Data center and failure prediction method, electronic device, medium, program product

Abstract

The application discloses a kind of data center and fault prediction method, electronic equipment, medium, program product, applied to fault diagnosis technical field.Therein, method includes that fault prediction task is divided into data collection subtask, data analysis subtask, fault diagnosis subtask and result generation subtask;Data collection subtask is sent to data collection intelligent agent to gather the operating state data of number server, and original operating state data and data analysis subtask are sent to data analysis intelligent agent to carry out data processing to operating state data, operating state data and fault diagnosis subtask are sent to fault diagnosis intelligent agent to carry out fault diagnosis, finally, fault diagnosis result and result generation subtask are sent to result generation intelligent agent to generate fault prediction result.The application can solve the problem that related technology cannot accurately predict complex fault, and can effectively improve the fault prediction precision of data center.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and in particular to a data center and fault prediction method, electronic equipment, non-volatile storage media, and computer program products. Background Technology

[0002] As data centers continue to grow in scale, fault diagnosis and prediction of servers can quickly identify and resolve potential data center problems, prevent system crashes and data loss, and ensure business continuity and stability.

[0003] To ensure the normal operation of data centers, related technologies rely on artificial intelligence algorithms to quickly predict whether servers in the data center will malfunction. However, this method trains a separate model for each type of failure task, which is highly accurate for predicting single types of server failure tasks, but cannot accurately predict complex failure tasks, and cannot effectively guarantee accurate prediction of data center failures.

[0004] Therefore, effectively improving the accuracy of fault prediction in data centers is a technical problem that needs to be solved by those skilled in the art.

[0005] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this application, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] This invention provides a data center and a fault prediction method, electronic equipment, non-volatile storage medium, and computer program product, which effectively improves the fault prediction accuracy of data centers.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] This invention provides a fault prediction method, comprising:

[0009] The fault prediction task is divided into data collection subtasks, data analysis subtasks, fault diagnosis subtasks, and result generation subtasks. The data collection subtasks are sent to a data collection agent to collect raw operational status data from each server in the data center. When the data collection agent completes its data collection subtask, it sends the raw operational status data and the data analysis subtask to the data analysis agent, which processes the raw operational status data to obtain operational status data that meets preset quality conditions. When the data analysis agent completes its data analysis subtask... The task involves sending the operational status data and the fault diagnosis subtask to the fault diagnosis intelligent agent, enabling the fault diagnosis intelligent agent to perform fault diagnosis on the operational status data. When the fault diagnosis intelligent agent completes the fault diagnosis subtask, it sends the fault diagnosis result and the result generation subtask to the result generation intelligent agent, enabling the result generation intelligent agent to generate a fault prediction result based on the fault diagnosis result and prior knowledge. The data collection intelligent agent, the data analysis intelligent agent, the fault diagnosis intelligent agent, and the result generation intelligent agent are all language model-based agents that autonomously complete their tasks by configuring corresponding functions for each.

[0010] In a first exemplary embodiment, the fault diagnosis intelligent agent includes at least a first fault diagnosis intelligent agent, a second fault diagnosis intelligent agent, and a hybrid fault diagnosis intelligent agent. The data collection intelligent agent includes at least a first data collection intelligent agent and a second data collection intelligent agent. Sending the operating status data and the fault diagnosis sub-task to the fault diagnosis intelligent agent so that the fault diagnosis intelligent agent can perform fault diagnosis on the operating status data includes: sending the first operating status data collected by the first data collection intelligent agent to the first fault diagnosis intelligent agent, sending the second operating status data collected by the second data collection intelligent agent to the second fault diagnosis intelligent agent, and sending the first operating status data and the second operating status data to the hybrid fault diagnosis intelligent agent so that the first fault diagnosis intelligent agent, the second fault diagnosis intelligent agent, and the hybrid fault diagnosis intelligent agent can simultaneously perform fault diagnosis on the operating status data.

[0011] In a second exemplary embodiment, there are multiple fault diagnosis agents. Sending the fault diagnosis result and the result generation subtask to the result generation agent includes: obtaining historical task execution data of each fault diagnosis agent; the historical task execution data includes at least the number of historical tasks, execution parameters of each historical task, behavior parameters of each historical task, and interaction history parameters; determining the decision weight for the corresponding fault diagnosis agent to execute the current task based on the historical task execution data of each fault diagnosis agent; and determining the fault diagnosis result input to the result generation agent based on the fault diagnosis result of each fault diagnosis agent and its corresponding decision weight.

[0012] In a third exemplary implementation, determining the decision weight for each fault diagnosis agent to execute the current task based on its historical task execution data includes: determining the task execution score of the current historical task based on its success rate, execution speed, and resource consumption efficiency; determining the performance factor of each fault diagnosis agent based on its historical task execution score and task difficulty score; determining the trust factor of each fault diagnosis agent based on its interaction success rate, protocol compliance rate, and communication success rate during the completion of its historical tasks; determining the experience value of each fault diagnosis agent based on its total historical task processing volume, the total number of target historical tasks meeting preset complexity conditions, and total running time; and determining the decision weight for executing the current task based on the performance factor, trust factor, experience value, and weight coefficient of each fault diagnosis agent.

[0013] In a fourth exemplary embodiment, the fault diagnosis agent runs multiple tasks simultaneously. The fault diagnosis agent performs fault diagnosis on the running status data, including: when insufficient computing resources and / or memory resources are detected, acquiring the workload information and task running information of the fault diagnosis agent; reading the task type and task impact of each task from the task running information, and determining the urgency score of each task based on the task type and task impact; determining the workload during the execution of each task based on the workload information, and determining the load score based on the workload quantity and workload resource requirements; determining the priority of each task based on the urgency score, load score, and their respective weight coefficients; and performing fault diagnosis on the running status data based on the priority of the fault diagnosis subtasks.

[0014] In a fifth exemplary embodiment, there are multiple fault diagnosis agents, and the result generation agents include expert agents and interactive agents. The result generation subtasks include a diagnosis result generation subtask and a report generation subtask. The fault diagnosis results and the result generation subtasks are sent to the result generation agents so that the result generation agents can generate fault prediction results based on the fault diagnosis results and prior knowledge. This includes: sending the fault diagnosis results of each fault diagnosis agent and the diagnosis result generation subtasks to the expert agents so that the expert agents can generate a total fault diagnosis result based on the fault diagnosis results of each fault diagnosis agent and an expert knowledge base; and when the diagnosis result generation subtasks are completed, sending the report generation subtasks and the total fault diagnosis result to the interactive agents so that the interactive agents can use a learning-based retrieval-enhanced generation method to generate fault prediction results based on the total fault diagnosis result and human interaction information.

[0015] In a sixth exemplary embodiment, the method further includes: acquiring workload and resource consumption information of the data collection agent, the data analysis agent, the fault diagnosis agent, and the result generation agent; determining the resource allocation weights of the data collection agent, the data analysis agent, the fault diagnosis agent, and the result generation agent at the current moment based on the workload and resource consumption information of the data collection agent, the data analysis agent, the fault diagnosis agent, and the result generation agent; and adjusting the resource allocation of the data collection agent, the data analysis agent, the fault diagnosis agent, and the result generation agent based on the current resource allocation weights.

[0016] In a seventh exemplary embodiment, the method further includes: sending the original operating status data to a first target terminal via a first learning interface, so that the first target terminal can check the original operating status data; when receiving feedback that the original operating status data contains errors, receiving the correct original operating status data fed back by the first target terminal via the first learning interface, and sending the first learning task and the correct original operating status data to the data collection agent, and sending the correct original operating status data as original operating status data to the data collection agent; sending the operating status data to a second target terminal via a second learning interface, so that the second target terminal can check the operating status data; when receiving feedback that the operating status data contains errors, receiving the correct operating status data fed back by the second target terminal via the second learning interface, and sending the second learning task and the correct operating status data to the data collection agent, and sending the correct operating status data as original operating status data to the data collection agent. Data is sent as operational status data to the fault diagnosis agent; the fault prediction result is sent to the third target end through the third learning interface, so that the third target end can review the fault prediction result; when feedback information indicating that the fault prediction result is incorrect is received, the correct fault prediction result fed back by the third target end is received through the third learning interface, and the third learning task and the correct fault prediction result are sent to the result generation agent, and the correct fault prediction result is output as the fault prediction result; the fault diagnosis result is sent to the fourth target end through the fourth learning interface, so that the fourth target end can correct the fault diagnosis result based on the chaos engineering self-evolution mechanism; when feedback information indicating that the fault diagnosis result is incorrect is received, the correct fault diagnosis result fed back by the fourth target end is received through the fourth learning interface, and the fourth learning task and the correct fault diagnosis result are sent to the fault diagnosis agent, and the correct fault diagnosis result is sent to the result generation agent.

[0017] The present invention also provides an electronic device including a processor, the processor being configured to implement the steps of any of the aforementioned fault prediction methods when executing a computer program stored in a memory.

[0018] The present invention also provides a non-volatile storage medium storing a computer program, which, when executed by a processor, implements the steps of any of the aforementioned fault prediction methods.

[0019] The present invention also provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of any of the aforementioned fault prediction methods.

[0020] Finally, this invention also provides a data center, including a fault prediction system; the fault prediction system includes a manager intelligent agent, a data collection intelligent agent, a data analysis intelligent agent, a fault diagnosis intelligent agent, an expert intelligent agent, and an interactive intelligent agent; wherein, the manager intelligent agent communicates with the data collection intelligent agent, the data analysis intelligent agent, the fault diagnosis intelligent agent, the expert intelligent agent, and the interactive intelligent agent respectively, and executes the steps of the fault prediction method as described above; the data collection intelligent agent is used to collect the original operating status data of each server in the data center after receiving a data collection sub-task; the data analysis intelligent agent is used to process the original operating status data after receiving a data analysis sub-task to obtain operating status data that meets preset quality conditions; the fault diagnosis intelligent agent is used to perform fault diagnosis on the operating status data after receiving a fault diagnosis sub-task; the expert intelligent agent is used to generate a total fault diagnosis result by integrating the fault diagnosis results of each fault diagnosis intelligent agent after receiving a diagnosis result generation sub-task; the interactive intelligent agent is used to generate a fault prediction result based on the total fault diagnosis result, prior knowledge, and interactive information after receiving a report generation sub-task.

[0021] In a first exemplary embodiment, the data collection agent includes an interface call function, a monitoring tool call function, and a tracking data function; wherein, the interface call function is used to obtain server operation log data by calling a remote control interface; the monitoring tool call function is used to obtain server performance indicator data by calling a monitoring tool interface; and the tracking data function is used to obtain call chain data corresponding to the target identifier implanted in the tracking system.

[0022] In a second exemplary embodiment, the data analysis agent includes data processing functions, log parsing tool invocation functions, and data cleaning tool invocation functions. Specifically, the data analysis agent filters runtime log data shorter than a preset length threshold by invoking the data processing function; it parses the remaining runtime log data using the log parsing tool invocation function, removing invalid and redundant data; it also deletes incomplete call chain data by invoking the data processing function; and it further deletes redundant data from the server performance indicator data by invoking the data processing function, and completes the server performance indicator data using the data cleaning tool invocation function.

[0023] In a third exemplary embodiment, the fault prediction system further includes a knowledge database; the knowledge database includes a structured knowledge sub-database and an unstructured knowledge sub-database; it acquires historical knowledge already learned by the manager agent and / or the data collection agent and / or the data analysis agent and / or the fault diagnosis agent and / or the expert agent and / or the interaction agent, and selects target historical knowledge that is not accessed within a preset time period from the historical knowledge; for the first type of target historical knowledge belonging to the structured data type, the first type of target historical knowledge is converted into vector data and stored in the structured knowledge sub-database; for the second type of target historical knowledge belonging to the unstructured data type, the second type of target historical knowledge is converted into text data and stored in the unstructured knowledge sub-database.

[0024] The advantage of the technical solution provided by this invention lies in decomposing the entire fault prediction process of a data center into sub-tasks, simplifying and optimizing the difficulty of fault prediction for complex computer systems. Each agent is responsible for only one sub-task, which not only reduces the workload of a single agent but also ensures the accuracy and efficiency of task execution, achieving efficient and high-precision fault prediction. Using an advanced large-scale language model as the core engine, multiple agents work collaboratively to accurately respond to complex fault scenarios that may occur during server operation, achieving high-precision prediction of complex fault tasks, effectively ensuring accurate prediction of data center faults, and ensuring the safe, stable, and reliable operation of the data center.

[0025] Furthermore, the present invention also provides corresponding electronic devices, non-volatile storage media, computer program products, and data centers for the fault prediction method, further making the method more practical. The electronic devices, non-volatile storage media, computer program products, and data centers have corresponding advantages.

[0026] The technical features mentioned above, those to be mentioned below, and those shown individually in the accompanying drawings can be arbitrarily combined with each other, provided that the combined technical features are not contradictory. All feasible combinations of features are the technical content explicitly described herein. Any one of the multiple sub-features contained in the same statement can be applied independently, without necessarily being applied together with other sub-features. It should be understood that the above general description and the following detailed description are merely exemplary and do not limit the invention. Attached Figure Description

[0027] To more clearly illustrate the technical solutions of the present invention or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0028] Figure 1 A flowchart illustrating a fault prediction method provided by the present invention;

[0029] Figure 2 A flowchart illustrating another fault prediction method provided by the present invention;

[0030] Figure 3 A schematic diagram of an exemplary architecture for a data center fault prediction system provided by the present invention;

[0031] Figure 4 Another exemplary architecture diagram of the data center fault prediction system provided by the present invention;

[0032] Figure 5 A structural diagram of a specific embodiment of the fault prediction device provided by the present invention;

[0033] Figure 6 This is a structural diagram of a specific embodiment of the electronic device provided by the present invention. Detailed Implementation

[0034] To enable those skilled in the art to better understand the technical solutions of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. In this specification and the aforementioned drawings, the terms "first," "second," "third," "fourth," etc., are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. The term "exemplary" means "serving as an example, embodiment, or illustration." Any embodiment described herein as "exemplary" is not necessarily to be construed as superior to or better than other embodiments.

[0035] As the core infrastructure of computer information systems, data centers are not only crucial hubs for storing and processing massive amounts of data, but also support various critical business applications and services. With the rapid development of artificial intelligence, the Internet of Things, and cloud technologies, data centers are growing in scale, and the requirements for availability and reliability are becoming increasingly stringent. To prevent computer information system crashes and data loss, ensure business continuity and stability, and guarantee the normal operation of data centers, fault diagnosis and prediction are performed on each server. Timely fault diagnosis allows for the early detection of malfunctioning equipment, rapid resolution of potential data center problems, and effective prevention of data center failures. For example, by monitoring and analyzing parameters such as server temperature, voltage, and vibration, potential hardware failures can be predicted, allowing for maintenance and replacement before failures occur, thus avoiding unexpected server downtime and data loss. Furthermore, by predicting server failures in advance, the frequency of emergency repairs and hardware replacements can be reduced, lowering maintenance costs, optimizing resource allocation, and improving energy efficiency and overall operational efficiency. For instance, predictive maintenance systems can automatically adjust workloads based on server health conditions, preventing global performance degradation due to single points of failure. Real-time monitoring of the health of servers and other infrastructure equipment can optimize energy use and reduce unnecessary energy consumption.

[0036] To ensure the normal operation of data centers, traditional methods typically rely on statistical analysis and expert rules for fault diagnosis or prediction, such as threshold-based fault diagnosis methods, statistical model-based fault diagnosis methods, and expert rule-based fault diagnosis methods. Among these, threshold-based fault diagnosis methods predefine a threshold for each performance indicator; if any indicator exceeds this threshold during server operation, the server is considered to have failed.

[0037] This method relies solely on preset thresholds to determine faults, limiting its information utilization and lacking a comprehensive and in-depth analysis of the system's operational status. This leads to a high probability of misdiagnosis or missed faults. Furthermore, setting reasonable thresholds requires extensive experience and expertise, and these thresholds need constant adjustment as the system's operating environment changes, not only failing to guarantee effective fault diagnosis but also making maintenance difficult. Statistical model-based fault diagnosis methods analyze the distribution of server operational data and identify abnormal data points. However, this method depends on assumptions about data distribution, which are complex and variable in real-world applications, making it difficult to meet these assumptions and potentially causing the model to fail. Moreover, a thorough understanding of the system's operating mechanisms and data characteristics is required before model construction, making statistical model building complex and time-consuming. Furthermore, this method experiences significant performance degradation or even malfunction when processing high-dimensional data, limiting its application in diagnosing complex system faults. Expert rule-based fault diagnosis methods transform expert knowledge and experience into rules and utilize these rules to diagnose server faults. While this method can efficiently leverage the deep knowledge and experience of domain experts to quickly locate problems, it suffers from several drawbacks. It struggles to fully capture and express complex, implicit, and context-dependent knowledge, leading to reduced accuracy in fault diagnosis. Furthermore, the rule base is costly to maintain, requiring continuous updates to adapt to technological advancements and environmental changes, and is prone to errors. Conflicts and redundancies between rules are complex and difficult to resolve, affecting the consistency of diagnostic results and further reducing accuracy. Additionally, the application of a fixed rule set makes this method unable to quickly adapt to new situations and unknown faults, potentially leading to misleading results. For fault diagnosis of complex systems, it requires handling intricate rule combinations and logical judgments, resulting in low efficiency, and the system complexity is directly proportional to the computational cost.

[0038] To address the problems of traditional fault diagnosis methods, related technologies utilize artificial intelligence algorithms for server fault prediction and diagnosis. Fault prediction methods based on artificial intelligence algorithms include those based on traditional machine learning and those based on deep learning. Traditional machine learning methods, such as XGboost (Extreme Gradient Boost) and LightGBM (Light Gradient Boosting Machine), are used for fault identification and prediction. While these methods effectively leverage data-driven advantages to improve prediction accuracy, they have high requirements for the quality and quantity of training data. In cases of missing or incomplete data, model performance significantly degrades, leading to inaccurate fault prediction results. Furthermore, these models require retraining or adjustment to maintain performance when facing new fault modes or changing system environments, resulting in poor flexibility and adaptability. In addition, to avoid model performance instability and computational latency, real-time performance and stability of the model need special attention in practical industrial applications, increasing maintenance costs and complexity. Deep learning-based methods, such as convolutional neural networks, long short-term memory networks, and Transformers, are used for fault identification and prediction. While these methods can improve the accuracy and automation of fault prediction, they are also highly dependent on training data. With limited data, model performance may significantly decline. Secondly, the hyperparameter tuning process for deep learning models is complex, requiring extensive experimentation and specialized knowledge, increasing the difficulty of model development and maintenance. Finally, regardless of the specific AI algorithm used, each task requires a separately trained neural network model for fault prediction. As data centers grow in scale and computer system complexity increases, fault prediction tasks become increasingly complex, making accurate prediction of data center faults impossible with these technologies.

[0039] In view of this, in order to solve the problems existing in related technologies, this invention decomposes the entire fault prediction process of a data center into a series of sub-tasks, simplifying and optimizing the difficulty of fault prediction for complex computer systems. Using an advanced large language model as the core engine, it drives multiple intelligent agents to work collaboratively. Each agent is responsible for only one subdivided sub-task, accurately responding to complex fault scenarios that may occur during server operation, achieving high-precision prediction of complex fault tasks, and ensuring the safe, stable, and reliable operation of the data center. After introducing the technical solution of this invention, various non-limiting embodiments of this invention are described in detail below. To better illustrate this invention, numerous specific details are given in the following detailed embodiments. Those skilled in the art should understand that this invention can be implemented without these specific details. In other examples, methods, means, components, and circuits well known to those skilled in the art are not described in detail in order to highlight the main points of this invention.

[0040] Please see first. Figure 1 , Figure 1This is a flowchart illustrating a fault prediction method provided in this embodiment. Based on any existing large language model, and equipped with planning, memory, and tool usage functions, the intelligent agent, leveraging language processing capabilities, autonomously completes tasks, resulting in a manager agent, a data collection agent, a data analysis agent, a fault diagnosis agent, and a result generation agent. These agents, through task planning, memory, and external tools, identify the tasks to be performed and the execution methods, thereby autonomously making decisions to complete the corresponding tasks. Planning refers to the process by which the agent decomposes complex tasks into a series of sub-tasks and formulates corresponding action plans. This not only helps the agent understand the task objectives but also allows for adjustments and optimizations based on feedback during execution, thus efficiently completing the task. The role of planning in the agent includes: dividing complex tasks into multiple sub-tasks for gradual resolution, similar to human problem-solving thinking; self-criticism and reflection based on existing actions, learning from mistakes to optimize subsequent actions; and adjustments and optimizations based on feedback from the external environment during execution to ensure efficient task completion. Memory includes short-term memory and long-term memory. Short-term memory refers to contextual information during task execution, including tool return values ​​and completed reasoning paths. Short-term memory is similar to working memory in the human brain, storing currently processed information and aiding in complex cognitive tasks. Short-term memory is typically limited by the length of the context window, has a limited scope of influence, and a short duration. Long-term memory can store information for a considerable period, ranging from days to decades, with virtually unlimited storage capacity. Long-term memory includes episodic memory and semantic memory, which are memories of facts and events that can be consciously recalled; it also includes unconsciously executed skills and routines. In other words, the manager agent, data collection agent, data analysis agent, fault diagnosis agent, and result generation agent of this invention are all based on language models and are intelligent agents that autonomously complete tasks by configuring corresponding functions for each. Regarding the functions configured for the manager agent, data collection agent, data analysis agent, fault diagnosis agent, and result generation agent, in addition to the functions described in the following embodiments, those skilled in the art can flexibly configure them according to actual application scenarios, which does not affect the implementation of this invention. After obtaining the manager agent, data collection agent, data analysis agent, fault diagnosis agent, and result generation agent, this embodiment uses the manager agent as the subject of execution to complete the fault prediction method. This embodiment may include the following:

[0041] S101: When a fault prediction task is detected, the fault prediction task is divided into a data collection subtask, a data analysis subtask, a fault diagnosis subtask, and a result generation subtask.

[0042] In this embodiment, the fault prediction task can be triggered by a fault prediction request or by a script according to preset rules. This step begins as soon as a fault prediction task is detected in the data center. This embodiment divides the fault prediction task into four subtasks: a data collection subtask, a data analysis subtask, a fault diagnosis subtask, and a result generation subtask. Different subtasks are executed by different intelligent agents. The data collection subtask triggers the data collection agent to perform data collection tasks, the data analysis subtask triggers the data analysis agent to perform data analysis tasks, the fault diagnosis subtask triggers the fault diagnosis agent to perform fault diagnosis tasks, and the result generation subtask triggers the result generation agent to generate prediction results.

[0043] S102: Send the data collection subtask to the data collection agent so that the data collection agent can collect the raw operating status data of each server in the data center.

[0044] This step involves equipping the data collection agent with planning, memory, and tool usage capabilities to complete data collection tasks. Planning refers to how and which tasks to collect. For example, a data collection task can be divided into multiple sub-tasks based on the device being collected, with each sub-task responsible for collecting operational data from one device. The tools configured for the data collection agent include, but are not limited to, external application programming interfaces (APIs), calculators, search tools, code executors, database query tools, and knowledge bases. External APIs allow access to external data, while calculators, search tools, code executors, and database query tools assist the data collection agent in performing calculations, searches, code execution, and database queries to obtain the necessary information and data. Knowledge bases enable rapid information retrieval and storage, achieving long-term retention and recall of information. Raw operational status data includes, but is not limited to, various performance metrics such as CPU utilization, memory usage, and disk I / O performance, as well as server operation logs and data tracked by the tracing system used for invocation.

[0045] S103: When the data collection agent completes the data collection sub-task, it sends the raw operating status data and the data analysis sub-task to the data analysis agent, so that the data analysis agent can process the raw operating status data to obtain operating status data that meets the preset quality conditions.

[0046] This step involves equipping the data analysis agent with planning, memory, and tool usage capabilities to complete data analysis tasks. Planning refers to how the agent should process the task and analyze the data. For example, data analysis tasks can be divided according to the type of data to be analyzed, grouping similar types of data into sub-tasks, each sub-task handling one type of data. The tools configured for the data analysis agent include, but are not limited to, external application programming interfaces (APIs), calculators, search tools, code executors, database query tools, and knowledge bases. APIs allow access to external data, while calculators, search tools, code executors, and database query tools assist the agent in performing calculations, searches, code execution, and database queries to obtain the necessary information and data. Knowledge bases enable rapid information retrieval and storage, achieving long-term information retention and recall. Preset quality conditions are pre-defined to measure the degree to which the data analysis agent processes the original operational data, i.e., the quality of the processed operational data, such as whether it contains redundant data, whether it is complete, whether there are missing data, and the minimum conditions to be met. Those skilled in the art can make flexible determinations based on the actual situation, and the present invention does not impose any limitations in this regard.

[0047] S104: When the data analysis agent completes the data analysis subtask, it sends the running status data and the fault diagnosis subtask to the fault diagnosis agent so that the fault diagnosis agent can perform fault diagnosis on the running status data.

[0048] This step involves equipping the fault diagnosis agent with planning, memory, and tool usage functions to complete fault diagnosis tasks. Planning refers to how the agent should perform fault diagnosis, and what data should be processed using which methods. For example, the fault diagnosis task can be decomposed according to the type of data to be diagnosed, dividing similar types of data into sub-tasks. Each sub-task is responsible for processing one type of data. For each sub-task, the appropriate model or tool for analysis is specified; for example, image data can be analyzed using an image processing model, and text data can be analyzed using a text data model. The tools configured for the fault diagnosis agent include, but are not limited to, external application programming interfaces (APIs), calculators, search tools, code executors, database query tools, and knowledge bases. External APIs can acquire external data, while calculators, search tools, code executors, and database query tools can assist the fault diagnosis agent in performing calculations, searches, code execution, and database queries to obtain the necessary information and data.

[0049] S105: When the fault diagnosis agent completes the fault diagnosis subtask, it sends the fault diagnosis result and the result generation subtask to the result generation agent so that the result generation agent can generate the fault prediction result based on the fault diagnosis result and prior knowledge.

[0050] This step involves equipping the result-generating agent with planning, memory, and tool usage capabilities to perform the task of generating fault prediction results. Planning refers to how the agent should generate prediction results and what prior knowledge it should acquire when performing the result generation task. The tools configured for the result-generating agent include, but are not limited to, external application programming interfaces (APIs), calculators, search tools, code executors, database query tools, and knowledge bases. External APIs allow for the acquisition of external data. Calculators, search tools, code executors, and database query tools assist the data-collecting agent in performing calculations, searches, code execution, and database queries to obtain the necessary information and data. Knowledge bases enable rapid information retrieval and storage, achieving the ability to retain and recall information over long periods. Prior knowledge refers to the external data needed to generate fault prediction results, including, but not limited to, expert experience, fault support records, product documents, and web crawler results.

[0051] In the technical solution provided in this embodiment, the entire fault prediction process of the data center is decomposed into sub-tasks, simplifying and optimizing the difficulty of fault prediction for complex computer systems. Each agent is responsible for only one sub-task, which not only reduces the workload of a single agent but also ensures the accuracy and efficiency of task execution, achieving efficient and high-precision fault prediction. Using an advanced large-scale language model as the core engine, multiple agents work collaboratively to accurately respond to complex fault scenarios that may occur during server operation, achieving high-precision prediction of complex fault tasks. This effectively ensures accurate prediction of data center faults, guaranteeing the safe, stable, and reliable operation of the data center.

[0052] To further improve fault prediction efficiency, based on the above embodiments, the present invention also provides multiple fault diagnosis agents that execute fault diagnosis subtasks in parallel, which may include the following:

[0053] like Figure 2As shown, the fault diagnosis intelligent agent includes at least a first fault diagnosis intelligent agent, a second fault diagnosis intelligent agent, and a hybrid fault diagnosis intelligent agent. The data collection intelligent agent includes at least a first data collection intelligent agent and a second data collection intelligent agent. The first operating status data collected by the first data collection intelligent agent is sent to the first fault diagnosis intelligent agent, the second operating status data collected by the second data collection intelligent agent is sent to the second fault diagnosis intelligent agent, and the first operating status data and the second operating status data are sent to the hybrid fault diagnosis intelligent agent, so that the first fault diagnosis intelligent agent, the second fault diagnosis intelligent agent, and the hybrid fault diagnosis intelligent agent can simultaneously perform fault diagnosis on the operating status data.

[0054] In this embodiment, the first fault diagnosis intelligent agent, the second fault diagnosis intelligent agent, and the hybrid fault diagnosis intelligent agent are configured with programming and communication capabilities. Through programming, they invoke relevant machine learning models; through communication, they explain the reasons for the diagnostic or predictive conclusions, making the final results interpretable. This addresses the problem that related artificial intelligence-based methods are often "black box" models, leading to a lack of interpretability. The first, second, and hybrid fault diagnosis intelligent agents can invoke various algorithms to analyze and predict the collected and cleaned operational status data to determine the existence and type of faults, thereby identifying whether the data center and individual servers are in normal or abnormal operation. The difference lies in that the first and second fault diagnosis intelligent agents analyze, diagnose, and predict single operational status data, while the hybrid fault diagnosis intelligent agent fuses all operational status data and analyzes, diagnoses, and predicts the fused data. If the runtime status data is runtime log data, models such as LightGBM and Long Short-Term Memory networks can be called from the codebase to perform fault judgment and prediction. If the runtime status data is call chain data, graph neural networks can be called from the codebase to perform graph modeling on the call chain data and provide fault diagnosis results. If the runtime status data is performance index data, methods such as Isolation Forest and Standard Deviation can be called from the codebase to judge outliers, and Prophet (a time series prediction model) can be used to diagnose and predict faults. The first fault diagnosis agent, the second fault diagnosis agent, and the hybrid fault diagnosis agent, through a high-efficiency data processing engine, can not only instantly call advanced algorithms to perform deep learning and pattern recognition on massive amounts of data to accurately determine whether a fault has occurred in the data center, but also further identify the fault type, severity, and potential impact range. Fault types, such as CPU overload, memory leak, or disk failure, can be pre-classified into multiple levels and configured with corresponding standards. When the first fault diagnosis agent, the second fault diagnosis agent, and the hybrid fault diagnosis agent detect an abnormal mode, alarm information carrying the fault type can be generated.

[0055] Building upon fault diagnosis, the first, second, and hybrid fault diagnosis agents can also incorporate predictive analytics algorithms. By combining historical fault trends with the current operational status of the data center, they can predict potential future faults. For example, these agents utilize time-series analysis of historical data and current operational status data to predict whether the data center is experiencing performance degradation, thereby enabling proactive preventative measures to avoid faults. Root cause analysis of operational status data, through correlation analysis, identifies the root causes of faults and proposes targeted solutions to ensure high availability and stability of the data center and improve its task execution efficiency.

[0056] Furthermore, the first fault diagnosis agent, the second fault diagnosis agent, and the hybrid fault diagnosis agent are configured with student-teacher planning to enable them to have self-optimization capabilities. Through continuous learning, they improve the accuracy and efficiency of fault diagnosis, enhance prediction accuracy and response speed, and ensure the stability, reliability, and high availability of data center operations.

[0057] Based on the above embodiments, in scenarios where multiple fault diagnosis agents perform fault diagnosis simultaneously, to further improve the accuracy of the final fault prediction result, this invention also assigns importance to each fault diagnosis agent in the fault diagnosis result decision-making process, which may include the following:

[0058] Obtain historical task execution data for each fault diagnosis agent; determine the decision weight for the corresponding fault diagnosis agent to execute the current task based on the historical task execution data of each fault diagnosis agent; determine the input results to generate the fault diagnosis results of the agent based on the fault diagnosis results of each fault diagnosis agent and their corresponding decision weights.

[0059] The historical task execution data may include at least the number of historical tasks, the execution parameters of each historical task, the behavioral parameters of each historical task, and the interaction history parameters. The historical task execution parameters may include at least the task success rate, the task execution speed, and the task resource consumption efficiency. The historical task behavioral parameters may include the number of times the rules were violated, the number of times the protocol was not followed, and the number of times the task was not executed. The interaction history parameters may include the interaction status of each fault diagnosis agent in the process of completing each historical task. For example, the decision weight can be jointly determined by performance factors, trust factors, and experience factors. The task execution score of the current historical task is determined based on the success rate, execution speed, and resource consumption efficiency of the current historical task. The performance factor of each fault diagnosis agent is determined based on the task execution score and task difficulty score of each historical task. The trust factor of each fault diagnosis agent is determined based on the interaction success rate, protocol compliance rate, and communication success rate of each agent in completing each historical task. The experience value of each fault diagnosis agent is determined based on the total amount of historical tasks processed, the total number of target historical tasks that meet the preset complexity conditions, and the total running time. The decision weight for executing the current task is determined based on the performance factor, trust factor, experience value, and their respective weight coefficients of each fault diagnosis agent.

[0060] As an efficient method for determining decision weights, this invention can pre-store the collaborative decision weight relationship locally and calculate the decision weights of each fault diagnosis agent by calling the collaborative decision weight relationship. The collaborative decision weight relationship can be expressed as:

[0061] ;

[0062] in, represents the decision weight of the i-th fault diagnosis agent in collaborative decision-making, where α, β, and γ are weight coefficients assigned according to their respective importance. It is the task execution score of the j-th task. is the difficulty weight of the j-th task, and m is the number of historical tasks used in the performance factor calculation. It is the number of successful interactions on the k-th time. Let p be the total number of interactions, q be the total number of historical tasks, and q be the total number of interactions. This is the complexity score for the l-th task. It is the total running time of the i-th fault diagnosis agent.

[0063] In this embodiment, the decision weight of the i-th fault diagnosis agent in collaborative decision-making can be determined by... Sure.

[0064] Let be the performance factor of the i-th fault diagnosis agent. Experience is the trust factor for the i-th fault diagnosis agent. i Let be the experience value of the i-th fault diagnosis agent. The performance factor can be evaluated based on the fault diagnosis agent's performance in historical tasks. For example, it can be determined based on factors such as task success rate, task completion speed, and resource consumption efficiency. The performance factor can be determined using the following relationship:

[0065] .

[0066] The trust factor can be evaluated based on the historical behavior and interaction history of the fault diagnosis agent, including its cooperation records with other agents, the degree of adherence to protocols, and the accuracy of historical communications. The trust factor of the fault diagnosis agent can be calculated using the following formula:

[0067] .

[0068] The experience value can be determined based on the number of tasks the fault diagnosis agent participates in, the types of complex problems it handles, and the runtime in the data center. The experience value can be calculated using the following formula:

[0069] .

[0070] Since each agent is based on a large-scale language model and performs tasks simultaneously, running the language model in parallel at the same time may lead to insufficient computational and memory resources, causing tasks to fail. This may include the following:

[0071] For scenarios where the fault diagnosis agent runs multiple tasks simultaneously, when insufficient computing and / or memory resources are detected, the workload and task execution information of the fault diagnosis agent are obtained; the task type and impact of each task are read from the task execution information, and the urgency score of each task is determined based on the task type and impact; the workload during the execution of each task is determined based on the workload information, and the load score is determined based on the workload quantity and workload resource requirements; the priority of each task is determined based on the urgency score, load score, and their respective weight coefficients; and fault diagnosis is performed on the running status data based on the priority of the fault diagnosis subtasks.

[0072] In this embodiment, whenever computing and memory resources are insufficient, each task running by an agent needs to have its priority calculated according to this embodiment, and the tasks will be executed sequentially according to their respective priorities. As an efficient priority calculation method, priority calculation formulas can be pre-stored locally, and the priority of each task can be calculated by calling the priority formulas. The priority calculation formula can be expressed as:

[0073] ;

[0074] in, It is the priority of the i-th task. For the current workload, α represents the urgency of the i-th task, and β are weighting coefficients.

[0075] To further improve the accuracy of prediction results and address the issue that related artificial intelligence-based methods are black-box models, leading to a lack of interpretability in prediction results, the present invention also provides the following implementation methods:

[0076] The fault diagnosis intelligent agent consists of multiple agents, and the result generation intelligent agent includes an expert agent and an interactive agent. The result generation subtasks include a diagnosis result generation subtask and a report generation subtask. The fault diagnosis results and diagnosis result generation subtasks of each fault diagnosis intelligent agent are sent to the expert agent, so that the expert agent can generate a total fault diagnosis result based on the fault diagnosis results of each fault diagnosis intelligent agent and the expert knowledge base. When the diagnosis result generation subtask is completed, the report generation subtask and the total fault diagnosis result are sent to the interactive agent, so that the interactive agent can use a learning retrieval enhancement generation method to generate a fault prediction result based on the total fault diagnosis result and human interaction information.

[0077] In this embodiment, considering the self-consistency of each fault diagnosis agent, that is, each independently performs fault diagnosis and fault prediction, which can easily lead to conflicts, this embodiment adopts an expert agent to communicate and collaborate effectively with each fault diagnosis agent, integrates the analysis results of each fault diagnosis agent, and combines relevant knowledge to form an accurate and comprehensive overall fault diagnosis result.

[0078] The interactive intelligent agent generates fault prediction results based on the overall fault diagnosis results of the expert intelligent agent and interacts with users, such as operations and maintenance personnel, answering their various questions. Through continuous interaction with operations and maintenance personnel, it provides them with necessary information and operational knowledge. To effectively alleviate the illusion problem of language models and generate reference solutions and fault causes that include fault repair suggestions, this embodiment's interactive intelligent agent utilizes RAG (Relevance Analysis and Generation) technology to retrieve and analyze multi-source information such as expert experience, fault support records, product documentation, and web crawler results. Through this process, the interactive intelligent agent generates a detailed fault prediction result, which can also indicate the specific fault causes and scope of impact. Simultaneously, the interactive intelligent agent can provide reasonable repair suggestions based on the fault prediction results and the actual situation. For example, for identified hardware faults, the interactive intelligent agent will provide steps for replacing relevant components; for software problems, it will provide corresponding repair or configuration adjustment guidelines. The interactive intelligent agent uses natural language processing technology to present these suggestions to operations and maintenance personnel in an easily understandable form, ensuring they can efficiently perform repair tasks and thus helping them quickly and efficiently resolve problems.

[0079] As can be seen from the above, the intelligent interaction method in this embodiment not only simplifies the complex diagnosis and repair process, but also improves the efficiency and quality of operation and maintenance work.

[0080] Since each agent is based on a large-scale language model and performs tasks simultaneously, the parallel execution of the language model may lead to insufficient computational and memory resources. To avoid errors or other mishaps due to insufficient resources, based on the above embodiments, this invention also provides an implementation method for dynamic resource allocation, which may include the following:

[0081] Obtain workload and resource consumption information for the data collection agent, data analysis agent, fault diagnosis agent, and result generation agent; based on the workload and resource consumption information of the data collection agent, data analysis agent, fault diagnosis agent, and result generation agent, determine the resource allocation weights for the data collection agent, data analysis agent, fault diagnosis agent, and result generation agent at the current moment; adjust the resource allocation of the data collection agent, data analysis agent, fault diagnosis agent, and result generation agent according to the current resource allocation weights.

[0082] Because it involves enterprise-owned private data, an open-source model was chosen as the base model and deployed on a private machine. Compared to closed-source commercial models, the open-source model has fewer parameters, resulting in lower performance. Furthermore, both open-source and closed-source commercial models are trained on general domains, which are unfamiliar with the vertical data center operations domain. This can lead to "illusions," meaning that the language models used by each agent may generate seemingly reasonable but actually inaccurate or completely wrong information for unfamiliar content, resulting in reduced accuracy of the final fault prediction results. Building upon the above embodiment's use of RAG to acquire external knowledge and improve the accuracy of fault prediction results, this invention also provides a method for each agent to improve performance through continuous learning, which may include the following:

[0083] The original operating status data is sent to the first target end through the first learning interface so that the first target end can check the original operating status data; when feedback information that there is an error in the original operating status data is received, the correct original operating status data fed back by the first target end is received through the first learning interface, and the first learning task and the correct original operating status data are sent to the data collection agent, and the correct original operating status data is sent to the data collection agent as the original operating status data. The operational status data is sent to the second target end through the second learning interface so that the second target end can check the operational status data. When feedback information indicating that the operational status data is incorrect is received, the correct operational status data is received from the second target end through the second learning interface, and the second learning task and the correct operational status data are sent to the data collection agent. The correct operational status data is then sent as operational status data to the fault diagnosis agent. The fault prediction result is sent to the third target end through the third learning interface so that the third target end can review the fault prediction result. When feedback information indicating that the fault prediction result is incorrect is received, the correct fault prediction result is received from the third target end through the third learning interface, and the third learning task and the correct fault prediction result are sent to the result generation agent. The correct fault prediction result is then output as the fault prediction result. The fault diagnosis result is sent to the fourth target end through the fourth learning interface, so that the fourth target end can correct the fault diagnosis result based on the chaos engineering self-evolution mechanism; when feedback information that the fault diagnosis result is incorrect is received, the correct fault diagnosis result fed back by the fourth target end is received through the fourth learning interface, and the fourth learning task and the correct fault diagnosis result are sent to the fault diagnosis agent, and the correct fault diagnosis result is sent to the result generation agent.

[0084] In this embodiment, each intelligent agent has a third-party interface, such as a Teacher-Student interface. The third party, i.e., each target end, can perform manual review or inspection, or review or inspection via a cloud server. If the third party detects errors in the results of each intelligent agent, it can correct them and feed the correct data back to the corresponding intelligent agent, allowing it to learn from the teacher's corrections. This not only prevents errors from propagating backward but also enables the intelligent agents to continuously learn and improve performance. For example, when the data collection intelligent agent acquires performance data and operation log data, the first learning interface allows manual inspection of the data's completeness and accuracy, ensuring it is error-free before handing it over to the data analysis intelligent agent. Secondly, during the data completion and redundant data removal process, the second learning interface allows manual review of the analysis results, ensuring that the data cleaning and preprocessing work meets standards. Regarding the interactive intelligent agent that interacts with maintenance personnel, its corresponding learning interface ensures that the generated diagnostic reports and repair suggestions are manually reviewed, avoiding unnecessary maintenance operations due to erroneous information. Manual feedback not only provides an immediate error correction mechanism but also enhances the self-learning ability and accuracy of each intelligent agent through continuous correction and optimization. During task allocation, resource management, and coordination with other agents by the management agent, its reserved learning interface also provides an auditing interface. This ensures that while the entire data center operates efficiently, potential problems can be identified and resolved in a timely manner, preventing data center failures and resource waste. For the fault diagnosis agent, since human intervention cannot easily detect errors, this embodiment employs a chaotic engineering self-evolution mechanism to simulate data center failures. When an error is detected in the fault diagnosis result, it can be corrected automatically by the results of chaotic engineering, achieving the purpose of self-evolution.

[0085] As can be seen from the above, this embodiment, by introducing a learning interface at each agent, helps to improve the performance of each agent, effectively prevents error propagation, improves the accuracy of the final fault prediction result, and ensures the stability and reliability of the entire data center.

[0086] The present invention also provides a data center, which may include a fault prediction system. The fault prediction system may be deployed on the master node server of the data center or on the server with the best performance in the data center, which does not affect the implementation of the present invention.

[0087] Please see Figure 3The fault prediction system may include a managerial agent, a data collection agent, a data analysis agent, a fault diagnosis agent, an expert agent, and an interactive agent. The managerial agent communicates with each of the data collection, data analysis, fault diagnosis, expert, and interactive agents. The data collection agent communicates with the data analysis agent, the data analysis agent communicates with the fault diagnosis agent, the fault diagnosis agent communicates with the expert agent, and the expert agent communicates with the interactive agent. The number of data collection, data analysis, and fault diagnosis agents can be set according to actual needs. Figure 4 As shown, three data collection agents, three data analysis agents, and three fault diagnosis agents can be set up, with each agent corresponding to the other.

[0088] In this embodiment, the manager agent executes the steps of the fault prediction method described in any of the above embodiments. It coordinates the data collection agent, data analysis agent, fault diagnosis agent, expert agent, and interactive agent, enabling interaction between these agents and ensuring smooth execution of tasks such as data collection, analysis, diagnosis, and prediction. For example, after the data collection agent completes data collection, the manager agent notifies the data analysis agent to begin data processing. After the fault diagnosis agent outputs the fault diagnosis result, the manager agent transmits the result to the interactive agent interacting with maintenance personnel to generate detailed prediction results and repair suggestions. By coordinating the collaborative work of the data collection agent, data analysis agent, fault diagnosis agent, expert agent, and interactive agent, the manager agent can effectively improve the overall operating efficiency and response speed of the data center. The system comprises four intelligent agents: a data collection agent, which collects raw operational status data from each server in the data center upon receiving a data collection subtask; a data analysis agent, which processes the raw operational status data upon receiving a data analysis subtask to obtain operational status data that meets preset quality conditions; a fault diagnosis agent, which performs fault diagnosis on the operational status data upon receiving a fault diagnosis subtask; an expert agent, which generates a total fault diagnosis result by integrating the fault diagnosis results from each fault diagnosis agent upon receiving a diagnosis result generation subtask; and an interaction agent, which generates a fault prediction result based on the total fault diagnosis result, prior knowledge, and interaction information upon receiving a report generation subtask.

[0089] For example, the managerial agent is the core of the entire fault prediction system, responsible for at least task allocation, resource management, and coordination of interactions between agents. Regarding task allocation: the managerial agent can assign different tasks to appropriate agents based on the data center's needs and task priorities. For instance, when performance data needs to be collected, the managerial agent sends the data collection task to the data collection agent for execution; when data analysis is required, the task is delegated to the data analysis agent. The managerial agent manages and optimizes all computing and memory resources in the data center: monitoring the workload and resource consumption of each agent to ensure the reasonable allocation and efficient utilization of computing and memory resources. For example, when there are agents with excessive load, the managerial agent can adopt methods such as load balancing to achieve load balancing, or dynamically adjust resource allocation as necessary to maintain system stability and efficiency. By allocating tasks, managing resources, and coordinating interactions, the managerial agent ensures the efficient operation of the data center, guaranteeing its stability, reliability, and availability.

[0090] For example, the data collection agent can be configured with plans and tools to possess at least interface call functionality, monitoring tool call functionality, and data tracking functionality. The interface call functionality is used to obtain server operation log data by calling remote control interfaces, such as RedFish (out-of-band management protocol). The monitoring tool call functionality is used to obtain server performance indicator data by calling monitoring tool interfaces, such as having the ability to call tools like Rometheus (service monitoring system and time-series database). The data tracking functionality is used to obtain call chain data corresponding to target identifiers implanted in the tracking system, such as through programming. In this embodiment, the data collection agent can collect various performance indicator data such as CPU utilization, memory usage, and disk performance, as well as server operation log data and call chain data by calling various protocol frameworks and APIs (Application Programming Interfaces) of the server. For operation log data, standard RedFish interfaces or custom HTTP (Hypertext Transfer Protocol) interfaces can be used for programming to collect data. Tracking systems such as Jaeger and Zipkin embed specific identification information, such as IDs, for tracking at specific locations, and then the data collection agent collects data through programming. For performance metric data, it can be collected by calling the interfaces of monitoring tools such as Prometheus and Graphite, and through programming.

[0091] For example, the data analysis agent includes data processing functions, log parsing tool invocation functions, and data cleaning tool invocation functions. The data processing function can be programmed using programming capabilities, and the log parsing tool is named LogParser (tool name). Specifically, the data analysis agent filters runtime log data shorter than a preset length threshold by invoking the data processing function, thus preventing incomplete information in excessively short runtime log data from affecting the accuracy of subsequent data analysis. The log parsing tool invocation function parses the remaining runtime log data, removing invalid and redundant data. Since call chain information may be incomplete for various reasons, the data analysis agent can also delete incomplete call chain data by invoking the data processing function. Furthermore, the data analysis agent removes redundant data from server performance indicator data by invoking the data processing function and completes the server performance indicator data by invoking the data cleaning tool invocation function, such as using pandas (tool name) in Python (a compiled language) to complete missing data.

[0092] In this embodiment, the data analysis agent is used to process and optimize the raw operational status data collected by the data collection agent. Considering that the raw operational status data may have problems such as missing, redundant, incomplete, or inconsistent data, in order to avoid these low-quality data affecting the accuracy and effectiveness of subsequent fault diagnosis, this embodiment allows the data analysis agent to first call various data cleaning and preprocessing tools to fill in the missing data. For example, interpolation or prediction algorithms based on historical data can be used to fill in the missing data. Secondly, the data analysis agent can remove redundant data by finding duplicate records and merging similar data, thereby reducing the overhead of data storage and processing. For operational log data, in order to further analyze and mine the data, the data analysis agent also needs to parse the operational log data, extract key fields, format timestamps, identify event patterns, and convert unstructured log data into structured data.

[0093] Considering that each agent in the fault prediction system generates new knowledge during the learning process, including short-term and long-term knowledge, the fault prediction system automatically memorizes short-term knowledge, while it cannot memorize long-term knowledge. However, long-term knowledge can play a certain role in other dialogues and tasks, thereby supplementing the insufficiency of short-term knowledge memorization. This allows the fault prediction system to maintain the integrity of its memory over a considerable period. Based on the above embodiments, the fault prediction system of the present invention may further include a knowledge database; the knowledge database includes a structured knowledge sub-base and an unstructured knowledge sub-base; it acquires historical knowledge already learned by the manager agent and / or data collection agent and / or data analysis agent and / or fault diagnosis agent and / or expert agent and / or interactive agent, and selects target historical knowledge that is not accessed within a preset time period from the historical knowledge; for the first type of target historical knowledge belonging to the structured data type, the first type of target historical knowledge is converted into vector data and stored in the structured knowledge sub-base; for the second type of target historical knowledge belonging to the unstructured data type, the second type of target historical knowledge is converted into text data and stored in the unstructured knowledge sub-base.

[0094] In this embodiment, long-term knowledge refers to knowledge that has been learned but is no longer used in the near future. The preset time period can be flexibly adjusted according to the near-term duration of different application scenarios. Since fault prediction systems cannot be directly memorized, this embodiment uses a knowledge database for storage. For structured knowledge, to facilitate its conversion into vectors, it can be stored directly using a format such as Markdown (a lightweight markup language). For unstructured knowledge, it can be stored as text. To ensure data accuracy, long-term knowledge can be periodically manually verified. After confirmation, it is converted into vectors and stored in the knowledge database for future use or in other tasks.

[0095] As shown above, this embodiment achieves unified management of server faults in the data center based on multi-agent collaboration. It solves the problem of individual fault management modules operating independently and unable to coordinate, integrating fault management into a large system module that works intelligently and collaboratively. The fault prediction system uses an advanced large language model as its core engine, driving multiple agents to work collaboratively. It can flexibly call upon diverse tools and interfaces to accurately address complex fault scenarios that may occur during server operation. Through the powerful capabilities of deep learning and natural language processing, it can significantly improve the accuracy of fault identification and prediction, thereby effectively solving the difficulty of fault diagnosis for complex tasks in the data center, greatly reducing the workload of fault diagnosis and prediction, and simultaneously making all aspects of operation and maintenance an organic whole, effectively avoiding the fragmentation of the entire data center.

[0096] This invention also provides a corresponding apparatus for the fault prediction method, further enhancing the method's practicality. The apparatus can be described from both a functional module perspective and a hardware perspective. The fault prediction apparatus provided by this invention is described below. This apparatus is used to implement the fault prediction method provided by this invention. In this embodiment, the fault prediction apparatus may include or be divided into one or more program modules. These program modules are stored in a storage medium and executed by one or more processors to complete the fault prediction method disclosed in Embodiment 1. The program module referred to in this embodiment is a series of computer program instruction segments capable of performing a specific function, which is more suitable than the program itself for describing the execution process of the fault prediction apparatus in the storage medium. The following description will specifically introduce the functions of each program module in this embodiment. The fault prediction apparatus described below can be referred to in correspondence with the fault prediction method described above.

[0097] From the perspective of functional modules, see Figure 5 , Figure 5 This is a structural diagram of the fault prediction device provided in this embodiment under one specific implementation. The device may include:

[0098] The task decomposition module 501 is used to divide the fault prediction task into data collection subtasks, data analysis subtasks, fault diagnosis subtasks and result generation subtasks when a fault prediction task is detected.

[0099] The data collection module 502 is used to send data collection subtasks to the data collection agent so that the data collection agent can collect the raw operating status data of each server in the data center.

[0100] The data collection and analysis module 503 is used to send the raw operating status data and the data analysis sub-task to the data analysis agent when the data collection agent completes the data collection sub-task, so that the data analysis agent can process the raw operating status data to obtain operating status data that meets the preset quality conditions.

[0101] The fault diagnosis module 504 is used to send the running status data and fault diagnosis subtask to the fault diagnosis agent when the data analysis agent completes the data analysis subtask, so that the fault diagnosis agent can perform fault diagnosis on the running status data.

[0102] The fault prediction result output module 505 is used to send the fault diagnosis result and the result generation subtask to the result generation agent when the fault diagnosis agent completes the fault diagnosis subtask, so that the result generation agent can generate the fault prediction result based on the fault diagnosis result and prior knowledge. Among them, the data collection agent, data analysis agent, fault diagnosis agent and result generation agent are agents based on language models, which can autonomously complete tasks by configuring corresponding functions for each of them.

[0103] For example, in some embodiments of this example, the fault diagnosis module 504 is further configured to: when the fault diagnosis agent includes at least a first fault diagnosis agent, a second fault diagnosis agent, and a hybrid fault diagnosis agent, and the data collection agent includes at least a first data collection agent and a second data collection agent, send the first operating status data collected by the first data collection agent to the first fault diagnosis agent, send the second operating status data collected by the second data collection agent to the second fault diagnosis agent, and send the first operating status data and the second operating status data to the hybrid fault diagnosis agent, so that the first fault diagnosis agent, the second fault diagnosis agent, and the hybrid fault diagnosis agent simultaneously perform fault diagnosis on the operating status data.

[0104] For example, in some other embodiments of this embodiment, the fault prediction result output module 505 is further configured to: when there are multiple fault diagnosis agents, obtain the historical task execution data of each fault diagnosis agent; the historical task execution data includes at least the number of historical tasks, the execution parameters of each historical task, the behavior parameters of each historical task, and the interaction history parameters; determine the decision weight for the corresponding fault diagnosis agent to execute the current task based on the historical task execution data of each fault diagnosis agent; and determine the fault diagnosis result of the input result generating agent based on the fault diagnosis result of each fault diagnosis agent and its corresponding decision weight.

[0105] As an exemplary implementation of the above embodiments, the fault prediction result output module 505 can also be used to: determine the task execution score of the current historical task based on the current historical task success rate, the current historical task execution speed, and the current historical task resource consumption efficiency; determine the performance factor of each fault diagnosis agent based on the task execution score and task difficulty score corresponding to each historical task of each fault diagnosis agent; determine the trust factor of each fault diagnosis agent based on the interaction success rate, protocol compliance rate, and communication success rate of each fault diagnosis agent in completing each historical task; determine the experience value of each fault diagnosis agent based on the total amount of historical tasks processed by each fault diagnosis agent, the total number of target historical tasks that meet the preset complexity conditions, and the total running time; and determine the decision weight for executing the current task based on the performance factor, trust factor, and experience value of each fault diagnosis agent and their respective weight coefficients.

[0106] For example, in some other embodiments of this embodiment, the fault diagnosis module 504 is further configured to: when the fault diagnosis agent runs multiple tasks simultaneously, and when insufficient computing resources and / or memory resources are detected, obtain the workload information and task operation information of the fault diagnosis agent; read the task type and task impact of each task from the task operation information, and determine the urgency score of each task based on the task type and task impact of each task; determine the workload during the execution of each task based on the workload information, and determine the load score based on the workload quantity and workload resource requirements; determine the priority of each task based on the urgency score, load score, and their respective weight coefficients; and perform fault diagnosis on the running status data based on the priority of the fault diagnosis subtasks.

[0107] For example, in some other embodiments of this embodiment, the fault prediction result output module 505 is further configured to: when there are multiple fault diagnosis agents, the result generation agents include expert agents and interactive agents, and the result generation subtasks include diagnosis result generation subtasks and report generation subtasks, send the fault diagnosis results and diagnosis result generation subtasks of each fault diagnosis agent to the expert agent, so that the expert agent generates a total fault diagnosis result based on the fault diagnosis results of each fault diagnosis agent and the expert knowledge base; when the diagnosis result generation subtask is completed, send the report generation subtask and the total fault diagnosis result to the interactive agent, so that the interactive agent uses a learning retrieval enhancement generation method to generate a fault prediction result based on the total fault diagnosis result and human interaction information.

[0108] For example, in some other embodiments of this example, the above-described apparatus may further include a resource allocation module, which may be used to: acquire workload and resource consumption information of the data collection agent, data analysis agent, fault diagnosis agent, and result generation agent; determine the resource allocation weights of the data collection agent, data analysis agent, fault diagnosis agent, and result generation agent at the current time based on the workload and resource consumption information of the data collection agent, data analysis agent, fault diagnosis agent, and result generation agent; and adjust the resource allocation of the data collection agent, data analysis agent, fault diagnosis agent, and result generation agent according to the current resource allocation weights.

[0109] For example, in other embodiments of this example, the above-described apparatus may further include a learning module, which can be used to: send raw operating status data to a first target end through a first learning interface, so that the first target end can check the raw operating status data; when receiving feedback information that the raw operating status data is incorrect, receive the correct raw operating status data fed back by the first target end through the first learning interface, and send the first learning task and the correct raw operating status data to the data collection agent, and send the correct raw operating status data as raw operating status data to the data collection agent; send the operating status data to a second target end through a second learning interface, so that the second target end can check the operating status data. When feedback indicating an error in the running status data is received, the system receives the correct running status data from the second target end through the second learning interface, and sends the second learning task and the correct running status data to the data collection agent. The correct running status data is then sent as running status data to the fault diagnosis agent. The fault prediction result is sent to the third target end through the third learning interface for review. When feedback indicating an error in the fault prediction result is received, the system receives the correct fault prediction result from the third target end through the third learning interface, and sends the third learning task and the correct fault prediction result to the result generation agent. The correct fault prediction result is then output as the fault prediction result. The fault diagnosis result is sent to the fourth target end through the fourth learning interface, allowing the fourth target end to correct the fault diagnosis result based on a chaotic engineering self-evolution mechanism. When feedback indicating an error in the fault diagnosis result is received, the system receives the correct fault diagnosis result from the fourth target end through the fourth learning interface, and sends the fourth learning task and the correct fault diagnosis result to the fault diagnosis agent. The correct fault diagnosis result is then sent to the result generation agent.

[0110] The fault prediction device mentioned above is described from the perspective of functional modules. Furthermore, the present invention also provides an electronic device, which is described from the perspective of hardware. Figure 6 This is a schematic diagram of the structure of an electronic device provided in one embodiment of the present invention. For example... Figure 6 As shown, the electronic device includes a memory 60 for storing a computer program; and a processor 61 for executing the computer program to implement the steps of the fault prediction method as described in any of the above embodiments.

[0111] The processor 61 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 61 may also be a controller, microcontroller, microprocessor, or other data processing chip. The processor 61 may be implemented using at least one hardware form selected from DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), and PLA (Programmable Logic Array). The processor 61 may also include a main processor and a coprocessor. The main processor, also known as a CPU (Central Processing Unit), is used to process data in the wake-up state; the coprocessor is a low-power processor used to process data in the standby state. In some embodiments, the processor 61 may integrate a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content to be displayed on the screen. In some embodiments, the processor 61 may also include an AI (Artificial Intelligence) processor, which is used to handle computational operations related to machine learning.

[0112] The memory 60 may include one or more computer non-volatile storage media, which may be non-transitory. The memory 60 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices or flash memory devices. In some embodiments, the memory 60 may be an internal storage unit of an electronic device, such as a server hard drive. In other embodiments, the memory 60 may be an external storage device of an electronic device, such as a plug-in hard drive on a server, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Furthermore, the memory 60 may include both internal and external storage units of the electronic device. The memory 60 can be used not only to store application software and various types of data installed on the electronic device, such as code in the process of executing the fault prediction method, but also to temporarily store data that has been output or will be output. In this embodiment, the memory 60 is used to store at least the following computer program 601, which, after being loaded and executed by the processor 61, can implement the relevant steps of the fault prediction method disclosed in any of the foregoing embodiments. In addition, the resources stored in the memory 60 may also include an operating system 602 and data 603, and the storage method may be temporary storage or permanent storage. The operating system 602 may include Windows, Unix, Linux, etc. The data 603 may include, but is not limited to, data corresponding to fault prediction results.

[0113] In some embodiments, the above-mentioned electronic device may further include a display screen 62, an input / output interface 63, a communication interface 64 (or network interface), a power supply 65, and a communication bus 66. The display screen 62 and the input / output interface 63, such as a keyboard, are user interfaces. Exemplary user interfaces may also include standard wired interfaces, wireless interfaces, etc. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a display screen or display unit, used to display information processed in the electronic device and to display a visual user interface. The communication interface 64 may exemplary include wired and / or wireless interfaces, such as a Wi-Fi interface, a Bluetooth interface, etc., typically used to establish communication connections between the electronic device and other electronic devices. The communication bus 66 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. This bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 6 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0114] Those skilled in the art will understand that Figure 6 The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, such as sensors 67 that perform various functions.

[0115] It is understood that if the fault prediction method in the above embodiments is implemented as a software functional unit and sold or used as an independent product, it can be stored in a non-volatile storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the related technology, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and executes all or part of the steps of the methods in the various embodiments of the present invention. The aforementioned storage medium includes, but is not limited to, various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), electrically erasable programmable ROM, registers, hard disks, multimedia cards, card-type memories (e.g., SD or DX memories), magnetic memories, removable disks, CD-ROMs, magnetic disks, or optical disks. Based on this, the present invention also provides a non-volatile storage medium storing a computer program, which, when executed by a processor, performs the steps of the fault prediction method as described in any of the above embodiments.

[0116] It is understood that if the fault prediction method in the above embodiments is implemented as a software functional unit and sold or used as an independent product, the computer software product may not need to be stored in a physical storage medium. For example, it can be directly transmitted to a computer or other device with information processing capabilities via a wired or wireless network to execute all or part of the steps of the methods in the various embodiments of the present invention. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the related technology, or all or part of the technical solution, can be embodied in the form of a software product. Based on this, the present invention also provides a computer program product storing a computer program, which, when executed by a processor, performs the steps of the fault prediction method as described in any of the above embodiments.

[0117] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the hardware disclosed in the embodiments, including electronic devices, non-volatile storage media, and computer program products, the descriptions are relatively simple since they correspond to the methods disclosed in the embodiments; relevant details can be found in the method section.

[0118] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0119] The foregoing has provided a detailed description of the data center and fault prediction method, electronic device, non-volatile storage medium, and computer program product provided by this invention. Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are merely for the purpose of helping to understand the method and core ideas of this invention. It should be noted that, based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention. Several improvements and modifications can be made to this invention without departing from the principles of this invention, and these improvements and modifications also fall within the scope of protection of this invention.

Claims

1. A fault prediction method, characterized in that, include: The manager's intelligent agent divides the fault prediction task into data collection subtasks, data analysis subtasks, fault diagnosis subtasks, and result generation subtasks. Based on the data center's needs and task priorities, different tasks are assigned to the corresponding intelligent agents, and the workload and resource consumption of each intelligent agent are monitored. When there are intelligent agents with excessive load, a load balancing method is adopted to achieve load balancing. The manager agent sends the data collection subtask to the data collection agent. The data collection agent calls the remote control interface through the interface call function to obtain the server's operation log data, calls the monitoring tool interface through the monitoring tool call function to obtain the server's performance index data, and obtains the call chain data corresponding to the target identifier implanted in the tracking system through the tracking data function, so that the data collection agent can collect the original operating status data of each server in the data center. When the data collection agent completes the data collection subtask, the manager agent sends the raw operational status data and the data analysis subtask to the data analysis agent. The data analysis agent filters operational log data whose length is less than a preset length threshold by calling the data processing function, and performs log parsing on the remaining operational log data by calling the log parsing tool to remove invalid and redundant data, delete incomplete call chain data, delete redundant data in server performance index data, and complete the server performance index data by calling the data cleaning tool, so that the data analysis agent can process the raw operational status data to obtain operational status data that meets the preset quality conditions. When the data analysis agent completes the data analysis sub-task, the manager agent sends the operation status data and fault diagnosis sub-task to the fault diagnosis agent. The fault diagnosis agent includes at least a first fault diagnosis agent, a second fault diagnosis agent, and a hybrid fault diagnosis agent that simultaneously diagnose faults in the operation status data. The first fault diagnosis agent and the second fault diagnosis agent analyze, diagnose, and predict single operation status data, respectively. The hybrid fault diagnosis agent merges all operation status data and analyzes, diagnoses, and predicts the merged operation status data. The first fault diagnosis intelligent agent, the second fault diagnosis intelligent agent, and the hybrid fault diagnosis intelligent agent have programming and communication capabilities through configuration planning. They can call the relevant machine learning models required through programming and explain the reasons for the diagnosis or prediction conclusions through communication. When the fault diagnosis agent completes the fault diagnosis sub-task, the manager agent sends the fault diagnosis results and diagnosis result generation sub-tasks of each fault diagnosis agent to the expert agent, so that the expert agent can generate the overall fault diagnosis result based on the fault diagnosis results of each fault diagnosis agent and the expert knowledge base. Once the diagnosis result generation subtask is completed, the manager agent will send the report generation subtask and the total fault diagnosis result to the interactive agent, so that the interactive agent can use the learning retrieval enhancement generation method to generate fault prediction results based on the total fault diagnosis result and human interaction information. The fault diagnosis result is sent to the fourth target end through the fourth learning interface, so that the fourth target end can correct the fault diagnosis result based on the chaos engineering self-evolution mechanism; when feedback information that the fault diagnosis result is incorrect is received, the correct fault diagnosis result fed back by the fourth target end is received through the fourth learning interface, and the fourth learning task and the correct fault diagnosis result are sent to the fault diagnosis agent, and the correct fault diagnosis result is sent to the result generation agent. Among them, the data collection agent, data analysis agent, fault diagnosis agent, and result generation agent are language-based agents that autonomously complete tasks by configuring corresponding functions for each of them. Through task planning, memory, and external tool capabilities, they identify the tasks to be executed and the execution methods, and autonomously make decisions to complete the corresponding tasks.

2. The fault prediction method according to claim 1, characterized in that, The data collection agent includes at least a first data collection agent and a second data collection agent. It sends the operational status data and the fault diagnosis subtask to the fault diagnosis agent, enabling the fault diagnosis agent to perform fault diagnosis on the operational status data, including: The first operational status data collected by the first data collection agent is sent to the first fault diagnosis agent, the second operational status data collected by the second data collection agent is sent to the second fault diagnosis agent, and the first operational status data and the second operational status data are sent to the hybrid fault diagnosis agent.

3. The fault prediction method according to claim 1, characterized in that, The step of sending the fault diagnosis results and the result generation subtask to the result generation agent includes: Obtain historical task execution data for each fault diagnosis agent; the historical task execution data includes at least the number of historical tasks, execution parameters of each historical task, behavioral parameters of each historical task, and interaction history parameters. Based on the historical task execution data of each fault diagnosis agent, determine the decision weight for the corresponding fault diagnosis agent to execute the current task; Based on the fault diagnosis results of each fault diagnosis agent and their corresponding decision weights, the fault diagnosis results of the agent generated by inputting the results are determined.

4. The fault prediction method according to claim 3, characterized in that, The step of determining the decision weight for each fault diagnosis agent to execute the current task based on the historical task execution data of each fault diagnosis agent includes: The task execution score of the current historical task is determined based on the success rate, execution speed, and resource consumption efficiency of the current historical task. The performance factor of each fault diagnosis agent is determined based on the task execution score and task difficulty score of each historical task. The trust factor of each fault diagnosis agent is determined based on the interaction success rate, protocol compliance rate, and communication success rate of each agent in completing each historical task. The experience value of each fault diagnosis agent is determined based on the total amount of historical tasks processed by each agent, the total number of target historical tasks that meet the preset complexity conditions, and the total running time. The decision weights for executing the current task are determined based on the performance factors, trust factors, and experience values ​​of each fault diagnosis agent and their respective weight coefficients.

5. The fault prediction method according to claim 1, characterized in that, The fault diagnosis agent runs multiple tasks simultaneously, and performs fault diagnosis on the operational status data, including: When insufficient computing and / or memory resources are detected, the workload information and task execution information of the fault diagnosis agent are obtained. The task type and impact of each task are read from the task operation information, and the urgency score of each task is determined based on the task type and impact. The workload during the execution of each task is determined based on the workload information, and the workload score is determined based on the number of workloads and the workload resource requirements. The priority of each task is determined based on its urgency score, workload score, and corresponding weighting coefficient. Based on the priority of the fault diagnosis subtask, fault diagnosis is performed on the running status data.

6. The fault prediction method according to claim 1, characterized in that, Also includes: Obtain the workload and resource consumption information of the data collection agent, the data analysis agent, the fault diagnosis agent, and the result generation agent; Based on the workload and resource consumption information of the data collection agent, the data analysis agent, the fault diagnosis agent, and the result generation agent, the resource allocation weights of the data collection agent, the data analysis agent, the fault diagnosis agent, and the result generation agent at the current moment are determined; Based on the current resource allocation weights, adjust the resource allocation of the data collection agent, the data analysis agent, the fault diagnosis agent, and the result generation agent.

7. The fault prediction method according to any one of claims 1 to 6, characterized in that, Also includes: The original operating status data is sent to the first target terminal through the first learning interface so that the first target terminal can check the original operating status data; When a feedback message indicating that the original running status data is incorrect is received, the correct original running status data fed back by the first target end is received through the first learning interface, and the first learning task and the correct original running status data are sent to the data collection agent, and the correct original running status data is sent to the data collection agent as the original running status data. The operation status data is sent to the second target end through the second learning interface so that the second target end can check the operation status data; when feedback information that the operation status data is incorrect is received, the correct operation status data fed back by the second target end is received through the second learning interface, and the second learning task and the correct operation status data are sent to the data collection agent, and the correct operation status data is sent as operation status data to the fault diagnosis agent; The fault prediction result is sent to the third target end through the third learning interface so that the third target end can review the fault prediction result; when feedback information that the fault prediction result is incorrect is received, the correct fault prediction result fed back by the third target end is received through the third learning interface, and the third learning task and the correct fault prediction result are sent to the result generating agent, and the correct fault prediction result is output as the fault prediction result.

8. An electronic device, characterized in that, It includes a processor and a memory, wherein the processor is configured to implement the steps of the fault prediction method as described in any one of claims 1 to 7 when executing a computer program stored in the memory.

9. A non-volatile storage medium, characterized in that, The non-volatile storage medium stores a computer program that, when executed by a processor, implements the steps of the fault prediction method as described in any one of claims 1 to 7.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the fault prediction method according to any one of claims 1 to 7.

11. A data center, characterized in that, Including fault prediction systems; The fault prediction system includes a manager intelligent agent, a data collection intelligent agent, a data analysis intelligent agent, a fault diagnosis intelligent agent, an expert intelligent agent, and an interactive intelligent agent. The manager agent communicates with the data collection agent, the data analysis agent, the fault diagnosis agent, the expert agent, and the interaction agent, respectively, and executes the steps of the fault prediction method as described in any one of claims 1 to 7. The data collection agent is used to collect the raw operating status data of each server in the data center after receiving the data collection subtask. The data analysis intelligent agent is used to process the original running status data after receiving the data analysis subtask to obtain running status data that meets the preset quality conditions. The fault diagnosis intelligent agent is used to perform fault diagnosis on the running status data when it receives a fault diagnosis subtask. The expert intelligent agent is used to generate a subtask when it receives a diagnosis result, and to generate a total fault diagnosis result by integrating the fault diagnosis results of each fault diagnosis intelligent agent. The interactive intelligent agent is used to generate a fault prediction result based on the overall fault diagnosis result, prior knowledge, and interactive information when it receives a report generation subtask.

12. The data center according to claim 11, characterized in that, The data collection agent includes interface call function, monitoring tool call function and data tracking function; Specifically, the interface call function is used to obtain server operation log data by calling the remote control interface; the monitoring tool call function is used to call the monitoring tool interface to obtain server performance indicator data; and the tracing data function is used to obtain call chain data corresponding to the target identifier implanted in the tracing system.

13. The data center according to claim 11, characterized in that, The data analysis intelligent agent includes data processing functions, log parsing tool invocation functions, and data cleaning tool invocation functions; The data analysis agent filters out running log data whose length is less than a preset length threshold by calling the data processing function; it then uses the log parsing tool to parse the remaining running log data and remove invalid and redundant data. The data analysis agent also deletes incomplete call chain data by invoking data processing functions; The data analysis agent also removes redundant data from the server performance index data by calling the data processing function, and completes the server performance index data by calling the data cleaning tool.

14. The data center according to claim 11, characterized in that, The fault prediction system also includes a knowledge database; the knowledge database includes a structured knowledge sub-base and an unstructured knowledge sub-base. Acquire the historical knowledge that the manager agent and / or the data collection agent and / or the data analysis agent and / or the fault diagnosis agent and / or the expert agent and / or the interaction agent has learned, and select target historical knowledge that is not accessed within a preset time period from the historical knowledge; For the first type of target historical knowledge that belongs to the structured data type, the first type of target historical knowledge is converted into vector data and stored in the structured knowledge sub-base; For the second type of target historical knowledge that belongs to the unstructured data type, the second type of target historical knowledge is converted into text data and stored in the unstructured knowledge sub-base.

Images