Cloud server health management method and system based on large model

By building a GRU and Transformer model based on large models, combining the performance status data and maintenance records of cloud servers, real-time health management of cloud servers is realized, solving the problems of cloud server failure prediction and maintenance guarantee, and improving operation and maintenance efficiency and service stability sex.

CN119065920APending Publication Date: 2024-12-03SHANDONG LANGCHAO YUNTOU INFORMATION TECH CO LTD
View PDF 0 Cites 1 Cited by

Patent Information

Application Number
CN202410986902.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-23
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The existing technology is difficult to effectively manage the health status of cloud servers, which makes it difficult to predict faults and repair guarantees, affecting the quality of cloud computing service and operation and maintenance efficiency.

Method used

Using a large-scale model-based method, by building a GRU model and a Transformer model, combining the performance status data and maintenance records of cloud servers, real-time performance warning and fault prediction are achieved, and abnormal state analysis and maintenance suggestions are provided.

Benefits of technology

It realizes accurate warning and fault prediction of cloud server status, improves the processing efficiency and diagnostic effects of operation and maintenance personnel, reduces production costs, and ensures the stable operation of cloud servers.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119065920A_ABST
    Figure CN119065920A_ABST
Patent Text Reader

Abstract

The invention discloses a cloud server health management method and system based on a large model, and relates to the technical field of server health management, and the method comprises the steps: collecting the performance state data of a cloud server, and the maintenance record and historical report information of the cloud server; preprocessing the performance state data, and randomly dividing the performance state data into a training set and a test set; constructing a GRU model based on error compensation, training the GRU model by using the training set, and evaluating the GRU model by using the test set; establishing a performance early warning model by using the output of the GRU model to judge the running state of the cloud server, and giving an alarm when detecting that the cloud server is abnormal; carrying out format conversion on the maintenance record and the historical report information for finely adjusting the large model; and inputting the performance state data during alarming into the fine-tuned large model so as to output an abnormal state prediction result, an abnormal generation reason and related maintenance suggestions. According to the invention, accurate early warning of the cloud server state can be realized, and the processing efficiency of operation and maintenance personnel is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of server health management, and in particular to a cloud server health management method and system based on a large model. Background Art

[0002] The continuous advancement of computer technology has accelerated the rapid development of all walks of life and promoted the arrival of the intelligent era. Data elements are the key to the intelligent era. How to extract useful information from massive data has become an important step in the intelligent transformation. How to support the processing and calculation of massive data requires powerful computer processing capabilities. Cloud servers can fully integrate computing resources and have the characteristics of strong portability, low resource prices, and high degree of personalization. Users can choose to rent corresponding computing resources according to their own needs. The performance of cloud servers directly affects the service quality and customer experience of cloud computing. Therefore, healthy management of cloud servers can avoid significant losses caused by server failures.

[0003] As an important technological breakthrough in the field of artificial intelligence, big models have gradually become the focus of attention in the scientific and academic circles in recent years. With the advent of the era of intelligence and big data, people have gradually realized the limitations of traditional models in processing massive data. Therefore, building larger and more complex models has become the key to solving this problem. Big models are usually based on deep learning algorithms and simulate human cognitive processes by building huge neural networks. This model requires a huge data set for training in order to learn and understand various complex patterns and associations. With the continuous increase of training data, the performance of big models has gradually improved, and they can more accurately identify, analyze and predict various phenomena. In practical applications, big models have shown great potential. Whether in natural language processing, image recognition or speech recognition, big models have achieved remarkable results. It can understand human language, identify objects in images, and recognize commands in speech, providing humans with more intelligent and convenient services.

[0004] It is urgent to use big model technology to achieve health management of cloud servers. How to deal with the increasingly cumbersome health management of cloud server equipment, solve the maintenance and guarantee work of the cloud server throughout its life cycle, provide operation and maintenance personnel with rich and extensive problem-solving methods, improve the stability and security of cloud server operation, and reduce production costs are the top priorities. Summary of the invention

[0005] In view of the needs and deficiencies of current technological development, the present invention provides a cloud server health management method and system based on a large model, which is used to achieve real-time and accurate prediction and health management of cloud server performance and ensure the stable operation of cloud servers.

[0006] In the first aspect, the present invention provides a cloud server health management method based on a large model. The technical solution adopted to solve the above technical problems is as follows:

[0007] A cloud server health management method based on a large model comprises the following steps:

[0008] S1. Collecting performance status data of the cloud server, as well as maintenance records and historical report information of the cloud server, wherein the historical report information includes historical abnormal operation status, causes of abnormality, and maintenance measures taken;

[0009] S2, preprocessing the collected cloud server performance status data, and randomly dividing the preprocessed cloud server performance status data into a training set and a test set;

[0010] S3. Build a GRU model based on error compensation, use the training set to train the GRU model, and use the test set to evaluate the generalization ability and prediction accuracy of the GRU model;

[0011] S4. Use the output of the GRU model to establish a performance warning model to determine the operating status of the cloud server and issue an alarm when an abnormality is detected in the cloud server;

[0012] S5. Use the Transformer-based pre-trained big model as the base model, convert the format of the collected cloud server maintenance records and historical report information data, and use the format-converted data to fine-tune the big model; input the performance status data of the cloud server when the performance warning model alarms into the fine-tuned big model, and the fine-tuned big model outputs the abnormal status prediction results, the causes of the abnormality, and related maintenance suggestions.

[0013] Optionally, executing step S3, the constructed GRU model includes N GRU submodules, each of which is a standard GRU network capable of processing sequence data and capturing long-term dependencies in time series;

[0014] The input of each GRU submodule is the original input data X of the training set;

[0015] The prediction output of the first GRU submodule is the cloud server performance indicator data The target output, i.e. the actual output, is the cloud server performance indicator data Y;

[0016] Output the prediction of the first GRU submodule The residual of the target output Y As the target output of the second GRU submodule, the predicted output of the second GRU submodule is

[0017] Output the prediction of the second GRU submodule and target output ΔY 1 The residual As the target output of the third GRU submodule, the predicted output of the third GRU submodule is

[0018] Continue to perform the above operations until the Nth GRU submodule, and output the prediction of the N-1th GRU submodule and target output ΔY N-2 The residual As the target output of the Nth GRU submodule, the predicted output of the Nth GRU submodule is

[0019] Finally, we get N prediction outputs The N prediction outputs are accumulated to form the final prediction result of the GRU model.

[0020] Optionally, the performance of the GRU model is tested using a test set, and the mean square error function is used as the evaluation indicator during the test:

[0021]

[0022] Among them, RMSE represents the root mean square error, M represents the number of samples contained in the test set, and y i represents the target output value, Represents the predicted output value;

[0023] During the entire test process, the RMSE value is within the set threshold range, indicating that the GRU model test is qualified.

[0024] Optionally, executing step S4, setting a threshold for distinguishing between a normal state and an abnormal state, and a time threshold for continuous operation of an abnormal state in the performance warning model;

[0025] If the performance warning model determines that the cloud server is in an abnormal state based on the output of the GRU model, and the duration of the cloud server running in this state does not exceed the set time threshold, then the cloud server is determined to be in a normal state;

[0026] If the performance warning model determines that the cloud server is in an abnormal state based on the output of the GRU model, and the duration of the cloud server running in this state exceeds the set time threshold, it is determined that the cloud server is about to become abnormal. At this time, the performance warning model finds out the potential factors affecting the performance of the cloud server based on the changes in the GRU model input data, and inputs them into the fine-tuned large model.

[0027] Optionally, step S2 is performed to pre-process the collected cloud server performance status data, specifically including:

[0028] Data cleaning: Check the integrity of the data, remove or fill missing values, remove irrelevant features, and handle outliers;

[0029] Data standardization: standardize the data to make it have a standard normal distribution;

[0030] Data coding: Encode categorical data;

[0031] Sequence data reconstruction: For time series data, according to the requirements of the GRU model, organize the data into a sequence form to ensure that the shape of the input data meets the expectations of the GRU model;

[0032] Subsequently, the preprocessed cloud server performance status data are randomly divided into training sets and test sets in a ratio of 8:2 or 7:3.

[0033] In the second aspect, the present invention provides a cloud server health management system based on a large model, and the technical solution adopted to solve the above technical problems is as follows:

[0034] A cloud server health management system based on a large model, comprising:

[0035] A data collection module is used to collect the performance status data of the cloud server, as well as the maintenance records and historical report information of the cloud server, wherein the historical report information includes the historical abnormal operation status, the cause of the abnormality and the maintenance measures taken;

[0036] A preprocessing module, used to preprocess the collected cloud server performance status data;

[0037] A data partitioning module is used to randomly partition the pre-processed cloud server performance status data into a training set and a test set;

[0038] Model building module, used to build a GRU model based on error compensation;

[0039] The training and testing module is used to train the GRU model using the training set and to evaluate the generalization ability and prediction accuracy of the GRU model using the test set;

[0040] A model building module is used to build a performance warning model using the output of the GRU model;

[0041] Performance warning model, used to determine the operating status of the cloud server and issue an alarm when an abnormality is detected in the cloud server;

[0042] A format conversion module is used to convert the collected cloud server maintenance records and historical report information data into different formats;

[0043] The model fine-tuning module is used to use the Transformer-based pre-trained large model as the base model and use the format-converted data to fine-tune the large model;

[0044] The fine-tuned large model is used to take the performance status data of the cloud server when the performance warning model issues an alarm as input, and output the abnormal status prediction results, the causes of the abnormality, and related maintenance suggestions.

[0045] Optionally, the GRU model constructed by the model building module includes N GRU sub-modules, each of which is a standard GRU network capable of processing sequence data and capturing long-term dependencies in time series;

[0046] The input of each GRU submodule is the original input data X of the training set;

[0047] The prediction output of the first GRU submodule is the cloud server performance indicator data The target output, i.e. the actual output, is the cloud server performance indicator data Y;

[0048] Output the prediction of the first GRU submodule The residual of the target output Y As the target output of the second GRU submodule, the predicted output of the second GRU submodule is

[0049] Output the prediction of the second GRU submodule and target output ΔY 1 The residual As the target output of the third GRU submodule, the predicted output of the third GRU submodule is

[0050] Continue to perform the above operations until the Nth GRU submodule, and output the prediction of the N-1th GRU submodule and target output ΔY N-2 The residual As the target output of the Nth GRU submodule, the predicted output of the Nth GRU submodule is

[0051] Finally, we get N prediction outputs The N prediction outputs are accumulated to form the final prediction result of the GRU model.

[0052] Optionally, the training test module uses a test set to test the performance of the GRU model, and the mean square error function is used as an evaluation indicator during the test:

[0053]

[0054] Among them, RMSE represents the root mean square error, M represents the number of samples contained in the test set, and y i represents the target output value, Represents the predicted output value;

[0055] During the entire test process, the RMSE value is within the set threshold range, indicating that the GRU model test is qualified.

[0056] Optionally, when using the model building module to build the performance warning model, a threshold for distinguishing between a normal state and an abnormal state, and a time threshold for continuous operation of an abnormal state are set;

[0057] If the performance warning model determines that the cloud server is in an abnormal state based on the output of the GRU model, and the duration of the cloud server running in this state does not exceed the set time threshold, then the cloud server is determined to be in a normal state;

[0058] If the performance warning model determines that the cloud server is in an abnormal state based on the output of the GRU model, and the duration of the cloud server running in this state exceeds the set time threshold, it is determined that the cloud server is about to become abnormal. At this time, the performance warning model finds out the potential factors affecting the performance of the cloud server based on the changes in the GRU model input data, and inputs them into the fine-tuned large model.

[0059] Optionally, the preprocessing module involved preprocesses the collected cloud server performance status data, and the specific process includes:

[0060] Data cleaning: Check the integrity of the data, remove or fill missing values, remove irrelevant features, and handle outliers;

[0061] Data standardization: standardize the data to make it have a standard normal distribution;

[0062] Data coding: Encode categorical data;

[0063] Sequence data reconstruction: For time series data, according to the requirements of the GRU model, organize the data into a sequence form to ensure that the shape of the input data meets the expectations of the GRU model;

[0064] Subsequently, the data partitioning module randomly divides the preprocessed cloud server performance status data into a training set and a test set in a ratio of 8:2 or 7:3.

[0065] The cloud server health management method and system based on a large model of the present invention has the following beneficial effects compared with the prior art:

[0066] 1. The present invention fully combines the advantages of traditional machine learning and large models to manage the health status of cloud servers in real time. It can realize the integration of different status information of cloud servers and output effective cloud server health management strategies, including server fault prediction and early warning and health status protection, to improve diagnostic effects and maintenance efficiency.

[0067] 2. The present invention can realize accurate early warning of the status of cloud servers, tell operation and maintenance personnel the possible factors causing fault problems and give professional and feasible guiding suggestions and implementation measures, thereby improving the processing efficiency of operation and maintenance personnel. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Attached Figure 1 is a flow chart of a method according to Embodiment 1 of the present invention;

[0069] Attached Figure 2 is a module connection block diagram of the second embodiment of the present invention;

[0070] Attached Figure 3 It is a framework diagram of the GRU model in an embodiment of the present invention;

[0071] Attached Figure 4 It is a framework diagram of a GRU submodule in the GRU model in an embodiment of the present invention. DETAILED DESCRIPTION

[0072] In order to make the technical solution, the technical problem solved and the technical effect of the present invention more clearly understood, the technical solution of the present invention is clearly and completely described below in conjunction with specific embodiments.

[0073] Embodiment 1:

[0074] Combined with Figure 1 This embodiment proposes a cloud server health management method based on a large model, which includes the following steps:

[0075] S1. Collecting performance status data of the cloud server, as well as maintenance records and historical report information of the cloud server, wherein the historical report information includes historical abnormal operation status, causes of abnormality, and maintenance measures taken.

[0076] S2. Preprocess the collected cloud server performance status data, and randomly divide the preprocessed cloud server performance status data into a training set and a test set.

[0077] The preprocessing process specifically includes:

[0078] (1) Data cleaning: Check the integrity of the data, remove or fill missing values, remove irrelevant features, and handle outliers;

[0079] (2) Data standardization: standardize the data to make it have a standard normal distribution;

[0080] (3) Data coding: Encoding categorical data;

[0081] (4) Sequence data reconstruction: For time series data, according to the requirements of the GRU model, the data is organized into a sequence form to ensure that the shape of the input data meets the expectations of the GRU model.

[0082] The preprocessed cloud server performance status data is randomly divided into training set and test set in the ratio of 8:2 or 7:3.

[0083] S3. Build a GRU model based on error compensation, use the training set to train the GRU model, and use the test set to evaluate the generalization ability and prediction accuracy of the GRU model.

[0084] Reference Figure 4 The GRU (Gated Recurrent Unit) model is a recurrent neural network (RNN) architecture designed to solve the long-term dependency problem in standard RNNs. GRU introduces a gating mechanism to control the flow of information, allowing the model to better capture long-term dependencies in sequence data. The core idea of ​​GRU is to use update gates (UpdateGate) and reset gates (Reset Gate) to control the transmission of information. The update gate determines how much previous information needs to be retained in the current state, while the reset gate determines how much previous information to ignore. Both gates are calculated using a sigmoid function, and their output values ​​are between 0 and 1, indicating the degree of information retention or ignoring.

[0085] The workflow of GRU is roughly as follows:

[0086] Reset gate: First, the reset gate calculates a reset gate value based on the current input and the hidden state at the previous moment. This value determines how much information in the hidden state at the previous moment needs to be ignored.

[0087] Candidate hidden state: Next, the model uses the previous hidden state after the reset gate and the current input to calculate the candidate hidden state. This candidate hidden state contains the information of the current input and the information of the previous hidden state after the reset.

[0088] Update gate: The update gate then calculates an update gate value based on the current input and the hidden state at the previous moment. This value determines how much information in the current hidden state comes from the hidden state at the previous moment, and how much information comes from the newly calculated candidate hidden state.

[0089] Final hidden state: Finally, the model uses the update gate to combine the hidden state of the previous moment and the candidate hidden state to calculate the final hidden state of the current moment. This hidden state contains both the information of the current input and the information of the previous sequence, so that it can better handle long-term dependency problems. The output expression of each unit gate is as follows:

[0090] zt=σ(W z ·[h t-1 , x t ]) Formula (1),

[0091] r t =σ(W r ·[h t-1 , x t ]) Formula (2),

[0092]

[0093] Combined with Figure 3 ,The GRU model constructed in this embodiment includes N GRU submodules, each GRU submodule is a standard GRU network that can process sequence data and capture long-term dependencies in time series;

[0094] The input of each GRU submodule is the original input data X of the training set;

[0095] The prediction output of the first GRU submodule (GRU Module 1) is the cloud server performance indicator data The target output, i.e. the actual output, is the cloud server performance indicator data Y;

[0096] Output the prediction of the first GRU submodule (GRU Module 1) The residual of the target output Y As the target output of the second GRU submodule (GRU Module 2), the predicted output of the second GRU submodule (GRUModule 2) is

[0097] Output the prediction of the second GRU submodule (GRU Module 2) and target output ΔY 1 The residual As the target output of the third GRU submodule (GRU Module 3), the predicted output of the third GRU submodule (GRU Module 3) is

[0098] Continue to perform the above operations until the Nth GRU submodule (GRU Module N), and output the prediction of the N-1th GRU submodule (GRU Module N-1) and target output ΔY N-2 The residual As the target output of the Nth GRU submodule (GRU Module N), the predicted output of the Nth GRU submodule (GRU Module N) is

[0099] Finally, we get N prediction outputs The N prediction outputs are accumulated to form the final prediction result of the GRU model.

[0100] This example uses a test set to test the performance of the GRU model, and uses the mean square error function as an evaluation indicator during the test:

[0101]

[0102] Among them, RMSE represents the root mean square error, M represents the number of samples contained in the test set, and y i represents the target output value, Represents the predicted output value;

[0103] During the entire test process, the RMSE value is within the set threshold range, indicating that the GRU model test is qualified.

[0104] S4. Use the output of the GRU model to establish a performance warning model to determine the operating status of the cloud server and issue an alarm when an abnormality is detected in the cloud server.

[0105] In the performance warning model, the threshold for distinguishing between normal state and abnormal state, as well as the time threshold for the continuous operation of abnormal state are set;

[0106] If the performance warning model determines that the cloud server is in an abnormal state based on the output of the GRU model, and the duration of the cloud server running in this state does not exceed the set time threshold, then the cloud server is determined to be in a normal state;

[0107] If the performance warning model determines that the cloud server is in an abnormal state based on the output of the GRU model, and the duration of the cloud server running in this state exceeds the set time threshold, it is determined that the cloud server is about to become abnormal. At this time, the performance warning model finds out the potential factors affecting the performance of the cloud server based on the changes in the GRU model input data, and inputs them into the fine-tuned large model.

[0108] S5. Use the Transformer-based pre-trained large model as the base model, such as ChatGLM and BaiChuan, convert the collected cloud server maintenance records and historical report information data into different formats, and use the converted data to fine-tune the large model. Input the performance status data of the cloud server when the performance warning model issues an alarm and the potential factors affecting the performance of the cloud server into the fine-tuned large model. The fine-tuned large model outputs the abnormal status prediction results, the causes of the abnormality, and the relevant suggestions for maintenance.

[0109] It should be added that: the performance status data of the cloud server when the performance warning model issues an alarm is a quantitative indicator that reflects the real-time or time-period operation status of the cloud server, usually including CPU usage, memory usage, disk I / O speed, network latency, etc. They are a direct description of the current status of the server and can provide an instant picture of the server's health status; they are used to evaluate whether the performance of the cloud server at a specific point in time has reached the expected standard, and whether there are existing problems that need to be resolved;

[0110] Potential factors that affect the performance of cloud servers refer to internal and external conditions or environmental factors that may affect the performance of cloud servers, such as hardware configuration, operating system settings, application design, network bandwidth, and environmental conditions of data centers. These factors do not directly reflect current performance, but affect the performance of servers in the long term or periodically. By analyzing and adjusting these factors, it is possible to prevent performance problems that may occur in the future or conduct root cause analysis of existing problems. For example, upgrading hardware configuration, optimizing network settings, or refactoring software code to improve performance.

[0111] Embodiment 2:

[0112] Combined with Figure 2 , this embodiment proposes a cloud server health management system based on a large model, which includes:

[0113] A data collection module is used to collect the performance status data of the cloud server, as well as the maintenance records and historical report information of the cloud server, wherein the historical report information includes the historical abnormal operation status, the cause of the abnormality and the maintenance measures taken;

[0114] A preprocessing module, used to preprocess the collected cloud server performance status data;

[0115] A data partitioning module is used to randomly partition the pre-processed cloud server performance status data into a training set and a test set;

[0116] Model building module, used to build a GRU model based on error compensation;

[0117] The training and testing module is used to train the GRU model using the training set and to evaluate the generalization ability and prediction accuracy of the GRU model using the test set;

[0118] A model building module is used to build a performance warning model using the output of the GRU model;

[0119] Performance warning model, used to determine the operating status of the cloud server and issue an alarm when an abnormality is detected in the cloud server;

[0120] A format conversion module is used to convert the collected cloud server maintenance records and historical report information data into different formats;

[0121] Model fine-tuning module, which is used to use the pre-trained large model based on Transformer as the base model, such as ChatGLM and BaiChuan, and use the format-converted data to fine-tune the large model;

[0122] The fine-tuned large model is used to take the performance status data of the cloud server when the performance warning model issues an alarm as input, and output the abnormal status prediction results, the causes of the abnormality, and related maintenance suggestions.

[0123] In this embodiment, the preprocessing module involved preprocesses the collected cloud server performance status data, and the specific process includes:

[0124] Data cleaning: Check the integrity of the data, remove or fill missing values, remove irrelevant features, and handle outliers;

[0125] Data standardization: standardize the data to make it have a standard normal distribution;

[0126] Data coding: Encode categorical data;

[0127] Sequence data reconstruction: For time series data, according to the requirements of the GRU model, organize the data into a sequence form to ensure that the shape of the input data meets the expectations of the GRU model;

[0128] Subsequently, the data partitioning module randomly divides the preprocessed cloud server performance status data into a training set and a test set in a ratio of 8:2 or 7:3.

[0129] As we all know, the combination of Figure 4The GRU (Gated Recurrent Unit) model is a recurrent neural network (RNN) architecture designed to solve the long-term dependency problem in standard RNNs. GRU introduces a gating mechanism to control the flow of information, allowing the model to better capture long-term dependencies in sequence data. The core idea of ​​GRU is to use the update gate and reset gate to control the transmission of information. The update gate determines how much previous information needs to be retained in the current state, while the reset gate determines how much previous information to ignore. Both gates are calculated using a sigmoid function, and their output values ​​are between 0 and 1, indicating the degree of information retention or ignoring.

[0130] The workflow of GRU is roughly as follows:

[0131] Reset gate: First, the reset gate calculates a reset gate value based on the current input and the hidden state at the previous moment. This value determines how much information in the hidden state at the previous moment needs to be ignored.

[0132] Candidate hidden state: Next, the model uses the previous hidden state after the reset gate and the current input to calculate the candidate hidden state. This candidate hidden state contains the information of the current input and the information of the previous hidden state after the reset.

[0133] Update gate: The update gate then calculates an update gate value based on the current input and the hidden state at the previous moment. This value determines how much information in the current hidden state comes from the hidden state at the previous moment, and how much information comes from the newly calculated candidate hidden state.

[0134] Final hidden state: Finally, the model uses the update gate to combine the hidden state of the previous moment and the candidate hidden state to calculate the final hidden state of the current moment. This hidden state contains both the information of the current input and the information of the previous sequence, so that it can better handle long-term dependency problems. The output expression of each unit gate is as follows:

[0135] z t =σ(W z ·[h t-1 , x t ]) Formula (1),

[0136] r t =σ(W r ·[h t-1 , x t ]) Formula (2),

[0137]

[0138] Combined with Figure 3 ,The GRU model constructed in this embodiment includes N GRU submodules, each GRU submodule is a standard GRU network that can process sequence data and capture long-term dependencies in time series;

[0139] The input of each GRU submodule is the original input data X of the training set;

[0140] The prediction output of the first GRU submodule (GRU Module 1) is the cloud server performance indicator data The target output, i.e. the actual output, is the cloud server performance indicator data Y;

[0141] Output the prediction of the first GRU submodule (GRU Module 1) The residual of the target output Y As the target output of the second GRU submodule (GRU Module 2), the predicted output of the second GRU submodule (GRUModule 2) is

[0142] Output the prediction of the second GRU submodule (GRU Module 2) and target output ΔY 1 The residual As the target output of the third GRU submodule (GRU Module 3), the predicted output of the third GRU submodule (GRU Module 3) is

[0143] Continue to perform the above operations until the Nth GRU submodule (GRU Module N), and output the prediction of the N-1th GRU submodule (GRU Module N-1) and target output ΔY N-2 The residual As the target output of the Nth GRU submodule (GRU Module N), the predicted output of the Nth GRU submodule (GRU Module N) is

[0144] Finally, we get N prediction outputs The N prediction outputs are accumulated to form the final prediction result of the GRU model.

[0145] The training test module involved in this embodiment uses a test set to test the performance of the GRU model. During the test, the mean square error function is used as an evaluation indicator:

[0146]

[0147] Among them, RMSE represents the root mean square error, M represents the number of samples contained in the test set, and y i represents the target output value, Represents the predicted output value;

[0148] During the entire test process, the RMSE value is within the set threshold range, indicating that the GRU model test is qualified.

[0149] In this embodiment, when the performance warning model is established using the model establishment module, a threshold for distinguishing between a normal state and an abnormal state, and a time threshold for continuous operation of an abnormal state are set;

[0150] If the performance warning model determines that the cloud server is in an abnormal state based on the output of the GRU model, and the duration of the cloud server running in this state does not exceed the set time threshold, then the cloud server is determined to be in a normal state;

[0151] If the performance warning model determines that the cloud server is in an abnormal state based on the output of the GRU model, and the duration of the cloud server running in this state exceeds the set time threshold, it is determined that the cloud server is about to become abnormal. At this time, the performance warning model finds out the potential factors affecting the performance of the cloud server based on the changes in the GRU model input data, and inputs them into the fine-tuned large model.

[0152] It should be added that: the performance status data of the cloud server when the performance warning model issues an alarm is a quantitative indicator that reflects the real-time or time-period operation status of the cloud server, usually including CPU usage, memory usage, disk I / O speed, network latency, etc. They are a direct description of the current status of the server and can provide an instant picture of the server's health status; they are used to evaluate whether the performance of the cloud server at a specific point in time has reached the expected standard, and whether there are existing problems that need to be resolved;

[0153] Potential factors that affect the performance of cloud servers refer to internal and external conditions or environmental factors that may affect the performance of cloud servers, such as hardware configuration, operating system settings, application design, network bandwidth, and environmental conditions of data centers. These factors do not directly reflect current performance, but affect the performance of servers in the long term or periodically. By analyzing and adjusting these factors, it is possible to prevent performance problems that may occur in the future or conduct root cause analysis of existing problems. For example, upgrading hardware configuration, optimizing network settings, or refactoring software code to improve performance.

[0154] In summary, the cloud server health management method and system based on a large model of the present invention can realize the integration of different status information of the cloud server and output an effective cloud server health management strategy, including server fault prediction and warning and health status assurance, thereby realizing accurate warning of the cloud server status and improving the processing efficiency of operation and maintenance personnel.

[0155] The above specific examples are used to explain the principles and implementation methods of the present invention in detail. These examples are only used to help understand the core technical content of the present invention. Based on the above specific embodiments of the present invention, any improvements and modifications made by technicians in this technical field without departing from the principles of the present invention should fall within the scope of patent protection of the present invention.

Claims

1. A cloud server health management method based on a large model, characterized in that: The steps include: S1. Collecting performance status data of the cloud server, as well as maintenance records and historical report information of the cloud server, wherein the historical report information includes historical abnormal operation status, causes of abnormality, and maintenance measures taken; S2, preprocessing the collected cloud server performance status data, and randomly dividing the preprocessed cloud server performance status data into a training set and a test set; S3. Build a GRU model based on error compensation, use the training set to train the GRU model, and use the test set to evaluate the generalization ability and prediction accuracy of the GRU model; S4. Use the output of the GRU model to establish a performance warning model to determine the operating status of the cloud server and issue an alarm when an abnormality is detected in the cloud server; S5. Use the Transformer-based pre-trained large model as the base model, convert the collected cloud server maintenance records and historical report information data into different formats, and use the converted data to fine-tune the large model. The performance status data of the cloud server when the performance warning model issues an alarm is input into the fine-tuned large model, and the fine-tuned large model outputs the abnormal status prediction results, the causes of the abnormality and related maintenance suggestions.

2. A cloud server health management method based on a large model according to claim 1, characterized in that: Execute step S3, the constructed GRU model includes N GRU submodules, each of which is a standard GRU network that can process sequence data and capture long-term dependencies in time series; The input of each GRU submodule is the original input data of the training set; The prediction output of the first GRU submodule is the cloud server performance indicator data The target output, i.e. the actual output, is the cloud server performance indicator data; Output the prediction of the first GRU submodule The residual of the target output As the target output of the second GRU submodule, the predicted output of the second GRU submodule is Output the prediction of the second GRU submodule The residual with the target output As the target output of the third GRU submodule, the predicted output of the third GRU submodule is Continue to perform the above operations until the Nth GRU submodule, and output the prediction of the N-1th GRU submodule The residual with the target output As the target output of the Nth GRU submodule, the predicted output of the Nth GRU submodule is Finally, we get N prediction outputs The N prediction outputs are accumulated to form the final prediction result of the GRU model.

3. A cloud server health management method based on a large model according to claim 2, characterized in that: The performance of the GRU model is tested using the test set, and the mean square error function is used as the evaluation indicator during the test: Among them, RMSE represents the root mean square error, M represents the number of samples contained in the test set, represents the target output value, Represents the predicted output value; During the entire test process, the RMSE value is within the set threshold range, indicating that the GRU model test is qualified.

4. A cloud server health management method based on a large model according to claim 1, characterized in that: Execute step S4 to set a threshold for distinguishing between a normal state and an abnormal state, and a time threshold for continuous operation of an abnormal state in the performance warning model; If the performance warning model determines that the cloud server is in an abnormal state based on the output of the GRU model, and the duration of the cloud server running in this state does not exceed the set time threshold, then the cloud server is determined to be in a normal state; If the performance warning model determines that the cloud server is in an abnormal state based on the output of the GRU model, and the duration of the cloud server running in this state exceeds the set time threshold, it is determined that the cloud server is about to become abnormal. At this time, the performance warning model finds out the potential factors affecting the performance of the cloud server based on the changes in the GRU model input data, and inputs them into the fine-tuned large model.

5. A cloud server health management method based on a large model according to claim 1, characterized in that: Execute step S2 to pre-process the collected cloud server performance status data, specifically including: Data cleaning: Check the integrity of the data, remove or fill missing values, remove irrelevant features, and handle outliers; Data standardization: standardize the data to make it have a standard normal distribution; Data coding: Encode categorical data; Sequence data reconstruction: For time series data, according to the requirements of the GRU model, organize the data into a sequence form to ensure that the shape of the input data meets the expectations of the GRU model; Subsequently, the preprocessed cloud server performance status data are randomly divided into training sets and test sets in a ratio of 8:2 or 7:

3.

6. A cloud server health management system based on a large model, characterized in that: It includes: A data collection module is used to collect the performance status data of the cloud server, as well as the maintenance records and historical report information of the cloud server, wherein the historical report information includes the historical abnormal operation status, the cause of the abnormality and the maintenance measures taken; A preprocessing module, used to preprocess the collected cloud server performance status data; A data partitioning module is used to randomly partition the pre-processed cloud server performance status data into a training set and a test set; Model building module, used to build a GRU model based on error compensation; The training and testing module is used to train the GRU model using the training set and to evaluate the generalization ability and prediction accuracy of the GRU model using the test set; A model building module is used to build a performance warning model using the output of the GRU model; Performance warning model, used to determine the operating status of the cloud server and issue an alarm when an abnormality is detected in the cloud server; A format conversion module is used to convert the collected cloud server maintenance records and historical report information data into different formats; The model fine-tuning module is used to use the Transformer-based pre-trained large model as the base model and use the format-converted data to fine-tune the large model; The fine-tuned large model is used to take the performance status data of the cloud server when the performance warning model issues an alarm as input, and output the abnormal status prediction results, the causes of the abnormality, and related maintenance suggestions.

7. A cloud server health management system based on a large model according to claim 6, characterized in that: The GRU model constructed by the model building module includes N GRU submodules, each of which is a standard GRU network that can process sequence data and capture long-term dependencies in time series; The input of each GRU submodule is the original input data of the training set; The prediction output of the first GRU submodule is the cloud server performance indicator data The target output, i.e. the actual output, is the cloud server performance indicator data; Output the prediction of the first GRU submodule The residual of the target output As the target output of the second GRU submodule, the predicted output of the second GRU submodule is Output the prediction of the second GRU submodule The residual with the target output As the target output of the third GRU submodule, the predicted output of the third GRU submodule is Continue to perform the above operations until the Nth GRU submodule, and output the prediction of the N-1th GRU submodule The residual with the target output As the target output of the Nth GRU submodule, the predicted output of the Nth GRU submodule is Finally, we get N prediction outputs The N prediction outputs are accumulated to form the final prediction result of the GRU model.

8. A cloud server health management system based on a large model according to claim 7, characterized in that: The training test module uses the test set to test the performance of the GRU model. During the test, the mean square error function is used as the evaluation index: Among them, RMSE represents the root mean square error, M represents the number of samples contained in the test set, represents the target output value, Represents the predicted output value; During the entire test process, the RMSE value is within the set threshold range, indicating that the GRU model test is qualified.

9. A cloud server health management system based on a large model according to claim 6, characterized in that: When using the model building module to build a performance warning model, set the threshold for distinguishing between normal and abnormal states, as well as the time threshold for the abnormal state to continue running; If the performance warning model determines that the cloud server is in an abnormal state based on the output of the GRU model, and the duration of the cloud server running in this state does not exceed the set time threshold, then the cloud server is determined to be in a normal state; If the performance warning model determines that the cloud server is in an abnormal state based on the output of the GRU model, and the duration of the cloud server running in this state exceeds the set time threshold, it is determined that the cloud server is about to become abnormal. At this time, the performance warning model finds out the potential factors affecting the performance of the cloud server based on the changes in the GRU model input data, and inputs them into the fine-tuned large model.

10. A cloud server health management system based on a large model according to claim 6, characterized in that: The preprocessing module preprocesses the collected cloud server performance status data, and the specific process includes: Data cleaning: Check the integrity of the data, remove or fill missing values, remove irrelevant features, and handle outliers; Data standardization: standardize the data to make it have a standard normal distribution; Data coding: Encode categorical data; Sequence data reconstruction: For time series data, organize the data into a sequence form according to the requirements of the GRU model to ensure that the shape of the input data meets the expectations of the GRU model; Subsequently, the data partitioning module randomly divides the preprocessed cloud server performance status data into a training set and a test set in a ratio of 8:2 or 7:3.

Citation Information

Cited By

  • Server health prediction method and system based on multi-time sequence characteristics

    CN122431983A