085] For example, if the first historical period corresponds to 7 PM to 10 PM on November 10, 2024, then the historical monthly active baseline user count for the first historical period is the average daily active baseline user count from 7 PM to 10 PM every day in October 2024. Based on historical load impact data, historical ad placement volume, historical weather temperature, and historical rainfall are determined. Historical ad placement volume refers to the number of ads (in impressions) placed by the system operator before the historical period. Historical weather temperature and historical rainfall are the average weather temperature and rainfall on the dates of the historical period. Historical monthly active baseline users, historical ad placement volume, historical weather temperature, and historical rainfall all influence historical concurrent user counts to some extent. For example, more ads placed attract more new users, resulting in a higher number of concurrent users. Based on the correlation of the above data, a concurrent user count relationship function can be determined.

[0086] According to an embodiment of the present invention, step S43 includes: determining the equation of undetermined coefficients of the concurrent user number relationship function according to formula (1).

[0087] (1)

[0088] in, Let be the historical concurrent users in the i-th historical period. To preset the threshold for the number of concurrent users, Let be the historical monthly active baseline number of users corresponding to the i-th historical period. This represents the historical advertising volume for the i-th historical period. To preset the threshold for ad delivery volume, Let be the historical weather temperature for the i-th historical period. To preset the weather temperature threshold, Let be the historical rainfall for the i-th historical period. To preset the weather rainfall threshold, , , , , , , , and The first undetermined coefficient of the equation of undetermined coefficients;

[0089] Based on the historical concurrent user count, the historical monthly active baseline user count, the historical advertising volume, the historical weather temperature, and the historical weather rainfall, the first undetermined coefficient is solved to obtain the solution value of the first undetermined coefficient;

[0090] Based on the solved values ​​of the first undetermined coefficients and the equation of the undetermined coefficients, the function relating the number of concurrent users is determined.

[0091] According to one embodiment of the present invention, This is the ratio of the historical concurrent users in the i-th historical period to the preset concurrent user threshold. It can be set to 1000 people. This represents the dimensionless historical concurrent user count. This represents the dimensionless historical monthly active user base. This represents the overall trend of the number of concurrent users in the i-th historical period changing relative to the historical monthly active user base in the i-th historical period.

[0092] According to one embodiment of the present invention, in middle, This is the ratio of the historical advertising volume in the i-th historical period to the preset advertising volume threshold. It can be set to 1000 times. This represents the dimensionless historical advertising volume; the larger the ratio, the greater the historical advertising volume. This indicates a positive correlation between the historical concurrent users in the i-th historical period and the historical advertising volume in the i-th historical period. For example, the more advertising volume, the more new users are attracted, and the higher the number of concurrent users. This represents the relative difference between the historical weather temperature of the i-th historical period and the preset weather temperature threshold. The larger this ratio, the more likely the i-th historical period is experiencing high or low temperatures. The preset weather temperature threshold can be set to 18 degrees Celsius. This indicates a positive correlation between the historical concurrent user count in the i-th historical period and the relative difference between the historical weather temperature in the i-th historical period and the preset weather temperature threshold. For example, when the temperature is high or low, the frequency of users going out will decrease, and the frequency of users accessing the system interface will relatively increase. The more concurrent users there are, the higher the frequency of concurrent users will be. This is the ratio of the historical rainfall in the i-th historical period to the preset rainfall threshold, which can be set to 1 mm. This represents dimensionless historical rainfall; the larger the ratio, the greater the historical rainfall. This indicates a positive correlation between the historical concurrent user count over the i-th historical period and the historical rainfall over the i-th historical period. For example, the greater the rainfall and the more severe the weather (such as heavy rain), the less frequently users go out, and the more frequently they access the system interface, resulting in a higher number of concurrent users. Based on this correlation, the equation for the undetermined coefficients of the concurrent user count relationship function can be obtained.

[0093] According to one embodiment of the present invention, the equation involving the undetermined coefficients can be fitted based on multiple parameters, namely, based on historical concurrent users, historical monthly active baseline users, historical advertising volume, historical weather temperature, and historical weather rainfall, to solve for the aforementioned multiple first undetermined coefficients. There are nine first undetermined coefficients, namely... , , , , , , , and Based on the historical concurrent user count, historical monthly active baseline user count, historical advertising volume, historical weather temperature, and historical weather rainfall over at least nine historical periods, the above nine first undetermined coefficients are solved to obtain the solution values ​​of the above nine first undetermined coefficients. The solution values ​​of the above nine first undetermined coefficients are then substituted into the undetermined coefficient equation to determine the concurrent user count relationship function.

[0094] In this way, the relationship function of concurrent users can be determined based on historical concurrent user count, historical monthly active baseline user count, historical advertising volume, historical weather temperature, and historical rainfall. This accurately describes the impact of historical advertising volume, historical weather temperature, and historical rainfall on historical concurrent user count, improving the accuracy and objectivity of the relationship function of concurrent user count.

[0095] According to an embodiment of the present invention, in step S5, the predicted concurrent user count for the prediction period is determined based on the real-time load impact data and the concurrent user count relationship function.

[0096] Figure 3 An exemplary schematic diagram illustrates the prediction of the number of concurrent users for a prediction period according to an embodiment of the present invention.

[0097] According to an embodiment of the present invention, step S5 includes:

[0098] Step S51: Obtain the current active baseline number of users corresponding to the prediction period;

[0099] Step S52: Determine the planned advertising volume based on the real-time load impact data;

[0100] Step S53: Determine weather forecast data based on the real-time load impact data;

[0101] Step S54: Determine the predicted weather temperature and predicted weather rainfall based on the weather forecast data;

[0102] Step S55: Determine the predicted concurrent users for the prediction period based on the current active baseline user count, the planned advertising volume, the predicted weather temperature, the predicted weather rainfall, and the relationship function of the concurrent user count.

[0103] For example, when the prediction period is from 7 PM to 10 PM on the first day of the "Double Eleven" event in 2025, the following steps are taken: First, the average monthly active users of the system interface for the previous month are obtained as the current active baseline user count for that prediction period. Second, the operation plan of the system interface operator is obtained, and the planned advertising volume is determined based on the operation plan. Third, weather forecast data for the corresponding date in the prediction period is obtained through meteorological software, and the temperature and rainfall for that date are determined based on the weather forecast data. Fourth, the predicted temperature and predicted rainfall are then substituted into the concurrent user count relationship function to determine the predicted concurrent user count for the prediction period.

[0104] According to an embodiment of the present invention, in step S6, the system performance prediction model is trained based on the historical system interface architecture data, the historical concurrent user count, and the historical system performance data to obtain the trained system performance prediction model.

[0105] For example, a system performance prediction model is a type of neural network model, which includes: a data preprocessing and input layer, a feature extraction layer, a feature fusion layer, and a decision and output layer. The system performance prediction model is trained using historical system interface architecture data, historical concurrent user counts, and historical system performance data to obtain a trained system performance prediction model, which can then predict the system performance under load conditions.

[0106] Figure 4 An exemplary schematic diagram illustrating the process of obtaining a trained system performance prediction model according to an embodiment of the present invention is shown.

[0107] According to an embodiment of the present invention, step S6 includes:

[0108] Step S61: Determine the historical system performance coefficients based on the historical system performance data;

[0109] Step S62: Process the historical concurrent user count and the system interface architecture data according to the historical system performance prediction model to obtain the sample system performance coefficient. The historical system interface architecture data includes: the historical maximum number of threads in the thread pool, the historical maximum number of connections in the database connection pool, and the historical number of containerized replicas.

[0110] Step S63: Determine the training loss function of the system performance prediction model based on the historical system performance coefficients, the sample system performance coefficients, the historical concurrent user counts, and the historical system interface architecture data;

[0111] Step S64: Train the system performance prediction model according to the training loss function of the system performance prediction model to obtain the trained system performance prediction model.

[0112] For example, based on historical system performance data, the system performance under various system architecture and load conditions is evaluated to determine historical system performance coefficients. The historical concurrent user count and system interface architecture data are processed using the historical system performance prediction model to obtain sample system performance coefficients. Here, the maximum number of threads in the historical thread pool represents the server's ability to process requests simultaneously during a historical period; the maximum number of connections in the historical database connection pool represents the number of database connections allowed to be established by the server during a historical period; and the number of historical containerized replicas (e.g., the number of replicas) represents the number of service instances in the system during a historical period. Based on the historical system performance coefficients, sample system performance coefficients, historical concurrent user counts, and historical system interface architecture data, the training loss function of the system performance prediction model is determined. The system performance prediction model is then trained using this training loss function to improve its accuracy in predicting system performance, resulting in a trained system performance prediction model.

[0113] According to an embodiment of the present invention, step S61 includes:

[0114] Step S611: Based on the historical system performance data, determine the historical response time, historical application instance indicators, historical operating system CPU utilization, and historical database connection count.

[0115] Step S612: Determine the historical user perception layer performance coefficient based on the historical response time;

[0116] Step S613: Determine the historical application layer performance coefficients based on the historical application instance metrics;

[0117] Step S614: Determine the historical system resource layer performance coefficient based on the historical operating system CPU utilization rate;

[0118] Step S615: Determine the historical downstream dependency layer performance coefficient based on the historical database connection count;

[0119] Step S616: Determine the historical system performance coefficient based on the historical user perception layer performance coefficient, the historical application layer performance coefficient, the historical system resource layer performance coefficient, and the historical downstream dependency layer performance coefficient.

[0120] For example, based on historical system performance data from multiple historical periods, determine the historical response time (e.g., P99), historical application instance metrics (e.g., heap memory utilization), historical operating system CPU utilization, and historical database connection count for multiple historical periods. Based on the historical response time, determine the historical user perception layer performance coefficient. For instance, when the P99 quantile value for a historical period remains stable at a low level without abnormal fluctuations, it indicates that the user perception layer is functioning normally, and the historical user perception layer performance coefficient at that moment is 3. When the P99 response time shows a slow but continuous increase, it indicates an impending bottleneck in the system, and the historical user perception layer performance coefficient at that moment is 2. When the P99 quantile value rises sharply, it indicates a system failure, and the historical user perception layer performance coefficient at that moment is... The historical user-perceived layer performance coefficient is 1. The historical user-perceived layer performance coefficient for a historical period is the minimum of the historical user-perceived layer performance coefficients at multiple points within that historical period. Based on historical application instance metrics, the historical application layer performance coefficient is determined. For example, when heap memory utilization fluctuates regularly within a certain range (e.g., 40%-70%), exhibiting a sawtooth pattern (memory is reclaimed after GC), it indicates that the application code and framework are in normal condition, and the historical application layer performance coefficient at that moment is 20. If heap memory utilization remains consistently high (e.g., consistently exceeding 90%), frequently reaches 100%, or the post-GC reclamation effect is poor, and the curve continues to climb, it indicates that the application code and framework are in abnormal condition. The historical application layer performance coefficient at this moment is 10, and the historical application layer performance coefficient for a historical period is the minimum of the historical application layer performance coefficients at multiple moments within that historical period. The historical system resource layer performance coefficient is determined based on the historical operating system CPU utilization. For example, if the historical operating system CPU utilization is higher than 80%, it indicates that computing resources are strained and the server is experiencing an anomaly; the historical system resource layer performance coefficient at this moment is 100. Conversely, if the CPU utilization is lower, it indicates that the server is functioning normally; the historical system resource layer performance coefficient at this moment is 200, and the historical system resource layer performance coefficient for a historical period is the minimum of the historical system resource layer performance coefficients at multiple moments within that historical period. The historical database connection count is used to determine the historical... The downstream dependency layer performance coefficient is used as an example. When the number of database connections approaches or reaches the maximum connection limit, it indicates that the database performance has reached a bottleneck. The historical downstream dependency layer performance coefficient at this moment is 1000. Conversely, it indicates that the database is in normal condition, and the historical downstream dependency layer performance coefficient at this moment is 2000. The historical system performance coefficient is determined by summing the historical user perception layer performance coefficient, historical application layer performance coefficient, historical system resource layer performance coefficient, and historical downstream dependency layer performance coefficient. The specific abnormal situation can be determined by the digits of the units, tens, hundreds, and thousands digits of the historical system performance coefficient. For example, when the digit of the thousands digit of the historical system performance coefficient is 1, it indicates that the database performance has reached a bottleneck.

[0121] According to an embodiment of the present invention, step S63 includes: determining the training loss function of the system performance prediction model according to formula (2). ,

[0122] (2)

[0123] in, Let be the historical system performance coefficient for the i-th historical period. Let be the sample system performance coefficients for the i-th historical period. Let be the historical concurrent users in the i-th historical period. To preset the threshold for the number of concurrent users, The maximum number of threads in the historical thread pool for the i-th historical period. To preset the thread count threshold, The maximum number of connections in the historical database connection pool for the i-th historical period. To preset the maximum number of connections threshold, Let be the number of containerized replicas of the history in the i-th historical period. The preset replica count threshold is denoted as n, where n is the number of historical periods, i ≤ n, and both i and n are positive integers.

[0124] According to one embodiment of the present invention, This represents the relative difference between the maximum number of threads in the historical thread pool during the i-th historical period and a preset thread count threshold. The preset thread count threshold can be set to 80. The larger this ratio is, the more or less the maximum number of threads in the historical thread pool during the i-th historical period is either too large or too small. This is the ratio of the historical concurrent users in the i-th historical period to the preset concurrent user threshold. This can be set to 1000 users. The larger this ratio, the more concurrent users there will be in the i-th historical period. This represents the relative difference between the maximum number of connections in the historical database connection pool during the i-th historical period and the preset maximum connection threshold. The preset maximum connection threshold can be set to 200. The larger this ratio, the more excessive or insufficient the maximum number of connections in the historical database connection pool is. This represents the relative difference between the number of historical containerized replicas in the i-th historical period and a preset replica count threshold. The preset replica count threshold can be set to 4. The larger this ratio, the more likely the number of historical containerized replicas in the i-th historical period is to be too large or too small. The relative differences between the historical maximum number of threads in the thread pool and the preset thread count threshold, the historical number of concurrent users, the historical maximum number of database connection pool connections and the preset maximum number of connection threshold, and the historical number of containerized replicas and the preset replica threshold are negatively correlated with the performance coefficient of the sample system. For example, the larger the relative difference between the historical maximum number of threads in the thread pool and the preset thread count threshold, the more likely the historical maximum number of threads in the thread pool is too small or too large. When the historical maximum number of threads in the thread pool is too small, if a large number of requests flood in simultaneously, the thread pool cannot create more threads to handle them, causing requests to be rejected immediately or queued for a long time, resulting in worse system performance when dealing with load. The smaller the system performance coefficient, the more memory each thread will consume (mainly the thread stack, typically around 1MB by default) when the historical maximum number of threads in the thread pool is too large. The more threads there are, the more memory is consumed. Simultaneously, thread creation, destruction, and context switching also consume additional CPU resources, leading to poorer system performance under load, and a smaller sample system performance coefficient. Similarly, the larger the historical number of concurrent users, the greater the load the system handles, resulting in relatively poorer system performance (manifested as increased response time, increased CPU utilization, etc.), and a smaller sample system performance coefficient. When the historical maximum number of connections in the database connection pool is less than the preset maximum connection threshold... The greater the relative difference, the smaller or larger the maximum number of connections in the historical database connection pool. When the maximum number of connections in the historical database connection pool is too small, it means that the application server cannot have more threads performing database operations simultaneously, resulting in lower overall system concurrency and worse system performance under load, leading to a smaller performance coefficient for the sample system. When the maximum number of connections in the historical database connection pool is too large, multiple connections operating on the database simultaneously, especially when updating the same data or competing for the same resource, will greatly increase the probability of lock waiting. This will cause a sharp increase in request response time, while throughput will decrease, resulting in worse system performance under load and a smaller performance coefficient for the sample system. The greater the relative difference between the historical containerized replica count and the preset replica count threshold, the more likely the historical containerized replica count is to be too small or too large. When the historical containerized replica count is too small, it cannot handle more user requests simultaneously, leading to poorer system performance under load and a lower performance coefficient for the sample system. Conversely, when the historical containerized replica count is too large, the increased number of replicas makes state synchronization and data consistency more complex and difficult, further worsening system performance under load and resulting in a lower performance coefficient for the sample system. Therefore, items related to the historical maximum thread count, historical concurrent user count, historical maximum database connection count, and historical containerized replica count are placed in the denominator to represent... , , and The smaller the data, the larger the performance coefficient of the sample system, and the greater the impact on the error of the performance coefficient of the sample system.

[0125] According to one embodiment of the present invention, The error between the sample system performance coefficients and the historical system performance coefficients in the i-th historical period is calculated using... The training loss function is obtained by weighted summing the errors between the performance coefficients of the sample system and the historical system performance coefficients. During training, this training loss function is reduced, thereby improving the accuracy of the system performance prediction model in predicting system performance coefficients and ultimately enhancing the overall precision of the system performance prediction model.

[0126] In this way, the training loss function of the system performance prediction model can be determined based on historical system performance coefficients, sample system performance coefficients, historical concurrent user counts, and historical system interface architecture data. During the calculation process, the impact of the system architecture data and historical concurrent user counts on system performance can be used to determine the influence of the above data on the error of the sample system performance coefficients. Based on this influence and the relative error of the sample system performance coefficients, the training loss function can be set, thereby reducing the training loss function during the training process of the system performance prediction model and improving the accuracy of the system performance prediction model in a more targeted manner.

[0127] According to an embodiment of the present invention, in step S7, the predicted number of concurrent users and the system interface architecture data are processed according to the trained system performance prediction model to obtain the predicted system performance coefficients.

[0128] According to one embodiment of the present invention, the predicted concurrent user count and system interface architecture data are processed according to the trained system performance prediction model to obtain the predicted system interface performance status during the prediction period, that is, the predicted system performance coefficient.

[0129] According to one embodiment of the present invention, in step S8, a system interface verification report is generated based on the predicted system performance coefficients.

[0130] For example, specific abnormal situations can be determined based on the digits of the units, tens, hundreds, and thousands digits of the predicted system performance coefficient. For instance, if the digit of the thousands digit of the predicted system performance coefficient is 1, it indicates that the database performance may reach a bottleneck during the prediction period, and a corresponding verification report will be generated to remind staff to perform maintenance and upgrades in advance.

[0131] The AI-based system interface verification method according to embodiments of the present invention can accurately analyze the relationship between load impact data and concurrent user counts, and then predict the concurrent user count for a prediction period based on this relationship. When predicting the performance of the system interface during the prediction period, it uses system architecture data and the impact of historical concurrent user counts on system performance to correct and train the system performance prediction model, determine the predicted system performance coefficients, and further generate a system interface verification report based on the predicted system performance coefficients, thereby improving the accuracy of system interface verification. In determining the concurrent user count relationship function, it can be based on historical concurrent user counts, historical monthly active baseline user counts, historical advertising volume, historical weather temperature, and historical rainfall, accurately describing the impact of historical advertising volume, historical weather temperature, and historical rainfall on historical concurrent user counts, thus improving the accuracy and objectivity of the concurrent user count relationship function. When determining the training loss function of a system performance prediction model, it can be based on historical system performance coefficients, sample system performance coefficients, historical concurrent user counts, and historical system interface architecture data. During the calculation process, the impact of the system architecture data and historical concurrent user counts on system performance can be used to determine the influence of the above data on the error of the sample system performance coefficients. Based on this influence and the relative error of the sample system performance coefficients, the training loss function can be set to reduce the training loss function during the training process, thereby more effectively improving the accuracy of the system performance prediction model.

[0132] Figure 5 An exemplary block diagram of an artificial intelligence-based system interface verification system according to an embodiment of the present invention is shown, the system comprising:

[0133] The historical data module is used to obtain historical concurrent user counts, historical system performance data, and historical load impact data for multiple historical periods.

[0134] The architecture data module is used to obtain system interface architecture data and historical system interface architecture data;

[0135] The impact data module is used to obtain real-time load impact data;

[0136] The relational function module is used to determine the relational function for the number of concurrent users based on the historical concurrent user count and the historical load impact data;

[0137] The predictive user module is used to determine the predicted concurrent user count for the prediction period based on the real-time load impact data and the concurrent user count relationship function.

[0138] The model training module is used to train the system performance prediction model based on the historical system interface architecture data, the historical concurrent user count, and the historical system performance data, so as to obtain the trained system performance prediction model.

[0139] The prediction coefficient module is used to process the predicted concurrent user count and the system interface architecture data according to the trained system performance prediction model to obtain the predicted system performance coefficients.

[0140] The verification report module is used to generate a system interface verification report based on the predicted system performance coefficients.

[0141] This invention can be a method, apparatus, system, and / or computer program product. The computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for performing various aspects of the invention.

[0142] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are merely examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and explained in the embodiments, and any variations or modifications may be made to the implementation of the present invention without departing from the stated principles.

Claims

1. A system interface verification method based on artificial intelligence, characterized in that, include: Obtain historical concurrent user counts, historical system performance data, and historical load impact data across multiple historical periods; Obtain system interface architecture data and historical system interface architecture data; Obtain real-time load impact data; Based on the historical concurrent user count and the historical load impact data, determine the concurrent user count relationship function; Based on the real-time load impact data and the concurrent user number relationship function, determine the predicted concurrent user number for the prediction period; Based on the historical system interface architecture data, the historical concurrent user count, and the historical system performance data, the system performance prediction model is trained to obtain the trained system performance prediction model. The predicted concurrent user count and the system interface architecture data are processed based on the trained system performance prediction model to obtain the predicted system performance coefficients. Based on the predicted system performance coefficients, generate a system interface verification report; Based on the historical concurrent user count and the historical load impact data, determine the concurrent user count relationship function, including: Obtain the historical baseline number of monthly active users for each historical period; Based on the historical load impact data, determine the historical advertising volume, historical weather temperature, and historical weather rainfall. The relationship function for concurrent users is determined based on the historical concurrent user count, the historical monthly active baseline user count, the historical advertising volume, the historical weather temperature, and the historical weather rainfall. Based on the historical concurrent user count, the historical monthly active baseline user count, the historical advertising volume, the historical weather temperature, and the historical weather rainfall, a function relating concurrent user count is determined, including: according to the formula: Determine the equation of undetermined coefficients for the function relating concurrent users, where, Let be the historical concurrent users in the i-th historical period. To preset the threshold for the number of concurrent users, Let be the historical monthly active baseline number of users corresponding to the i-th historical period. This represents the historical advertising volume for the i-th historical period. To preset the threshold for ad delivery volume, Let be the historical weather temperature for the i-th historical period. To preset the weather temperature threshold, Let be the historical rainfall for the i-th historical period. To preset the weather rainfall threshold, , , , , , , , and The first undetermined coefficient of the equation of undetermined coefficients; Based on the historical concurrent user count, the historical monthly active baseline user count, the historical advertising volume, the historical weather temperature, and the historical weather rainfall, the first undetermined coefficient is solved to obtain the solution value of the first undetermined coefficient; Based on the solved values ​​of the first undetermined coefficients and the equation of the undetermined coefficients, the function relating the number of concurrent users is determined; Based on the historical system interface architecture data, the historical concurrent user count, and the historical system performance data, a system performance prediction model is trained to obtain a trained system performance prediction model, including: Based on the historical system performance data, determine the historical system performance coefficients; The historical concurrent user count and the system interface architecture data are processed according to the historical system performance prediction model to obtain the sample system performance coefficient. The historical system interface architecture data includes: the historical maximum number of threads in the thread pool, the historical maximum number of connections in the database connection pool, and the historical number of containerized replicas. Based on the historical system performance coefficients, the sample system performance coefficients, the historical concurrent user counts, and the historical system interface architecture data, determine the training loss function for the system performance prediction model; The system performance prediction model is trained according to the training loss function of the system performance prediction model to obtain the trained system performance prediction model. Based on the historical system performance coefficients, the sample system performance coefficients, the historical concurrent user counts, and the historical system interface architecture data, the training loss function for the system performance prediction model is determined, including: according to the formula: Determine the training loss function of the system performance prediction model ,in, Let be the historical system performance coefficient for the i-th historical period. Let be the sample system performance coefficients for the i-th historical period. Let be the historical concurrent users in the i-th historical period. To preset the threshold for the number of concurrent users, The maximum number of threads in the historical thread pool for the i-th historical period. To preset the thread count threshold, The maximum number of connections in the historical database connection pool for the i-th historical period. To preset the maximum number of connections threshold, Let be the number of containerized replicas of the history in the i-th historical period. The preset replica count threshold is denoted as n, where n is the number of historical periods, i ≤ n, and both i and n are positive integers.

2. The system interface verification method based on artificial intelligence according to claim 1, characterized in that, Based on the real-time load impact data and the concurrent user count relationship function, the predicted concurrent user count for the prediction period is determined, including: Obtain the current active baseline number of users corresponding to the prediction period; Based on the real-time load impact data, determine the planned advertising volume; Based on the real-time load impact data, weather forecast data is determined; Based on the aforementioned weather forecast data, the predicted weather temperature and predicted weather rainfall are determined; The predicted concurrent user count for the prediction period is determined based on the relationship function between the current active baseline user count, the planned advertising volume, the predicted weather temperature, the predicted weather rainfall, and the concurrent user count.

3. The system interface verification method based on artificial intelligence according to claim 1, characterized in that, Based on the historical system performance data, the historical system performance coefficients are determined, including: Based on the historical system performance data, determine the historical response time, historical application instance metrics, historical operating system CPU utilization, and historical database connection count; Based on the historical response time, determine the historical user perception layer performance coefficient; Based on the historical application instance metrics, determine the historical application layer performance coefficients; Based on the historical operating system CPU utilization, determine the historical system resource layer performance coefficient; Based on the historical database connection count, determine the historical downstream dependency layer performance coefficient; The historical system performance coefficient is determined based on the historical user perception layer performance coefficient, the historical application layer performance coefficient, the historical system resource layer performance coefficient, and the historical downstream dependency layer performance coefficient.

4. An artificial intelligence-based system interface verification system, used to execute the method according to any one of claims 1-3, characterized in that, include: The historical data module is used to obtain historical concurrent user counts, historical system performance data, and historical load impact data for multiple historical periods. The architecture data module is used to obtain system interface architecture data and historical system interface architecture data; The impact data module is used to obtain real-time load impact data; The relational function module is used to determine the relational function for the number of concurrent users based on the historical concurrent user count and the historical load impact data; The predictive user module is used to determine the predicted concurrent user count for the prediction period based on the real-time load impact data and the concurrent user count relationship function. The model training module is used to train the system performance prediction model based on the historical system interface architecture data, the historical concurrent user count, and the historical system performance data, so as to obtain the trained system performance prediction model. The prediction coefficient module is used to process the predicted concurrent user count and the system interface architecture data according to the trained system performance prediction model to obtain the predicted system performance coefficients. The verification report module is used to generate a system interface verification report based on the predicted system performance coefficients.

Images