An automated inspection method and system based on large model and multi-agent collaboration

Abstract

The present invention provides an automated inspection method and system based on a large model and multi-agent collaboration, wherein the method includes: obtaining the IT resource inspection requirements of the chat user; determining the target agent to be called and the calling strategy corresponding to the target agent according to the IT resource inspection requirements; performing automated inspections in collaboration with multiple agents based on the large model and the calling strategy corresponding to the target agent. The present invention provides an automated inspection method and system based on a large model and multi-agent collaboration, which obtains the IT resource inspection requirements of the chat user through a chat, and determines the target agent to be called and the calling strategy of the target agent; introduces a large model, and controls the target agent to collaboratively perform corresponding inspection tasks based on the calling strategy, thereby realizing collaborative automated inspections, which are more intelligent and have higher IT resource inspection efficiency.

Description

Technical Field

[0001] The present invention relates to the field of automated inspection technology, and in particular to an automated inspection method and system based on a large model and multi-agent collaboration. Background Art

[0002] Large models, also called large pre-trained models or base models, are a type of machine learning model with a large number of parameters and complex computational structures. These models are usually built with deep neural networks and have billions or even hundreds of billions of parameters, designed to improve the expressiveness and predictive performance of the model and to handle more complex tasks and data. Large models provide powerful language understanding and generation capabilities, enabling agents to understand and execute complex instructions.

[0003] Multi-Agent Collaboration is a technology in the field of artificial intelligence that involves multiple agents working together to solve complex problems or complete complex tasks. In a multi-agent system, each agent can be regarded as an independent computing entity that can perceive the environment, make decisions, and perform actions. These agents can be software agents, robots, nodes in sensor networks, etc.

[0004] The invention patent with application number: CN202310614618.6 discloses an automatic network inspection system in a Kubernetes environment, wherein the system includes: a network inspection test module and an inspection result display module; the network inspection test module adopts the following steps during the network inspection test: obtaining and verifying the inspection lock, creating an inspection test environment, executing the inspection network test, writing the inspection results to the cache and pushing them to Prometheus, cleaning the inspection environment, and releasing the inspection lock; the inspection result display module is used to implement the / metrics interface, obtain the network inspection test results from the cache, and when accessing the / metrics interface, call the collect function through Prometheus to return the data to the client. The above invention is based on the Kubernetes environment and realizes an automatic network inspection system, which has the advantages of automated inspection, real-time result display, and comprehensive test cases, and improves network stability and reliability.

[0005] However, the above-mentioned existing technology can only perform automatic network inspection, and the inspection task is single. When the user has other inspection needs, manual operation is still required. It is not intelligent enough and the inspection efficiency is relatively low.

[0006] In view of this, there is an urgent need for an automated inspection method and system based on a large model and multi-agent collaboration to at least solve the above-mentioned deficiencies. Summary of the invention

[0007] One of the purposes of the present invention is to provide an automated inspection method and system based on a large model and multi-agent collaboration, which obtains the IT resource inspection needs of the chat user through chatting, and determines the target agent to be called and the calling strategy of the target agent; introduces a large model, and controls the target agent to collaboratively perform corresponding inspection tasks based on the calling strategy, so as to realize collaborative automated inspection, which is more intelligent and has higher IT resource inspection efficiency.

[0008] An embodiment of the present invention provides an automated inspection method based on a large model and multi-agent collaboration, comprising:

[0009] Get the IT resource inspection requirements of chat users;

[0010] According to IT resource inspection requirements, determine the target agent to be called and the calling strategy corresponding to the target agent;

[0011] Based on the big model, multiple agents can collaborate to conduct automated inspections according to the calling strategy corresponding to the target agent.

[0012] Preferably, obtaining the IT resource inspection requirements of the chat user includes:

[0013] Get the chat information of the chat user;

[0014] Determine IT resource inspection requirements based on chat information.

[0015] Preferably, obtaining the chat information of the chat user includes:

[0016] Obtain the historical IT resource viewing records of the chat user for the target time period preset before the current moment, and obtain the inspection content preference based on the historical IT resource viewing records;

[0017] Get chat information based on inspection content preferences;

[0018] Among them, obtaining the historical IT resource viewing records of the chat user for the target time preset before the current moment, and obtaining the inspection content preference according to the historical IT resource viewing records, including:

[0019] Determine the resource type to be viewed based on historical IT resource viewing records;

[0020] Obtain the operation and maintenance analysis model corresponding to the resource type;

[0021] Determine the first viewed resource item in the historical IT resource viewing record;

[0022] Traversing the first viewing resource items in sequence, and based on a preset second viewing resource item extraction template, attempting to extract the second viewing resource item of the first viewing resource item being traversed;

[0023] If the extraction attempt fails, determine the first viewing feature;

[0024] If the extraction attempt is successful, determining the second viewing feature;

[0025] The first viewing feature and the second viewing feature are sequentially input into the operation and maintenance analysis model in the order of feature extraction time to obtain the inspection content preference.

[0026] Preferably, according to the IT resource inspection requirements, determining the target agent to be called and the calling strategy corresponding to the target agent includes:

[0027] Constructing an agent determination information base; the agent determination information base includes: one-to-one correspondence between pre-selected agents and inspection description information, and the pre-selected agents include: server agent, database agent, dial test agent and log agent;

[0028] Analyze IT resource inspection requirements and determine the target agent and the corresponding call strategy based on the inspection description information.

[0029] Preferably, constructing an agent to determine the information base includes:

[0030] Get the agent type of the pre-selected agent;

[0031] Determine inspection targets based on agent type;

[0032] Determine the inspection items for inspection targets;

[0033] Get the startup description semantics of the inspection project;

[0034] According to the historical inspection records of the inspection items, the automated execution script of the inspection items of the inspection targets is obtained;

[0035] An agent determination information library is built based on the pre-selected agents, the startup description semantics of the inspection items, and the automated execution scripts.

[0036] Preferably, obtaining the automated execution script of the inspection item of the inspection target according to the historical inspection records of the inspection item includes:

[0037] Get the inspection type of historical inspection records; inspection types include: automatic inspection and manual inspection;

[0038] If the inspection type is automatic inspection, obtain the preset execution program of the automatic inspection process according to the corresponding historical inspection records and use it as the automatic execution script;

[0039] If the inspection type is manual inspection, parse the corresponding historical inspection records to obtain the inspection steps and the record items of each inspection step;

[0040] Determine the inspection template and the order identification of the inspection template according to the inspection steps and record items;

[0041] Get the sub-script of the inspection template and the script interface of the sub-script;

[0042] Get the automation execution script based on the sequence identifier, sub-script and script interface.

[0043] An automatic inspection method based on a large model and multi-agent collaboration is provided in an embodiment of the present invention, and further includes:

[0044] Visualize the inspection results of automated inspections.

[0045] An automatic inspection method based on a large model and multi-agent collaboration is provided in an embodiment of the present invention, and further includes:

[0046] Get the inspection log file;

[0047] Obtain error-level logs based on the inspection log file;

[0048] According to the error level log, analyze the causes of inspection errors and make adjustments.

[0049] An automatic inspection method based on a large model and multi-agent collaboration is provided in an embodiment of the present invention, and further includes:

[0050] Get the current inspection item;

[0051] Dynamically determine the risk value of the inspection item based on the historical warning information of the inspection target;

[0052] Compare the current inspection items and the execution paths of the inspection items in order from large to small risk values;

[0053] If the comparison results meet the supplementary inspection conditions, the corresponding inspection items will be used as supplementary inspection items, and supplementary inspections will be conducted on the supplementary inspection items;

[0054] Among them, the additional inspection conditions include:

[0055] The path node sequence of the execution path of the current inspection item is the same as the path node sequence of the execution path of the supplementary inspection item;

[0056] The execution path similarity value of the current inspection item and the supplementary inspection item is greater than or equal to a preset execution path similarity value threshold;

[0057] The risk value corresponding to the supplementary inspection item is greater than or equal to the preset risk value threshold.

[0058] Preferably, the risk value of the inspection item is dynamically determined based on the historical warning information of the inspection target, including:

[0059] Based on the preset warning event feature extraction template, the warning event features are extracted according to the historical warning information. The warning event features include: the time when the warning event occurred and the loss caused by the warning event;

[0060] Determine the severity value of the warning event based on the loss caused by the warning event;

[0061] Determine the correction coefficient of the severity value according to the time when the warning event occurs;

[0062] Determine the severity of the warning event based on the correction factor and severity value;

[0063] The sum of the severity associated with the historical warning information corresponding to the inspection item is calculated to obtain the risk value of the inspection item.

[0064] An embodiment of the present invention provides an automated inspection system based on a large model and multi-agent collaboration, comprising:

[0065] IT resource inspection demand acquisition subsystem, used to obtain the IT resource inspection demand of the chat user;

[0066] The calling strategy determination subsystem is used to determine the target agent to be called and the calling strategy corresponding to the target agent according to the IT resource inspection requirements;

[0067] The automated inspection subsystem is used to perform automated inspections with multiple agents in coordination based on the large model and the call strategy corresponding to the target agent.

[0068] Preferably, the IT resource inspection demand acquisition subsystem includes:

[0069] A chat information acquisition module is used to acquire chat information of chat users;

[0070] The IT resource inspection demand determination module is used to determine the IT resource inspection demand based on the chat information.

[0071] Preferably, the calling strategy determination subsystem includes:

[0072] An agent determination information base construction module is used to construct an agent determination information base; the agent determination information base includes: one-to-one corresponding pre-selected agents and inspection description information, and the pre-selected agents include: server agent, database agent, dial test agent and log agent;

[0073] The calling strategy determination module is used to analyze the IT resource inspection requirements and determine the target agent and the calling strategy corresponding to the target agent according to the inspection description information.

[0074] The beneficial effects of the present invention are:

[0075] The present invention obtains the IT resource inspection demand of the chat user through the chat mode, and determines the target agent to be called and the calling strategy of the target agent; introduces a large model, and controls the target agent to collaboratively perform corresponding inspection tasks based on the calling strategy, so as to realize collaborative automatic inspection, which is more intelligent and has higher IT resource inspection efficiency.

[0076] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in the present application documents.

[0077] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0079] Figure 1 A schematic diagram of an automated inspection method based on a large model and multi-agent collaboration in an embodiment of the present invention;

[0080] Figure 2 The figure is a schematic diagram of an automated inspection system based on a large model and multi-agent collaboration in an embodiment of the present invention. DETAILED DESCRIPTION

[0081] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0082] The embodiment of the present invention provides an automated inspection method based on a large model and multi-agent collaboration, such as Figure 1 As shown, including:

[0083] Obtain the IT resource inspection requirements of chat users; chat users are users who use chat to inspect business systems, servers, databases and other IT resources; IT resource inspection requirements are users' requirements for IT resource inspection and maintenance, such as health status, performance indicators and security status of servers, network equipment, software applications, etc., which are obtained by parsing the chat information of chat users on the chat port;

[0084] According to the IT resource inspection requirements, determine the target agent to be called and the calling strategy corresponding to the target agent; the target agent is: the intelligent agent used to perform the inspection task determined according to the user's IT resource inspection requirements; the calling strategy is: how to use or call the target agent;

[0085] Based on the big model, according to the call strategy corresponding to the target agent, multiple agents are coordinated to perform automated inspections. The big model is an AI model that controls multiple agents to collaboratively perform corresponding inspection tasks.

[0086] The working principle and beneficial effects of the above technical solution are:

[0087] In the present invention, the chat user needs to enter the corresponding command in the input box, for example: "Detect the operation status of the OA business system!", and the system will automatically start to inspect the OA business system and its associated servers and databases. The inspection process for detecting the operation status of the OA business system is as follows:

[0088] 1) Check the server's CPU utilization, memory utilization, and disk utilization indicators;

[0089] 2) Check the number of database connections, QPS, and disk usage indicators;

[0090] 3) Detect the access success rate and access delay indicators of the business system;

[0091] 4) Analyze the log files of the business system.

[0092] The chat port displays, for example:

[0093] AI: OK, we will start automatically inspecting the operation of the OA business system and will perform the following operations:

[0094] Check the server's CPU utilization, memory utilization, and disk utilization indicators.

[0095] Check the number of database connections, QPS, and disk usage.

[0096] Detect the access success rate and access delay indicators of the OA business system.

[0097] Analyze the log files of the OA business system.

[0098] The server's operation status in the last hour is as follows:

[0099] CPU Utilization: 43.2%

[0100] Memory Utilization: 36.5%

[0101] Disk usage: 55.45%; The above indicators show that the current server is running normally.

[0102] The database operation status in the last hour is as follows:

[0103] Number of connections: 24

[0104] QPS:208 times / s

[0105] Disk usage: 32.3%; The above indicators show that the current database is running normally.

[0106] The dial test results of the OA business system in the last hour are as follows:

[0107] Access success rate: 100%

[0108] Access delay: 34ms; The above indicators show that the OA business system access is normal.

[0109] The present invention obtains the IT resource inspection demand of the chat user through the chat mode, and determines the target agent to be called and the calling strategy of the target agent; introduces a large model, and controls the target agent to collaboratively perform corresponding inspection tasks based on the calling strategy, so as to realize collaborative automatic inspection, which is more intelligent and has higher IT resource inspection efficiency.

[0110] In one embodiment, obtaining the IT resource inspection requirements of the chat user includes:

[0111] Obtaining chat information of a chat user; wherein the chat information is: chat information of the chat user on the chat port;

[0112] Determine IT resource inspection requirements based on chat information, where the IT resource inspection requirements are: semantics of the chat information obtained by semantic analysis technology.

[0113] The working principle and beneficial effects of the above technical solution are:

[0114] The present invention obtains the inspection requirements of users by chatting, which is more intelligent.

[0115] In one embodiment, obtaining chat information of a chat user includes:

[0116] Obtain the chat user's historical IT resource viewing records for the target duration preset before the current time, and obtain the inspection content preference based on the historical IT resource viewing records; wherein the preset target duration is manually preset, such as: 1 hour; the historical IT resource viewing records are: records of viewing IT resources within the target duration before the current time; the inspection content preference is the inspection items that the chat user may want to perform predicted based on the historical IT resource viewing records;

[0117] Obtain chat information based on inspection content preferences; obtaining chat information based on inspection content preferences means that before a chat user initiates a chat, the inspection content preferences predicted based on historical IT resource viewing records will be recommended to the user at the top of the dialog window, and the user can select the recommended inspection target or enter the chat content by himself;

[0118] Among them, obtaining the historical IT resource viewing records of the chat user for the target time preset before the current moment, and obtaining the inspection content preference according to the historical IT resource viewing records, including:

[0119] Determine the resource type to be viewed based on the historical IT resource viewing records; the resource type to be viewed is: the resource type corresponding to the historical IT resource viewing records, such as: server, network equipment, etc.;

[0120] Obtain an operation and maintenance analysis model corresponding to the resource type being viewed; wherein the operation and maintenance analysis model is a neural network model trained based on records of operation and maintenance analysis performed by manually viewing the corresponding type of IT resources;

[0121] Determine the first viewed resource item in the historical IT resource viewing record; wherein the first viewed resource item is: data items of all IT resources viewed by the chat user obtained by parsing the historical IT resource viewing record;

[0122] The first viewing resource item is traversed sequentially, and based on a preset second viewing resource item extraction template, a second viewing resource item of the first viewing resource item being traversed is attempted to be extracted; wherein the second viewing resource item extraction template is used to constrain the extraction of the second viewing resource item that is viewed immediately after the first viewing resource item, where immediately after means that the viewing time interval is less than a preset time length threshold, such as: 5 seconds;

[0123] If the extraction attempt fails, determining a first viewing feature; wherein the first viewing feature is: a viewing resource data value of the first viewing resource item;

[0124] If the extraction attempt is successful, determining a second viewing feature; wherein the second viewing feature is: a viewing resource data value of the first viewing resource item and a viewing resource data value of a second viewing resource item viewed immediately after the first viewing resource item;

[0125] The first viewing feature and the second viewing feature are sequentially input into the operation and maintenance analysis model in the order of feature extraction time to obtain the inspection content preference.

[0126] The working principle and beneficial effects of the above technical solution are:

[0127] Generally, before conducting an inspection, a chat user may browse the inspection resources of interest first, and then conduct an intelligent and automated inspection. In this case, the previous browsing information will contain inspection tendency information, which can timely recommend the user's possible resource inspection needs to the chat user, making it more user-friendly.

[0128] Specifically, the resource type to be viewed is determined based on the historical IT resource viewing records. The operation and maintenance analysis logics for different resource viewing types are different, so an operation and maintenance analysis model corresponding to the resource viewing type is introduced. The operation and maintenance analysis model is a neural network model trained based on the records of the operation and maintenance analysis of the corresponding type of IT resources viewed manually, and the corresponding inspection items can be inferred based on the viewing content. Determine the first viewing resource item, which is the data item of all IT resources viewed by the user; traverse the first viewing resource item in sequence, obtain the second viewing resource item viewed immediately after the first viewing resource item, if the attempt to obtain fails, obtain the viewing resource data value of the first viewing resource item, if the attempt to obtain succeeds, obtain the viewing resource data value of the first viewing resource item and the viewing resource data value of the second viewing resource item viewed immediately after the first viewing resource item, and input the extracted data into the operation and maintenance analysis model in chronological order. The operation and maintenance analysis model can learn the independent data features of the first viewing feature and the combined data features of the second viewing feature. At the same time, the order of viewing feature input also includes the time series information of the viewing data, accurately predicting the inspection items that the chat user wants to perform in the future, and improving the accuracy of obtaining the inspection content preference.

[0129] In one embodiment, according to the IT resource inspection requirement, determining the target agent to be called and the calling strategy corresponding to the target agent includes:

[0130] Construct an agent determination information base; the agent determination information base includes: one-to-one correspondence between pre-selected agents and inspection description information, the pre-selected agents include: server agent, database agent, dial test agent and log agent; wherein, the pre-selected agent is: an intelligent agent designed to perform various inspection tasks in the system; the inspection description information is: the descriptive semantics of the corresponding project inspection through the pre-selected agent, for example: the inspection description information of the dial test agent includes: "Please check the reachability of the network connection", "Please check the quality of the network connection", etc.;

[0131] Analyze IT resource inspection requirements, and determine the target agent and the call strategy corresponding to the target agent according to the inspection description information. When determining the call strategy corresponding to the target agent, the inspection items of the target agent are determined according to the inspection description information, and the triggering scheme that triggers the target agent to execute the corresponding inspection items is used as the call strategy.

[0132] The working principle and beneficial effects of the above technical solution are:

[0133] The present invention constructs an agent determination information library, which includes: intelligent agents designed to perform various inspection tasks in the system and corresponding descriptive semantics for performing corresponding project inspections through pre-selected agents; parsing IT resource inspection requirements, determining the target agent for the inspection task to be executed through semantic matching technology based on the inspection description information, and determining the triggering scheme as the calling strategy based on the inspection items to be executed by the required target agent, thereby improving the suitability of the calling strategy determination process.

[0134] In one embodiment, the build agent determines the information repository, including:

[0135] Obtaining the agent type of the pre-selected agent; wherein the agent type is determined according to the resource type of the IT resource that the agent checks, such as server, database, etc.;

[0136] Determine the inspection target based on the agent type; the inspection target is: IT resources of various resource types;

[0137] Determine the inspection items of the inspection target; the inspection items are: specific inspection items of IT resources of various resource types, such as: CPU load, disk space and service response time, etc.;

[0138] Obtain the startup description semantics of the inspection project; wherein the startup description semantics is: the description semantics of the startup inspection project;

[0139] According to the historical inspection records of the inspection items, the automatic execution script of the inspection items of the inspection targets is obtained; wherein the historical inspection records are: the process records of the execution of the inspection items in the past; the automatic execution script is: the script or program for automatically executing the inspection items;

[0140] An agent determination information library is built based on the pre-selected agents, the startup description semantics of the inspection items, and the automated execution scripts.

[0141] The working principle and beneficial effects of the above technical solution are:

[0142] The present invention determines IT resources of corresponding resource types according to the agent type of the pre-selected agent, determines the inspection items of the IT resources; obtains the startup description semantics of the inspection items; introduces the historical inspection records of the inspection items, obtains the automated execution scripts of the inspection items of the IT resources, associates the pre-selected agent, the startup description semantics of the inspection items of the IT resources of corresponding resource types, and the automated execution scripts of the inspection items as the construction records for constructing an agent determination information library, integrates the construction records to obtain the agent determination information library, and makes the construction process more reasonable.

[0143] In one embodiment, obtaining an automated execution script of an inspection item of an inspection target according to historical inspection records of the inspection item includes:

[0144] Get the inspection type of the historical inspection record; the inspection types include: automatic inspection and manual inspection; among them, automatic inspection means: the inspection items recorded in the historical inspection record are automatically executed through a preset program; manual inspection means: the inspection items recorded in the historical inspection record are manually performed according to the prescribed process;

[0145] If the inspection type is automatic inspection, obtain the preset execution program of the automatic inspection process according to the corresponding historical inspection records and use it as the automatic execution script;

[0146] If the inspection type is manual inspection, parse the corresponding historical inspection records to obtain the inspection steps and the record items of each inspection step; wherein the record items are: the inspection data items corresponding to the inspection steps;

[0147] According to the inspection steps and record items, the inspection template and the sequence identifier of the inspection template are determined; wherein the inspection template is: a template that defines what kind of record items need to be recorded in the inspection steps, and is used to standardize the inspection process; the sequence identifier is determined according to the step sequence of the inspection steps;

[0148] Obtain the sub-script of the inspection template and the script interface of the sub-script; wherein the sub-script is: a modular execution script that executes the inspection steps and captures the data of the record items; the script interface is: an interface for the sub-script to interact with the external system or data source;

[0149] According to the sequence identifier, the sub-scripts and the script interface, an automated execution script is obtained, wherein the sub-scripts are integrated according to the script interface based on the sequence identifier to obtain the automated execution script.

[0150] The working principle and beneficial effects of the above technical solution are:

[0151] The present invention introduces the inspection type when acquiring the automated execution script according to the historical inspection records. When the inspection type is automatic inspection, the corresponding execution program is directly acquired as the automated execution script; when the inspection type is manual inspection, the corresponding inspection steps and the record items corresponding to each step are acquired, and the inspection template and the sequence identifier of the template that define what kind of record items need to be recorded for each inspection step are determined; the sub-script of the execution inspection template and the script interface of the sub-script are acquired, and the automated execution script is acquired according to the sequence identifier, the sub-script and the script interface, thereby improving the efficiency and rationality of acquiring the automated execution script.

[0152] The embodiment of the present invention provides an automated inspection method based on a large model and multi-agent collaboration, further comprising:

[0153] Visualize the inspection results of automated inspections, including server operation status, database operation status, business system dialing status, and log file analysis results.

[0154] The working principle and beneficial effects of the above technical solution are:

[0155] The present invention introduces the inspection results of automated inspection for visual display, which is more intuitive.

[0156] The embodiment of the present invention provides an automated inspection method based on a large model and multi-agent collaboration, further comprising:

[0157] Obtain the inspection log file; wherein the inspection log file is: a record file automatically generated by the system during the inspection process;

[0158] Obtain error-level logs according to the inspection log file; error-level logs refer to log entries that record errors or abnormal situations.

[0159] According to the error level log, analyze the causes of inspection errors and make adjustments.

[0160] The working principle and beneficial effects of the above technical solution are:

[0161] The present invention obtains error-level logs in the inspection log file, for example: the error log comes from a Java application, specifically involving the parsing problem of JWT (JSON Web Tokens). The error message indicates that a format problem was encountered when parsing JWT, so the adjustment plan includes: checking the JWT format to ensure that the JWT format sent to the server is correct, especially whether the dot (.) separator between Header, Payload and Signature exists; debugging and logging: adding logging in the JwtUti l s.getPay loadMap method to print out the JWT string that failed to parse, so as to further analyze the problem.

[0162] The embodiment of the present invention provides an automated inspection method based on a large model and multi-agent collaboration, further comprising:

[0163] Get the current inspection item; the current inspection item is: the inspection item that the system is currently inspecting;

[0164] According to the historical warning information of the inspection target, the risk value of the inspection item is dynamically determined; the historical warning information is: the alarms generated in the history of IT resources of the resource type being inspected; the risk value is: the quantitative value of the risk degree of the inspection item;

[0165] The execution paths of the current inspection items and the inspection items are compared in descending order according to the risk values; wherein the execution path is: running the script;

[0166] If the comparison results meet the supplementary inspection conditions, the corresponding inspection items will be used as supplementary inspection items, and supplementary inspections will be conducted on the supplementary inspection items;

[0167] Among them, the additional inspection conditions include:

[0168] The path node sequence of the execution path of the current inspection item is the same as the path node sequence of the execution path of the supplementary inspection item; wherein the path node is: a logical unit or component in the execution path;

[0169] The execution path similarity value of the current inspection item and the supplementary inspection item is greater than or equal to a preset execution path similarity value threshold; wherein the preset execution path similarity value threshold is manually preset;

[0170] The risk value corresponding to the supplementary inspection item is greater than or equal to a preset risk value threshold, wherein the preset risk value threshold is manually preset.

[0171] The working principle and beneficial effects of the above technical solution are:

[0172] This application dynamically calculates the risk value of the inspection item, and compares the current inspection item with the running script of the inspection item in order from large to small risk values. When the comparison result meets the supplementary inspection conditions, the corresponding inspection item is used as a supplementary inspection item and a supplementary inspection is performed;

[0173] The additional inspection conditions include:

[0174] Condition 1: The path node sequence of the execution path of the current inspection project is the same as the path node sequence of the execution path of the supplementary inspection project, that is, the logical units and components of the running code must be completely consistent, meeting the prerequisite for code reuse determination;

[0175] Condition 2: The execution path similarity value of the current inspection item and the supplementary inspection item is greater than or equal to the execution path similarity value threshold, that is, the running code must be similar to improve the accuracy of code reuse judgment;

[0176] Condition 3: The risk value corresponding to the supplementary inspection item is greater than or equal to the preset risk value threshold, that is, the risk value of the inspection item must be high enough to meet the necessity of supplementary inspection.

[0177] Supplementary inspections are carried out when the degree of code reuse is high and the risk value of the items to be supplemented is high enough, which improves the suitability of supplementary inspections.

[0178] In one embodiment, based on the historical warning information of the inspection target, the risk value of the inspection item is dynamically determined, including:

[0179] Based on the preset warning event feature extraction template, the warning event features are extracted according to the historical warning information. The warning event features include: the time when the warning event occurs and the loss caused by the warning event; wherein the preset warning event feature extraction template is: a template that characterizes the historical warning information;

[0180] According to the loss caused by the warning event, the severity value of the warning event is determined; wherein the severity value is quantitatively determined according to the severity of the loss caused by the warning event;

[0181] According to the time when the warning event occurs, determine the correction coefficient of the severity value; where the correction coefficient is: determine the warning time interval from the time when the warning event occurs to the current time, and divide the average warning time interval of the inspection items by the warning time interval to obtain the correction coefficient;

[0182] The severity of the warning event is determined based on the correction coefficient and the severity value; wherein the severity is: the correction coefficient multiplied by the severity value;

[0183] The sum of the severity associated with the historical warning information corresponding to the inspection item is calculated to obtain the risk value of the inspection item.

[0184] The working principle and beneficial effects of the above technical solution are:

[0185] When determining the risk value of an inspection project, the present application introduces a warning event feature extraction template to extract warning event features, and the warning event features include: the time of occurrence of the warning event and the loss caused by the warning event; determining the severity value of the warning event based on the severity of the loss caused by the warning event; determining the warning time interval from the time of occurrence of the warning event to the current time based on the time of occurrence of the warning event, and dividing the average warning time interval of the inspection project by the warning time interval to obtain a correction coefficient; determining the severity of the warning event based on the correction coefficient and the severity value, and calculating the sum of the severity associated with the historical warning information corresponding to the inspection project to obtain the risk value of the inspection project, thereby improving the accuracy of the risk value calculation of the inspection project.

[0186] The embodiment of the present invention provides an automated inspection system based on a large model and multi-agent collaboration, such as Figure 2 As shown, including:

[0187] IT resource inspection demand acquisition subsystem 1, used to obtain the IT resource inspection demand of the chat user;

[0188] The calling strategy determination subsystem 2 is used to determine the target agent to be called and the calling strategy corresponding to the target agent according to the IT resource inspection requirements;

[0189] The automated inspection subsystem 3 is used to perform automated inspections in collaboration with multiple agents based on the large model and the call strategy corresponding to the target agent.

[0190] In one embodiment, the IT resource inspection requirement acquisition subsystem includes:

[0191] A chat information acquisition module is used to acquire chat information of chat users;

[0192] The IT resource inspection demand determination module is used to determine the IT resource inspection demand based on the chat information.

[0193] In one embodiment, invoking a policy determination subsystem includes:

[0194] An agent determination information base construction module is used to construct an agent determination information base; the agent determination information base includes: one-to-one corresponding pre-selected agents and inspection description information, and the pre-selected agents include: server agent, database agent, dial test agent and log agent;

[0195] The calling strategy determination module is used to analyze the IT resource inspection requirements and determine the target agent and the calling strategy corresponding to the target agent according to the inspection description information.

[0196] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. An automated inspection method based on a large model and multi-agent collaboration, characterized in that: include: Obtain the historical IT resource viewing records of the chat user for the target duration preset before the current moment; Determine the resource type to be viewed based on historical IT resource viewing records; Obtain the operation and maintenance analysis model corresponding to the resource type; Determine the first viewed resource item in the historical IT resource viewing record; Traversing the first viewing resource items in sequence, and based on a preset second viewing resource item extraction template, attempting to extract the second viewing resource item of the first viewing resource item being traversed; If the extraction attempt fails, determining a first viewing feature, the first viewing feature being: a viewing resource data value of the first viewing resource item; If the extraction attempt is successful, determining the second viewing feature; The second viewing characteristic is: the viewing resource data value of the first viewing resource item and the viewing resource data value of the second viewing resource item viewed immediately after the first viewing resource item; The first viewing feature and the second viewing feature are sequentially input into the operation and maintenance analysis model in the order of feature extraction time to obtain the inspection content preference; the operation and maintenance analysis model learns the independent data features of the first viewing feature, the combined data features of the second viewing feature, and the order of viewing feature input to predict the inspection items that the chat user wants to perform later; Get chat information based on inspection content preferences; Determine IT resource inspection requirements based on chat information; According to IT resource inspection requirements, determine the target agent to be called and the calling strategy corresponding to the target agent; Based on the big model, according to the calling strategy corresponding to the target agent, multiple agents can collaborate to conduct automated inspections; Get the current inspection item; Based on the preset warning event feature extraction template, the warning event features are extracted according to the historical warning information. The warning event features include: the time when the warning event occurred and the loss caused by the warning event; Determine the severity value of the warning event based on the loss caused by the warning event; According to the time when the warning event occurs, determine the correction coefficient of the severity value; where the correction coefficient is: determine the warning time interval from the time when the warning event occurs to the current time, and divide the average warning time interval of the inspection items by the warning time interval to obtain the correction coefficient; Determine the severity of the warning event based on the correction factor and severity value; Calculate the sum of the severity associated with the historical warning information corresponding to the inspection item to obtain the risk value of the inspection item; Compare the current inspection items and the execution paths of the inspection items in order from large to small risk values; If the comparison results meet the supplementary inspection conditions, the corresponding inspection items will be used as supplementary inspection items, and supplementary inspections will be conducted on the supplementary inspection items; Among them, the additional inspection conditions include: The path node sequence of the execution path of the current inspection item is the same as the path node sequence of the execution path of the supplementary inspection item; The execution path similarity value of the current inspection item and the supplementary inspection item is greater than or equal to a preset execution path similarity value threshold; The risk value corresponding to the supplementary inspection item is greater than or equal to the preset risk value threshold.

2. The automated inspection method based on a large model and multi-agent collaboration as claimed in claim 1, characterized in that: According to the IT resource inspection requirements, determine the target agent to be called and the calling strategy corresponding to the target agent, including: Constructing an agent determination information base; the agent determination information base includes: one-to-one correspondence between pre-selected agents and inspection description information, and the pre-selected agents include: server agent, database agent, dial test agent and log agent; Analyze IT resource inspection requirements and determine the target agent and the corresponding call strategy based on the inspection description information.

3. The automated inspection method based on a large model and multi-agent collaboration as claimed in claim 2, characterized in that: Build an agent-defined information repository, including: Get the agent type of the pre-selected agent; Determine inspection targets based on agent type; Determine the inspection items for inspection targets; Get the startup description semantics of the inspection project; According to the historical inspection records of the inspection items, the automated execution script of the inspection items of the inspection targets is obtained; An agent determination information library is built based on the pre-selected agents, the startup description semantics of the inspection items, and the automated execution scripts.

4. The automated inspection method based on a large model and multi-agent collaboration as claimed in claim 3, characterized in that: According to the historical inspection records of the inspection items, the automatic execution script of the inspection items of the inspection targets is obtained, including: Get the inspection type of historical inspection records; inspection types include: automatic inspection and manual inspection; If the inspection type is automatic inspection, obtain the preset execution program of the automatic inspection process according to the corresponding historical inspection records and use it as the automatic execution script; If the inspection type is manual inspection, parse the corresponding historical inspection records to obtain the inspection steps and the record items of each inspection step; Determine the inspection template and the order identification of the inspection template according to the inspection steps and record items; Get the sub-script of the inspection template and the script interface of the sub-script; Get the automation execution script based on the sequence identifier, sub-script and script interface.

5. The automated inspection method based on large model and multi-agent collaboration as claimed in claim 1, characterized in that: Also includes: Visualize the inspection results of automated inspections.

6. The automated inspection method based on large model and multi-agent collaboration as claimed in claim 1, characterized in that: Also includes: Get the inspection log file; Obtain error-level logs based on the inspection log file; According to the error level log, analyze the causes of inspection errors and make adjustments.

7. An automated inspection system based on a large model and multi-agent collaboration, characterized in that: include: The IT resource inspection demand acquisition subsystem is used to obtain the historical IT resource viewing records of the chat user for the target time preset before the current moment; Determine the resource type to be viewed based on historical IT resource viewing records; Obtain the operation and maintenance analysis model corresponding to the resource type; Determine the first viewed resource item in the historical IT resource viewing record; Traversing the first viewing resource items in sequence, and based on a preset second viewing resource item extraction template, attempting to extract the second viewing resource item of the first viewing resource item being traversed; If the extraction attempt fails, determining a first viewing feature, the first viewing feature being: a viewing resource data value of the first viewing resource item; If the extraction attempt is successful, determining the second viewing feature; The second viewing characteristic is: the viewing resource data value of the first viewing resource item and the viewing resource data value of the second viewing resource item viewed immediately after the first viewing resource item; The first viewing feature and the second viewing feature are sequentially input into the operation and maintenance analysis model in the order of feature extraction time to obtain the inspection content preference; the operation and maintenance analysis model learns the independent data features of the first viewing feature, the combined data features of the second viewing feature, and the order of viewing feature input to predict the inspection items that the chat user wants to perform later; Get chat information based on inspection content preferences; Determine IT resource inspection requirements based on chat information; The calling strategy determination subsystem is used to determine the target agent to be called and the calling strategy corresponding to the target agent according to the IT resource inspection requirements; The automated inspection subsystem is used to conduct automated inspections with multiple agents in coordination based on the large model and the corresponding calling strategy of the target agent. The automated inspection system based on large models and multi-agent collaboration also performs the following operations: Get the current inspection item; Based on the preset warning event feature extraction template, the warning event features are extracted according to the historical warning information. The warning event features include: the time when the warning event occurred and the loss caused by the warning event; Determine the severity value of the warning event based on the loss caused by the warning event; According to the time when the warning event occurs, determine the correction coefficient of the severity value; where the correction coefficient is: determine the warning time interval from the time when the warning event occurs to the current time, and divide the average warning time interval of the inspection items by the warning time interval to obtain the correction coefficient; Determine the severity of the warning event based on the correction factor and severity value; Calculate the sum of the severity associated with the historical warning information corresponding to the inspection item to obtain the risk value of the inspection item; Compare the current inspection items and the execution paths of the inspection items in order from large to small risk values; If the comparison results meet the supplementary inspection conditions, the corresponding inspection items will be used as supplementary inspection items, and supplementary inspections will be conducted on the supplementary inspection items; Among them, the additional inspection conditions include: The path node sequence of the execution path of the current inspection item is the same as the path node sequence of the execution path of the supplementary inspection item; The execution path similarity value of the current inspection item and the supplementary inspection item is greater than or equal to a preset execution path similarity value threshold; The risk value corresponding to the supplementary inspection item is greater than or equal to the preset risk value threshold.

8. The automated inspection system based on large model and multi-agent collaboration as claimed in claim 7, characterized in that: The call strategy determines the subsystem, including: An agent determination information base construction module is used to construct an agent determination information base; the agent determination information base includes: one-to-one corresponding pre-selected agents and inspection description information, and the pre-selected agents include: server agent, database agent, dial test agent and log agent; The calling strategy determination module is used to analyze the IT resource inspection requirements and determine the target agent and the calling strategy corresponding to the target agent according to the inspection description information.

Images