Server health state diagnosis method based on GAT-LP algorithm

Abstract

The invention discloses a server health state diagnosis method based on a GAT-LP algorithm, and relates to the technical field of server health diagnosis, and the method comprises the steps: collecting multi-dimensional operation state data of a server, transmitting the multi-dimensional operation state data to a data analysis platform, and carrying out the preprocessing, inputting the preprocessed historical operation state data of the server into a GAT-LP network for training to obtain a GAT-LP network model, evaluating health conditions of the server and the process and generating an overall health condition according to a health state diagnosis result of the server and health score features extracted based on the GAT-LP network model, converting an evaluation result into a visual health state, and displaying the visual health state in the GAT-LP network model. And triggering an alarm mechanism when the health state of the server is lower than a preset threshold value. By constructing a multi-dimensional health assessment index system, the running state of the server is visually displayed, potential risks are found in advance, and prevention measures are made.

Description

Technical Field

[0001] The present invention relates to the technical field of server health diagnosis, and particularly to a method for diagnosing the health status of a server based on the GAT-LP algorithm. Background Art

[0002] In today's information technology field, as the central infrastructure supporting various network services and business operations, the stability and performance of servers have a decisive impact on service quality. With the rapid development of emerging technologies such as cloud computing, big data processing, and the Internet of Things, the complexity of tasks and data traffic carried by servers are increasing day by day, which poses higher requirements for the reliability and continuous availability of servers. Traditional server monitoring methods mainly rely on the monitoring of static indicators, such as CPU utilization rate, memory occupancy rate, disk I / O frequency, etc. Although these methods can provide a certain degree of feedback on the running state, they usually can only reflect the instantaneous resource usage and are difficult to capture the dynamic change trend of the server health status and its potential problems.

[0003] In recent years, methods based on machine learning and deep learning have gradually been applied to server health status diagnosis, and prediction models are constructed to estimate the health status of servers. For example, time series analysis models can be used to capture the time correlation of server performance indicators; while expert systems based on neural networks can identify the change patterns of server health during long-term operation and perform failure prediction. However, the above methods have certain limitations. Time series models mostly focus on the extraction of data features from a single source and lack in-depth exploration of the correlation between multi-source data; some existing solutions only perform health assessment based on the data at the current moment and fail to make full use of the dynamic information in historical data, thus limiting the comprehensiveness and timeliness of diagnosis. In addition, some solutions can only diagnose the long-term running state of servers, cannot effectively extract the relevant features of server state fluctuations, and their accuracy in high-load scenarios also needs to be improved. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method for diagnosing the health status of a server based on the GAT-LP algorithm to solve the problems that the health status of the server cannot be accurately evaluated in real time and multi-source data cannot be effectively integrated.

[0006] To solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides a method for diagnosing the health status of a server based on the GAT-LP algorithm, including,

[0008] Collect multi-dimensional operation status data of the acquisition server;

[0009] Transmit the multi-dimensional operation status data to the data analysis platform and perform preprocessing;

[0010] Input the preprocessed historical operation status data of the server into the GAT-LP network for training to obtain the GAT-LP network model;

[0011] According to the server health status diagnosis result, based on the health score features extracted by the GAT-LP network model, evaluate the health status of the server and processes and generate the overall health status;

[0012] Convert the evaluation result into a visual health status;

[0013] When the server health status is lower than the preset threshold, trigger the alarm mechanism, trace the change of the server operation status before the fault occurs after the fault occurs, generate a fault timeline, and generate a diagnosis suggestion and a repair plan according to the tracing result.

[0014] As a preferred solution of the server health status diagnosis method based on the GAT-LP algorithm described in the present invention, wherein: the step of collecting multi-dimensional operation status data of the acquisition server is specifically as follows,

[0015] Set the collection time interval, and regularly collect the total CPU occupancy rate, total memory occupancy rate, main process CPU occupancy rate, memory occupancy rate, network traffic and disk read and write conditions of the server;

[0016] Perform preliminary processing on the collected multi-dimensional operation status data at the edge computing node;

[0017] Perform data compression on the preliminarily processed data through a data compression algorithm and generate a compressed package;

[0018] Store the compressed package in the local database of the server.

[0019] As a preferred solution of the server health status diagnosis method based on the GAT-LP algorithm described in the present invention, wherein: the step of transmitting the multi-dimensional operation status data to the data analysis platform and performing preprocessing is specifically as follows,

[0020] Transmit the compressed package in the local database of the server to the data analysis platform;

[0021] After the data analysis platform receives the compressed package, use the data compression algorithm to decompress the compressed package;

[0022] Detect and correct outliers for the data after decompressing the compressed package through rule detection, and use the interpolation method to fill in the missing values to obtain the historical operation status data of the server;

[0023] Standardize the historical operating status data of the server.

[0024] As a preferred solution of the server health status diagnosis method based on the GAT-LP algorithm of the present invention, wherein: input the preprocessed historical operating status data of the server into the GAT-LP network for training to obtain the GAT-LP network model. The specific steps are as follows:

[0025] Convert the standardized historical operating status data of the server into a graph structure, and define the node feature vector and the adjacency matrix;

[0026] Input the node feature matrix vector and the adjacency matrix into the GAT network layer, calculate the correlation relationship between nodes, and generate the feature matrix;

[0027] Input the feature matrix output by the GAT layer into the LSTM network layer to capture the time series features;

[0028] Use the fully connected layer to reduce the dimension of the time series features of the LSTM layer;

[0029] Adopt the parrot optimization algorithm to find the optimal hyperparameter combination to obtain the GAT-LP network model.

[0030] As a preferred solution of the server health status diagnosis method based on the GAT-LP algorithm of the present invention, wherein: the specific steps of adopting the parrot optimization algorithm to find the optimal hyperparameter combination are as follows:

[0031] Define N to represent the number of individuals searching for the optimal hyperparameter configuration in the parrot optimization algorithm;

[0032] Initialize the hyperparameter search space and randomly generate N parrot individuals, and each parrot represents a set of hyperparameter configurations;

[0033] Evaluate and train the hyperparameters corresponding to each parrot, record the fitness value of the validation set, and select the parrot position corresponding to the maximum value of all parrot fitnesses through the fitness value of the validation set;

[0034] Update the parrot position according to the parrot position corresponding to the maximum value of all parrot fitnesses and the group average position, and perform any one of the behaviors of foraging, staying, communicating, and fearing to update the position;

[0035] Check whether the updated position of the parrot at the next moment exceeds the boundary value. If it exceeds, set it to the boundary value and recalculate the fitness value;

[0036] Repeat the iterative optimization process, output the optimal hyperparameter combination, and use the optimal hyperparameters to train the GAT-LP network.

[0037] As a preferred solution of the server health status diagnosis method based on the GAT-LP algorithm of the present invention, wherein: according to the server health status diagnosis result and the health score features extracted based on the GAT-LP network model, evaluate the health status of the server and processes and generate the overall health status. The specific steps are as follows,

[0038] Input the server historical operation status data and real-time operation status data into the GAT-LP network model, calculate and extract the health score features;

[0039] Based on the health score features extracted by the GAT-LP network model, evaluate the overall health status of the server and processes;

[0040] According to the evaluation result of the overall health status of the server and processes, generate the server and process health status report.

[0041] As a preferred solution of the server health status diagnosis method based on the GAT-LP algorithm of the present invention, wherein: the specific steps of converting the evaluation result into a visual health status are as follows,

[0042] Use a Web development framework and a front-end library to build a visual interface framework, select a visualization tool, and set a health status overview area, a detailed indicator display area, and an alarm prompt area in the visual interface framework;

[0043] Extract the health score features from the GAT-LP network model and map them to a color coding table, and convert the color coding table into a visual server and process health status report;

[0044] Integrate the visual server and process health status report data and perform weighted aggregation processing;

[0045] Use a Python library to generate a bar chart from the weighted aggregated server and process health status report data.

[0046] As a preferred solution of the server health status diagnosis method based on the GAT-LP algorithm of the present invention, wherein: when the server health status is lower than a preset threshold, trigger an alarm mechanism, trace the change of the server operation status before the fault occurs after the fault occurs, generate a fault timeline, and generate a diagnosis suggestion and a repair plan according to the tracing result. The specific steps are as follows,

[0047] Set an initial safety threshold M;

[0048] Through the time series analysis of the historical health index, dynamically adjust the initial safety threshold M;

[0049] The server calculates the historical health index data once every minute and compares it with the adjusted preset threshold M;

[0050] If the health index is lower than the preset threshold M, start the alarm process and generate a description document of the cause of the fault and diagnostic suggestions;

[0051] According to the degree of deterioration of the health index, trigger email, SMS and phone warnings in sequence to notify the technical support personnel;

[0052] Restart the server when the health index reaches the critical point;

[0053] After a fault occurs, trace the changes in the server running state before the fault occurs and generate a fault timeline;

[0054] Generate diagnostic suggestions and repair solutions according to the tracing results.

[0055] In a second aspect, the present invention provides a computer device, including a memory and a processor, the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the server health state diagnosis method based on the GAT-LP algorithm as described in the first aspect of the present invention is implemented.

[0056] In a third aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, wherein: when the computer program is executed by the processor, any step of the server health state diagnosis method based on the GAT-LP algorithm as described in the first aspect of the present invention is implemented.

[0057] The beneficial effects of the present invention are as follows: By setting the acquisition time interval, the present invention regularly acquires multi-dimensional server running state data and compresses and stores it, effectively reducing the transmission bandwidth requirement; after the compressed data is transmitted to the data analysis platform, through decompression, outlier correction, missing value filling and standardization processing, the data quality and consistency are improved. The preprocessed data is converted into a graph structure and input into the analysis process by defining node feature vectors and adjacency matrices. Combining the graph attention mechanism with time series analysis technology, the features of multi-source time series data are deeply mined. The parrot optimization algorithm is used to automatically tune parameters, avoid local optima, and improve the robustness of the analysis results. Further, by constructing a multi-dimensional health assessment index system, the server running state is intuitively displayed, potential risks are discovered in advance and preventive measures are formulated. When the health state is lower than the threshold, the alarm mechanism is automatically triggered to provide fault diagnosis suggestions; when the health index reaches the critical point, the server is automatically restarted, thereby ensuring service continuity and stable business operation. Description of the Drawings

[0058] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0059] Figure 1 It is a flowchart of the server health status diagnosis method based on the GAT-LP algorithm in Embodiment 1.

[0060] Figure 2 It is a flowchart of training the GAT-LP network to obtain the GAT-LP network model in Embodiment 1. Detailed implementation manners

[0061] To make the above objects, features, and advantages of the present invention more obvious and understandable, the following will make a detailed description of the specific implementation manners of the present invention in conjunction with the drawings in the specification.

[0062] In the following description, many specific details are set forth to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Those skilled in the art can make similar extensions without departing from the connotation of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0063] Secondly, the so-called "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that can be included in at least one implementation of the present invention. The "in one embodiment" that appears in different places in this specification does not all refer to the same embodiment, nor is it a separate or alternative embodiment that mutually excludes other embodiments.

[0064] Embodiment 1, referring to Figure 1 and Figure 2 , is the first embodiment of the present invention. This embodiment provides a server health status diagnosis method based on the GAT-LP algorithm, including the following steps:

[0065] S1. Collect the operation status data of the server in multiple dimensions.

[0066] Furthermore, set the collection time interval and regularly collect the total CPU occupancy rate, total memory occupancy rate, main process CPU occupancy rate, memory occupancy rate, network traffic, and disk read and write conditions of the server.

[0067] It should be noted that by setting the collection time interval and regularly collecting the multi-dimensional operation status data of the server, continuous and efficient data accumulation of the server performance is achieved. This step ensures that subsequent analysis can be based on comprehensive and accurate historical data, providing a solid data foundation for health status diagnosis. Specifically, by defining a fixed time interval (such as every minute or every hour), the system can regularly obtain key performance indicators (KPIs), including but not limited to CPU utilization, memory usage, network bandwidth consumption, and disk I / O rate, etc. This periodic data collection method not only covers the performance of the server under different loads but also captures instantaneous changes that may affect its stability, thus providing detailed data support for subsequent in-depth analysis.

[0068] Perform preliminary processing on the multi-dimensional operation status data collected at the edge computing node.

[0069] It should be noted that the collected data is first transmitted to the edge computing node for preliminary processing. The edge computing node uses predefined rules or simple machine learning models to detect outliers and diagnose local faults in the data. For example, it detects whether the CPU occupancy rate exceeds the preset threshold, detects memory leaks or abnormal memory occupancy, detects abnormal fluctuations in network traffic, and detects abnormal disk read and write speeds.

[0070] Compress the preliminarily processed data through a data compression algorithm and generate a compressed package.

[0071] It should be noted that by applying a data compression algorithm to the collected data, effective optimization of the data storage space is achieved, and the requirement for data transmission bandwidth is reduced. Data compression technologies (such as the LZ77 series algorithms) can significantly reduce the volume of the original data without losing information. For large-scale data centers, this process not only reduces the storage cost but also speeds up the data transmission speed, especially important when transmitting a large amount of monitoring data across regions or networks. In addition, the compressed data format is usually easier to manage and process, helping to improve the efficiency of the entire system.

[0072] Store the compressed package in the local database of the server.

[0073] It should be noted that by storing the compressed data packets in the local database of the server, efficient data management and rapid access are achieved. This step ensures that all collected data can be securely stored in a centralized location, facilitating subsequent retrieval, query, and analysis operations. The design of the local database takes into account the high concurrent read and write requirements and adopts an efficient indexing mechanism to accelerate data positioning, enabling rapid response to user requests even in the face of a large amount of historical records. In addition, local storage can also serve as a temporary cache, providing a stable intermediate layer before synchronizing with the remote data analysis platform to prevent data loss or delay caused by network fluctuations. This step not only enhances the robustness of the system but also lays a good foundation for subsequent preprocessing and model training.

[0074] S2. Transmit the multi-dimensional operation status data to the data analysis platform and perform preprocessing.

[0075] Furthermore, transmit the compressed package of the server local database to the data analysis platform.

[0076] Specifically, by transmitting the compressed package in the server local database to the data analysis platform, efficient migration of data from a distributed storage environment to a centralized processing environment is achieved. This step utilizes modern network transmission protocols and optimized data transmission algorithms to ensure the stability and speed during the transmission of a large amount of data. This process not only reduces bandwidth occupancy but also improves the security of data transmission, avoiding the risk of unauthorized access. Its role is to provide a reliable data source for subsequent data processing, ultimately ensuring data integrity and availability.

[0077] After receiving the compressed package, the data analysis platform uses a data compression algorithm to decompress the compressed package.

[0078] It should be noted that by the data analysis platform receiving and decompressing the compressed package from the server local database, rapid recovery of the original data is achieved. In this step, an efficient decompression algorithm (such as Lempel-Ziv-Welch, LZW) is adopted to ensure the speed and accuracy of decompression. In addition, the decompressed data can immediately enter the preprocessing stage without additional waiting time. The role of this step is to prepare for the subsequent data cleaning and standardization processing, ultimately achieving the effect of improving the overall process efficiency.

[0079] After decompressing the data in the compressed package, detect and correct outliers through rules and use interpolation methods to fill in missing values to obtain the historical operation status data of the server.

[0080] It should be noted that for the possible outliers and missing values in the data after decompressing the compressed package, the specific steps for correction and filling are as follows.

[0081] For the i-th data point Ni , find the 6 nearest neighboring points {N i1 , N i2 , …, N i6} in terms of time;

[0082] Calculate the mean absolute difference between N i and its 6 nearest neighboring points

[0083] Calculate the average value E of all avg ;

[0084] If there exists a data point N i such that then this data point is considered an outlier;

[0085] For outliers and missing values, take the average value of the two nearest points for interpolation and filling. Finally, complete server historical operation status data is obtained. The role of this process is to eliminate noise interference, improve the accuracy and reliability of model training, and ultimately achieve the purpose of providing high-quality input data.

[0086] Perform standardization processing on the server historical operation status data.

[0087] It should be noted that by performing standardization processing on the server historical operation status data, the present invention ensures that data in different dimensions has the same scale, thereby improving the consistency and stability of model training. Standardization processing makes it possible to compare features on the same scale, avoiding inappropriate impacts on the model caused by some features having too large or too small numerical ranges. The role of this step is to enhance the learning ability and generalization performance of the model, and ultimately achieve the effect of improving the diagnostic accuracy.

[0088] S3. Input the preprocessed server historical operation status data into the GAT-LP network for training to obtain the GAT-LP network model.

[0089] Furthermore, convert the standardized server historical operation status data into a graph structure, and define the node feature vector and the adjacency matrix.

[0090] Specifically, convert the server historical operation data, i.e., multi-dimensional time series data, into a graph structure. The nodes in the graph at time t are defined as the node feature vectors composed of the historical operation data of a certain type of server parameter (such as the total CPU occupancy rate C p ) at time t and a period of time before it, a total of F time instants. The edges of the graph are defined as the relationships between N different types of server parameters, which can be represented by the adjacency matrix A, where the element α ij in A represents the correlation between node i and node j, and αij = 1 indicates the strongest correlation, α ij = 0 indicates the weakest correlation;

[0091] This step not only preserves the time series characteristics of the original data but also provides a structured basis for calculating the relationships between subsequent nodes. Its function is to capture the complex correlations between data in different dimensions, enabling the model to more comprehensively understand the dynamic behavior of the server, thereby improving the accuracy of health status diagnosis.

[0092] Input the node feature matrix vector and the adjacency matrix into the GAT network layer to calculate the correlation between nodes and generate a feature matrix.

[0093] Specifically, input the defined node feature vectors to form an F×N-dimensional node feature matrix and the adjacency matrix A into the GAT network layer. Define the weight matrix W, which represents the relationship between the input feature matrix h and the output feature matrix h ′ Assume that node j is a neighbor of node i, then the correlation between node j and node i can be calculated using the attention mechanism, where the attention coefficient e ij is defined as

[0094]

[0095] where S is the attention scoring function, which consists of a feed-forward neural network stacked with the LeakyReLu function, and α ij is the attention coefficient after regularization processing. This operation makes the attention coefficient easier to calculate and compare. After the above operations, the attention coefficients of different nodes are obtained, which are used to predict the output features of each node.

[0096]

[0097] where Φ is the ReLu activation function, and j ∈ N i is limited to all nodes related to node i. Thus, the output of the GAT network layer is obtained, which is a new node feature matrix

[0098] Input the feature matrix output by the GAT layer into the LSTM network layer to capture time series features.

[0099] Specifically, the F×N - dimensional output vector at time t in (2) is pooled and fused into an F×1 - dimensional feature vector, which is input into two LSTM network layers. Each LSTM network layer is composed of several LSTM cells connected in series. Information is passed backward through the LSTM cells in sequence. Each LSTM cell contains three gating units, namely, the forget gate, the input gate, and the output gate, which can pass or forget historical information controllably. The forget gate is implemented by a Sigmoid network layer, and its input is the hidden state h at the previous time step t-1 and the input x at the current time step t , and the output value f t falls between 0 and 1, which is used to control how much content in the state needs to be forgotten and determines the proportion of information to be retained. The expression is

[0100] f t =σ(W f ·[h t-1 ,x t +b f );

[0101] where W f is the weight matrix of the forget gate, b f is the bias vector, and σ is the Sigmoid activation function

[0102] The input of the input gate is the hidden state h at the previous time step t-1 and the input x at the current time step t . First, the input passes through the tanh network layer to generate a state Then the Sigmoid network layer controls whether to use to update the state value and how much information to update, and obtains the output i t . The old state C of the input neuron t-1 is multiplied by f t to obtain the information after forgetting, and then added with the updated value to obtain a new state, which is used to be passed to the next unit. The expression is

[0103]

[0104] i t =σ(W i ·[h t-1 ,x t +b i );

[0105]

[0106] where W C is the weight matrix of the tanh network layer, b C is the bias vector, W iis the weight matrix of the Sigmoid network layer, and b i is the bias vector.

[0107] The output gate multiplies the current new state, which has been processed by the tanh network layer, with the Sigmoid network layer to control the output content. The expression is

[0108] o t = σ(W o [h t-1 , x t + b o );

[0109] h t = o t * tanh(C t );

[0110] Among them, W o is the weight matrix of the Sigmoid network layer, b o is the bias vector. The hidden state h t output by each LSTM cell in the first-layer LSTM network is used as the input of the second-layer LSTM cell. After all training data are input into the network in sequence, the hidden state h end of the last cell in the second-layer LSTM network is obtained;

[0111] Through this mechanism, the LSTM layer can effectively capture the long-term dependencies in the server running state, enabling the model to have stronger time series analysis capabilities, improving the accuracy of prediction, especially performing well when dealing with data with long time spans.

[0112] The time series features of the LSTM layer are processed by a fully connected layer for dimensionality reduction.

[0113] Specifically, the fully connected layer is a common layer structure in neural networks. Each neuron in it is connected to all neurons in the previous layer. The role of the fully connected layer is to perform linear transformation and non-linear activation on the input features, so as to extract higher-level features or perform dimensionality reduction. In the GAT-LP network model, the fully connected layer is used to perform dimensionality reduction on the time series features output by the LSTM layer. Each neuron in the fully connected layer is connected to all neurons in the previous layer. Through linear transformation and non-linear activation function (such as ReLU), the high-dimensional features are mapped to a low-dimensional space. The dimensionality-reduced feature vector will be used for subsequent health score feature calculation or classification tasks. In this step, the fully connected layer is used to perform dimensionality reduction on the time series features output by the LSTM layer, mapping the high-dimensional features to a low-dimensional space, so that subsequent health score feature calculation or classification tasks can utilize the fully connected layer to reduce the dimensionality of the high-dimensional data output by the LSTM network layer, and the hidden state h of the last cell of the second-layer LSTM networkend Connected to the fully connected layer, which contains a weight matrix W and a bias vector b, and can map the input to the final output Y. The expression is

[0114]

[0115] where H t is the input feature of the current layer of the neural network;

[0116] When training the above neural network, it is necessary to set the network hyperparameters. The hyperparameters include the learning rate lr ∈ LR, the batch size bs ∈ BS, the input sequence length lo ∈ LO, etc. Among them,

[0117] LR = {lr 1 , lr 2 , …, lr n};

[0118] BS = {bs 1 , bs 2 , …, bs n};

[0119] LO = {lo 1 , lo 2 , …, lo n};

[0120] respectively represent the hyperparameter search spaces of the learning rate lr, the batch size bs, and the input sequence length lo;

[0121] The role of the fully connected layer is to reduce the feature dimension, making the model easier to train and avoiding overfitting. The feature vector after dimensionality reduction can better summarize the running state of the server, reducing the impact of redundant information, while improving the computational efficiency of the model. In addition, dimensionality reduction can also enhance the generalization ability of the model, making it more stable when facing new data.

[0122] Define N to represent the number of individuals searching for the optimal hyperparameter configuration in the parrot optimization algorithm. Use the parrot optimization algorithm to find the optimal hyperparameter combination to obtain the GAT-LP network model.

[0123] Specifically, initialize the hyperparameter search space and randomly generate N parrot individuals, and each parrot represents a set of hyperparameter configurations;

[0124] Evaluate and train the hyperparameters corresponding to each parrot, record the fitness value of the validation set, and select the parrot position corresponding to the maximum value of all parrot fitnesses through the fitness value of the validation set;

[0125] Update the parrot position according to the parrot position corresponding to the maximum value of all parrot fitnesses and the group average position, and perform any one of the foraging, staying, communicating, and fear behaviors to update the position;

[0126] Check whether the updated position of the parrot at the next moment exceeds the boundary value. If it exceeds, set it to the boundary value and recalculate the fitness value;

[0127] Repeat the iterative optimization process, output the optimal hyperparameter combination, and use the optimal hyperparameters to train the GAT-LP network.

[0128] It should be noted that a parrot group composed of multiple parrot individuals is created, and the positions of the parrot group are initialized. The expression is,

[0129]

[0130] Among them, is the initial position of the i-th parrot, lb is the lower limit of the search space, and ub is the upper limit of the search space. Taking the learning rate lr as an example, set the lower limit of the search space lb = 10 -4 , and the upper limit of the search space ub = 10 -1 . Subsequently, each parrot randomly adopts one of the following behaviors,

[0131] Foraging behavior

[0132] In the foraging behavior, the parrot estimates the approximate position of the food by observing the position of the food or considering the position of the owner, and then flies towards its respective target. The expression is,

[0133]

[0134] Among them, represents the current position, represents the position updated subsequently, represents the average position within the parrot group, dim represents the dimension of the parameter to be optimized, Levy(dim) represents the Levy distribution, which is used to describe the flight of the parrot, X best represents the best position from the initial position to the current position, that is, the position of the owner. t represents the current iteration number, represents moving from its own position to the position of the owner, Max iter represents the preset maximum time step of iteration, represents moving from its own position to the group average position, and the parrot determines the food orientation through both;

[0135] The group average position is shown in the following formula,

[0136]

[0137] Among them, the Levy distribution is represented by the following rules, where γ is assigned a value of 1.5,

[0138]

[0139] Stay behavior,

[0140] The stay behavior of the parrot is defined as flying to any part of the owner's body and staying for a period of time, as shown in the following formula,

[0141]

[0142] where X best ·Levy(dim) represents moving towards the owner, and rand(0,1)·ones(1,dim) represents staying at a random position on the owner's body.

[0143] Communication behavior,

[0144] The communication behavior of the parrot includes flying towards the group and leaving the group after communication. The algorithm assumes that the probabilities of the two behaviors occurring are equal, as shown in the following formula,

[0145]

[0146] where, represents the process of a parrot joining the group communication, represents the process of a parrot leaving after group communication, and P represents a randomly generated choice, with a range of (0,1).

[0147] Fear behavior,

[0148] The fear behavior is defined as the behavior of the parrot moving away from strangers and approaching the owner, as shown in the following formula,

[0149]

[0150] where, represents the process of redirecting to fly towards the owner, represents the process of moving away from strangers.

[0151] After the parrot selects a certain behavior, the position of the parrot at the next moment is updated based on the selected behavior. If the updated position exceeds the boundary value of its corresponding hyperparameter, as shown in the following formula,

[0152]

[0153] then its position at the next moment is set to this boundary value, as shown in the following formula,

[0154]

[0155] Then, the neural network is trained using the hyperparameters corresponding to the updated parrot positions, and the fitness of each parrot in the parrot population is calculated. where N is the number of samples in the training set, and y i is the true label of the data in the training set, is the predicted value of the neural network. The parrot position corresponding to the maximum value of all parrot fitnesses is taken as the optimal solution of the neural network hyperparameters at the current moment. Subsequently, each parrot randomly selects one of the above four behaviors again, and this process is repeated until t reaches the maximum iteration time step Max. iter After optimizing each neural network hyperparameter, the neural network is finally trained using the optimal hyperparameter combination, and then the neural network can be used for server health status diagnosis.

[0156] Integrating the above process, a GAT-LP network model capable of diagnosing the server health status can be obtained, which improves the generalization ability and robustness of the model, ensuring that it can still maintain high performance in complex environments.

[0157] S4. According to the server health status diagnosis result, based on the health score features extracted by the GAT-LP network model, evaluate the health status of the server and processes and generate the overall health status.

[0158] Furthermore, input the server historical operation status data and real-time operation status data into the GAT-LP network model, calculate and extract the health score features.

[0159] Specifically, the system uses the pre-trained GAT-LP network model to process the standardized multi-dimensional historical operation status data and generate a feature vector that can reflect the current health status of the server and its processes. The GAT-LP model combines the graph attention network (GAT) and the long short-term memory network (LSTM), which can effectively capture the correlation between multi-source data and the dynamic changes in the time series. This process not only considers the immediate data but also combines the time series characteristics of historical data, making the health score features more comprehensive and accurate. Its role is to provide a solid data foundation for subsequent health assessments, ensuring the reliability and accuracy of the diagnosis results.

[0160] Based on the health score features extracted by the GAT-LP network model, evaluate the overall health status of the server and processes.

[0161] Specifically, in this step, a multi-dimensional health assessment index system is constructed based on the health score features obtained from the GAT-LP network model. This system includes the following key performance indicators: CPU occupancy rate, memory usage, disk read / write frequency, network traffic, process status, temperature, and power status. Each dimension reflects an aspect of the server's health status. By comprehensively considering the mutual influence between these dimensions, the GAT-LP network model can more comprehensively evaluate the health status of the server. For example, a high CPU occupancy rate may lead to an increase in memory usage, which in turn affects the disk read / write frequency and network traffic; abnormal network traffic may indicate a network attack or communication failure. Through the multi-dimensional assessment system, the GAT-LP network model can not only identify the status of individual components but also comprehensively evaluate the health level of the entire server system. In addition, the multi-dimensional assessment system can detect potential problems earlier, provide a more detailed analysis of the health status, thereby helping the operation and maintenance personnel take preventive measures in advance and improve the stability and reliability of the system.

[0162] Generate a server and process health status report based on the evaluation results of the overall health status of the server and processes.

[0163] Specifically, the historical operation status data or real-time operation status data is input into a specially designed GAT-LP network model. Combining the structural advantages of the GAT-LP network and the hyperparameter configuration optimized by the parrot optimization algorithm, the overall health status is evaluated. The GAT-LP network model not only relies on the powerful feature extraction ability of the GAT-LP network but also utilizes the best hyperparameter combination found by the parrot optimization algorithm to ensure the high generalization ability and robustness of the model in complex environments. The model outputs a numerical health score (ranging from 0 to 100) to quantify the health status of the server and its processes. The higher the score, the better the health status; the lower the score, the more potential problems exist. The health score can be presented in ways such as numerical scoring, color coding, and graphical display. By analyzing the degree of abnormality of each key indicator (such as CPU, memory, disk, etc.), possible failure causes are inferred. The failure causes are output in the form of descriptive text, such as "Performance degradation due to CPU overload" or "Resource exhaustion due to memory leak". Based on the analysis of the failure causes, specific diagnostic suggestions are generated, and the diagnostic suggestions are output in the form of operation guides, such as "Restart the relevant process" or "Increase memory resources". The health score features can be presented through numerical scoring: directly display a numerical value (such as "Health score feature: 85"), color coding: map the score to colors (such as green for healthy, yellow for warning, and red for serious problems), graphical display: use visualization tools such as bar charts and line charts to show the change trend of the score. The failure cause analysis can be presented through a list of abnormal indicators: list the degree of abnormality of each key indicator (such as "CPU occupancy rate: 95% (abnormal)"), and the failure cause description: provide a specific description of the failure cause (such as "Performance degradation due to CPU overload"). The diagnostic suggestions can be presented through a list of suggestions: list specific operation suggestions (such as "Restart the relevant process" or "Increase memory resources"), and operation guides: provide detailed operation steps or links to relevant documents. The operation and maintenance personnel can quickly locate problems and implement effective maintenance strategies based on the diagnostic suggestions. Finally, the overall health status report generated in this step not only provides intuitive health score features but also includes detailed failure cause analysis and diagnostic suggestions, which greatly facilitates the operation and maintenance personnel to quickly locate problems and implement effective maintenance strategies. This not only shortens the response time, reduces the losses caused by failures, but also enhances the self-healing ability of the system and ensures the continuity of business and service quality.

[0164] S5. Convert the evaluation result into a visual health status.

[0165] Furthermore, use a Web development framework and a front-end library to build a visual interface framework, select visualization tools, and set up a health status overview area, a detailed indicator display area, and an alarm prompt area in the visual interface framework.

[0166] Specifically, through this step, the present invention creates a visualization interface framework with a clear structure and user-friendliness. This framework not only integrates a variety of visualization tools but also particularly sets three key areas: the health status overview area, the detailed index display area, and the alarm prompt area. The health status overview area is used to provide a quick view of historical health index data; the detailed index display area shows the specific values and change trends of each key performance indicator; the alarm prompt area displays any potential problems or abnormal situations in real time. This design enables the operation and maintenance personnel to quickly obtain the operation overview of the server and delve into specific indicators when needed, thus improving the speed and accuracy of fault troubleshooting and maintenance decision-making.

[0167] Extract the health score features from the GAT-LP network model and map them to the color coding table, and convert the color coding table into a visual server and process health status report.

[0168] Specifically, in this step, the health score features are extracted from the GAT-LP network model and mapped to a predefined color coding table according to their values. Different colors represent different levels of health status (for example, green indicates normal, yellow indicates warning, and red indicates severe). This enables the historical health index data to be presented to users in an intuitive color form. In this way, even personnel without a professional background can identify the health status of the server at a glance, greatly simplifying the information interpretation process and enhancing the usability of the system. At the same time, the color coding mechanism also facilitates quickly locating the problem and improving the emergency response efficiency.

[0169] Integrate the visual server and process health status report data and perform weighted aggregation processing.

[0170] Specifically, in order to more accurately reflect the overall health status of the server, the present invention adopts a weighted aggregation method to integrate the health score features from different dimensions. The health score features of each dimension are the output results generated by the GAT-LP network model after processing the input multi-dimensional operation status data (such as CPU occupancy rate, memory usage, disk read and write frequency, etc.). Each dimension (such as CPU, memory, disk, etc.) corresponds to one or more health score features, which are used to quantify the health status of that dimension. The weight of each dimension can be set according to its importance to system stability to ensure that the finally generated historical health index data is more scientific and reasonable. This process not only comprehensively considers multiple factors, avoids the possible misguidance caused by a single indicator, but also effectively reduces noise interference through mathematical weighted aggregation, improving the stability and reliability of the evaluation results. In addition, this method can also highlight the abnormal situations that have a greater impact on the system, helping the operation and maintenance personnel to prioritize the most critical problems.

[0171] Use a Python library to generate a bar chart from the weighted aggregated server and process health report data.

[0172] Specifically, utilize the plotting library in Python to convert the historical health index data after weighted aggregation into an easily understandable bar chart. The form of the bar chart makes the changing trend of the historical data clear at a glance, facilitating users to compare the server performance in different time periods, and then discover potential development patterns or periodic problems. The graphical display not only enhances the readability of the data but also provides an intuitive basis for predicting the future health status. In addition, by regularly updating the bar chart, the operation and maintenance team can continuously monitor the long-term health trend of the server, adjust the maintenance strategy in a timely manner, and ensure the stable operation of the system and service quality.

[0173] S6. Trigger the alarm mechanism when the server health status is lower than the preset threshold.

[0174] Furthermore, set an initial safety threshold M.

[0175] Specifically, collect the historical health index data of the server, determine the average value and the fluctuation range as the baseline, adjust the baseline according to the business requirements, initially set the threshold M, and continuously monitor and dynamically adjust M during actual operation to ensure its effectiveness and applicability.

[0176] Dynamically adjust the initial safety threshold M through time series analysis of the historical health index.

[0177] It should be noted that a numerical value M is predefined as the standard for measuring the server health status. This threshold M is used to determine whether the health index of the server is within the normal range. When the health index of the server is lower than M, the alarm mechanism will be triggered to prompt the operation and maintenance personnel that there may be potential risks and preventive or corrective measures need to be taken. This step dynamically adjusts the safety threshold M by analyzing the time series of the server's past health index to adapt to the changes in the server's workload in different time periods. This method of setting the adaptive threshold M not only considers the static historical data but also combines the trend prediction of the time series, ensuring the rationality and timeliness of the threshold. Its role is to avoid false alarms or missed alarms that may be caused by a fixed threshold, improve the sensitivity and accuracy of the alarm mechanism, and ultimately achieve the beneficial effect of optimizing resource utilization and improving system reliability.

[0178] The server calculates the historical health index data every minute and compares it with the adjusted preset threshold M.

[0179] Specifically, historical health index data refers to the health status scoring data of the server within a certain past time period, which is generated by processing historical operation status data. The generation of historical health index data is based on the health scoring features generated by the GAT-LP network model. The GAT-LP network model processes the input multi-dimensional operation status data (such as CPU occupancy rate, memory usage, disk read and write frequency, etc.) to generate health scoring features for each dimension. Based on the health scoring features, the overall health score of the server is generated through methods such as weighted aggregation. The weight of each dimension can be set according to its importance to system stability. The health scores calculated every minute are accumulated in time series to generate historical health index data. The historical health index data calculated every minute can provide near-real-time feedback on the server operation status. By comparing the current health index with the adjusted security threshold, any potential problems can be quickly discovered. Historical health index data is used to dynamically adjust the preset threshold M to adapt to the workload changes of the server in different time periods, ensuring the rationality and timeliness of the threshold. By analyzing historical health index data, the changing trend of the server's health status can be better understood, and potential problems can be discovered in advance. The role of this step is to ensure that the alarm mechanism can react immediately when problems start to show up, thus buying more time for the operation and maintenance personnel to take corrective measures and prevent the problems from deteriorating further. Ultimately, this method greatly shortens the fault response time and improves the stability and reliability of the system.

[0180] If the health index is lower than the preset threshold M, start the alarm process and generate a description document of the fault cause and diagnostic suggestions.

[0181] Specifically, when the health index is lower than the preset threshold M, immediately start the alarm process and automatically generate a detailed document of the fault cause and diagnostic suggestions. The key to this step is to provide immediate and specific guiding information to help technical support personnel quickly locate the root cause of the problem and reduce the troubleshooting time. Its purpose is to provide clear operation guidelines for the maintenance team to ensure that they can solve problems efficiently and reduce the delays caused by human judgment errors. Ultimately, this process significantly improves the fault handling efficiency and enhances the self-repair ability of the system.

[0182] According to the degree of deterioration of the health index, trigger email, SMS, and phone warnings in sequence to notify technical support personnel.

[0183] Specifically, according to the severity of the deterioration of the health index, the system will automatically select appropriate warning methods, such as emails, text messages or phone calls, and gradually upgrade the notification level. This method ensures that warning messages can be promptly conveyed to relevant personnel. Especially in emergency situations, a phone warning can immediately draw attention. Its function lies in establishing a multi-level warning mechanism to ensure that warning messages will not be ignored. At the same time, it can also flexibly adjust the notification intensity according to different situations, ensuring the effectiveness of information transmission while avoiding unnecessary interference. Ultimately, this method effectively promotes rapid response and reduces the scope of the impact of faults.

[0184] Restart the server when the health index reaches the critical point.

[0185] Specifically, when the health index drops to the critical point, the system automatically executes the server restart operation. This measure aims to prevent the server from entering an uncontrollable state and restore normal services through proactive intervention. Its purpose is to ensure business continuity in extreme situations and minimize service interruption time. The restart action, as the last line of security, ensures that even in the most adverse circumstances, the server can quickly resume its normal working state, thereby protecting user data and service quality. Ultimately, this method strengthens the fault tolerance and self-healing capabilities of the system and improves the overall service level.

[0186] This embodiment also provides a computer device applicable to the situation of the server health status diagnosis method based on the GAT-LP algorithm, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the server health status diagnosis method based on the GAT-LP algorithm as proposed in the above embodiment.

[0187] This computer device can be a terminal. The computer device includes a processor, a memory, a communication interface, a display screen, and an input device connected through a system bus. Among them, the processor of this computer device is used to provide computing and control capabilities. The memory of this computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of this computer device is used to communicate with external terminals in a wired or wireless manner. The wireless manner can be achieved through WIFI, a carrier network, NFC (Near Field Communication), or other technologies. The display screen of this computer device can be a liquid crystal display screen or an electronic ink display screen. The input device of this computer device can be a touch layer covering the display screen, or a button, a trackball, or a touchpad set on the outer shell of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0188] This embodiment also provides a storage medium, on which a computer program is stored. When the program is executed by a processor, it implements the server health status diagnosis method based on the GAT-LP algorithm proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, a magnetic disk, or an optical disc.

[0189] In summary, by setting the acquisition time interval, the present invention periodically collects multi-dimensional operation status data of the server and compresses and stores it, effectively reducing the required transmission bandwidth. After the compressed data is transmitted to the data analysis platform, through decompression, outlier correction, missing value completion, and standardization processing, the data quality and consistency are improved. The preprocessed data is converted into a graph structure and input into the analysis process by defining node feature vectors and adjacency matrices. Combining the graph attention mechanism with time series analysis techniques, the features of multi-source time series data are deeply mined. The parrot optimization algorithm is used to automatically tune the parameters, avoid local optima, and improve the robustness of the analysis results. Further, by constructing a multi-dimensional health assessment index system, the operation status of the server is intuitively displayed, potential risks are discovered in advance, and preventive measures are formulated. When the health status is lower than the threshold, the alarm mechanism is automatically triggered to provide fault diagnosis suggestions. When the health index reaches the critical point, the server is automatically restarted, thereby ensuring service continuity and stable operation of the business.

[0190] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. A server health status diagnosis method based on the GAT-LP algorithm, characterized in that: include, Collect multi-dimensional operation status data of the server; Transmit multi-dimensional operating status data to the data analysis platform and perform pre-processing; The preprocessed server historical operation status data is input into the GAT-LP network for training to obtain the GAT-LP network model; According to the server health status diagnosis results, the health score features extracted based on the GAT-LP network model are used to evaluate the health status of the server and process and generate the overall health status; Convert the assessment results into visual health status; When the server health status is lower than the preset threshold, the alarm mechanism is triggered. After a failure occurs, the server operation status changes before the failure are traced back to generate a failure timeline. Based on the tracing results, diagnostic suggestions and repair plans are generated.

2. The server health status diagnosis method based on the GAT-LP algorithm according to claim 1, characterized in that: The specific steps of collecting multi-dimensional operation status data of the server are as follows: Set the collection time interval to regularly collect the server's total CPU usage, total memory usage, main process CPU usage, memory usage, network traffic, and disk read and write status; Perform preliminary processing on the collected multi-dimensional operating status data at the edge computing node; The data after preliminary processing is compressed by a data compression algorithm and a compressed package is generated; Store the compressed package in the server's local database.

3. The server health status diagnosis method based on the GAT-LP algorithm as claimed in claim 2, characterized in that: The multi-dimensional operating status data is transmitted to the data analysis platform and pre-processed. The specific steps are: Transfer the compressed package of the server local database to the data analysis platform; After receiving the compressed package, the data analysis platform uses the data compression algorithm to decompress the compressed package; After decompressing the compressed package, the data is corrected for abnormal values ​​through rule detection, and the missing values ​​are supplemented by interpolation method to obtain the historical operation status data of the server; Standardize the historical operation status data of the server.

4. The server health status diagnosis method based on the GAT-LP algorithm as claimed in claim 3, characterized in that: The pre-processed server historical operation status data is input into the GAT-LP network for training to obtain the GAT-LP network model. The specific steps are: Convert the standardized historical server operation status data into a graph structure and define node feature vectors and adjacency matrices; Input the node feature matrix vector and adjacency matrix into the GAT network layer, calculate the correlation between nodes and generate the feature matrix; The feature matrix output by the GAT layer is input into the LSTM network layer to capture the time series features; The time series features of the LSTM layer are processed by using a fully connected layer for dimensionality reduction; The Parrot optimization algorithm is used to find the optimal hyperparameter combination and obtain the GAT-LP network model.

5. The server health status diagnosis method based on the GAT-LP algorithm as claimed in claim 4, characterized in that: The Parrot optimization algorithm is used to find the optimal hyperparameter combination. The specific steps are: Define N to represent the number of individuals that search for the optimal hyperparameter configuration in the Parrot optimization algorithm; Initialize the hyperparameter search space and randomly generate N parrot individuals, each parrot represents a set of hyperparameter configurations; Evaluate and train the hyperparameters corresponding to each parrot, record the fitness value of the validation set, and select the parrot position corresponding to the maximum fitness value of all parrots based on the fitness value of the validation set; Update the parrot position according to the parrot position corresponding to the maximum fitness of all parrots and the average position of the group, and perform any of the foraging, staying, communication and fear behaviors to update the position; Check whether the parrot's updated position at the next moment exceeds the boundary value. If it exceeds, set it as the boundary value and recalculate the fitness value; The iterative optimization process is repeated to output the optimal hyperparameter combination, and the GAT-LP network is trained using the optimal hyperparameters.

6. The server health status diagnosis method based on the GAT-LP algorithm according to claim 5, characterized in that: According to the server health status diagnosis results, based on the health score features extracted by the GAT-LP network model, the health status of the server and the process is evaluated and the overall health status is generated. The specific steps are: Input the server's historical operating status data and real-time operating status data into the GAT-LP network model to calculate and extract health score features; Based on the health score features extracted by the GAT-LP network model, the overall health status of servers and processes is evaluated; Generates server and process health reports based on the assessment of the overall health of servers and processes.

7. The server health status diagnosis method based on the GAT-LP algorithm as claimed in claim 6, characterized in that: The specific steps of converting the evaluation results into a visualized health status are as follows: Use the Web development framework and front-end library to build a visualization interface framework, select visualization tools, and set up a health status overview area, detailed indicator display area, and alarm prompt area in the visualization interface framework; Extract health score features from the GAT-LP network model and map them to a color-coded table, and convert the color-coded table into a visual server and process health status report; Consolidate and perform weighted aggregation of visual server and process health report data; Use Python library to generate histograms from weighted aggregated server and process health report data.

8. The server health status diagnosis method based on the GAT-LP algorithm according to claim 7, characterized in that: The alarm mechanism is triggered when the server health status is lower than the preset threshold. After the failure occurs, the server operation status changes before the failure are traced back to generate a failure timeline, and diagnostic suggestions and repair plans are generated based on the tracing results. The specific steps are as follows: Set the initial safety threshold M; Through the time series analysis of historical health index, the initial safety threshold M is dynamically adjusted; The server calculates the historical health index data once a minute and compares it with the adjusted preset threshold M; If the health index is lower than the preset threshold M, the alarm process is initiated and a description document of the fault cause and diagnostic suggestions is generated; According to the degree of deterioration of the health index, email, SMS and phone warnings are triggered in sequence to notify technical support personnel; Restart the server when the health index reaches a critical point; After a failure occurs, trace back the server operation status changes before the failure and generate a failure timeline; Generate diagnostic suggestions and repair plans based on the traceability results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the server health status diagnosis method based on the GAT-LP algorithm described in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the server health status diagnosis method based on the GAT-LP algorithm described in any one of claims 1 to 8 are implemented.

Images