Data center IT equipment fault diagnosis method and device, equipment and storage medium

Abstract

The invention provides a data center IT equipment fault diagnosis method, apparatus and device, and a storage medium. The method comprises the steps of obtaining a multi-dimensional data set of each IT equipment of a data center; performing matching association on the multi-dimensional data set to obtain total information of each fault work order, and performing vectorization processing on the total information of each fault work order to obtain a total feature vector of each fault work order; according to the full feature vector of each fault work order, constructing a graph structure corresponding to each fault work order, and determining a node feature vector matrix and an adjacent matrix of the graph structure; inputting the node feature vector matrix and the adjacent matrix of the graph structure into a pre-constructed graph attention neural network model to obtain a fault root cause corresponding to the graph structure output by the graph attention neural network model; wherein the graph attention neural network model is obtained by training based on a sample node feature vector matrix and a sample adjacency matrix of the sample graph structure and a fault root cause label corresponding to the sample graph structure.

Description

Data center IT equipment fault diagnosis methods, devices, equipment and storage media Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a method, apparatus, device, and storage medium for diagnosing faults in data center IT equipment. Background Technology

[0002] Data center IT equipment fault diagnosis is a crucial step in ensuring stable data center operation and improving service availability. Currently, data center IT equipment maintenance primarily relies on manual experience. Maintenance personnel develop troubleshooting plans based on alarm information and past experience. This traditional maintenance method has the following drawbacks: First, the process is time-consuming and cumbersome, resulting in low fault location efficiency; second, considering only a single factor affects the accuracy of fault location, making it difficult to accurately assess the impact of equipment failure on business operations, which may seriously damage service quality and lead to continuous customer complaints. Summary of the Invention

[0003] This invention provides a method, apparatus, device, and storage medium for diagnosing faults in data center IT equipment, which addresses the shortcomings of existing technologies where fault diagnosis of data center IT equipment relies on human experience, considers only a single factor, and has low efficiency and accuracy in fault location.

[0004] In a first aspect, the present invention provides a method for diagnosing faults in IT equipment in a data center, comprising: acquiring a multidimensional dataset of each IT device in the data center; matching and associating the multidimensional dataset to obtain full information of each fault work order; vectorizing the full information of each fault work order to obtain a full feature vector of each fault work order; constructing a graph structure corresponding to each fault work order based on the full feature vector of each fault work order, and determining the node feature vector matrix and adjacency matrix of the graph structure; inputting the node feature vector matrix and adjacency matrix of the graph structure into a pre-constructed graph attention neural network model to obtain the root cause of the fault corresponding to the graph structure output by the graph attention neural network model; wherein, the graph attention neural network model is trained based on the sample node feature vector matrix and sample adjacency matrix of the sample graph structure, and the root cause label of the fault corresponding to the sample graph structure.

[0005] In some embodiments, the step of constructing a graph structure corresponding to each fault work order based on the full feature vector of each fault work order, and determining the node feature vector matrix and adjacency matrix corresponding to the graph structure, includes: determining multiple target IT devices involved in each fault work order based on the full feature vector of each fault work order; constructing a graph structure corresponding to each fault work order using the multiple target IT devices as nodes; determining the feature vector of each node based on the full feature vector of each fault work order to obtain the node feature vector matrix corresponding to the graph structure; determining whether the feature vectors of any two nodes in each node exist within the target time period; if so, adding an edge to the two nodes; traversing each node to determine all edges to obtain the adjacency matrix corresponding to the graph structure.

[0006] In some embodiments, the graph attention neural network model includes a graph attention layer based on a multi-head attention mechanism, a feature reduction layer, and a classification layer. Correspondingly, the step of inputting the node feature vector matrix and adjacency matrix of the graph structure into the pre-constructed graph attention neural network model to obtain the root cause of the fault corresponding to the graph structure output by the graph attention neural network model includes: inputting the node feature vector matrix and adjacency matrix of the graph structure into the graph attention layer to obtain the high-dimensional feature information of the graph structure output by the graph attention layer; inputting the high-dimensional feature information of the graph structure into the feature reduction layer to obtain the low-dimensional feature information of the graph structure output by the feature reduction layer; and inputting the low-dimensional feature information of the graph structure into the classification layer to obtain the root cause of the fault corresponding to the graph structure output by the classification layer.

[0007] In some embodiments, the step of inputting the node feature vector matrix and adjacency matrix of the graph structure into the graph attention layer to obtain the high-dimensional feature information of the graph structure output by the graph attention layer includes: determining the weight coefficients of each node and connected nodes of the graph structure based on the graph attention layer, calculating the attention mechanism features of each node, and obtaining the high-dimensional feature information of the graph structure based on the attention mechanism features of each node.

[0008] In some embodiments, the multidimensional dataset includes an alarm dataset, a fault ticket dataset, a performance dataset, a complaint dataset, a resource dataset, a business dataset, and a topology dataset.

[0009] In some embodiments, matching and associating the multidimensional datasets to obtain full information of each fault work order includes: matching and associating the multidimensional datasets based on the fault work order datasets; matching and associating the multidimensional datasets according to the location information of each IT device; and matching and associating the multidimensional datasets according to the corresponding time information.

[0010] In some embodiments, the full information of each fault work order is vectorized to obtain the full feature vector of each fault work order, including: extracting features from the full information of each fault work order to obtain the text features and non-text features of each fault work order; inputting the text features of each fault work order into a first vectorization model to obtain the first word vector of each fault work order output by the first vectorization model; inputting the text features of each fault work order into a second vectorization model to obtain the second word vector of each fault work order output by the second vectorization model; performing weighted aggregation on the first word vector and the second word vector of each fault work order to obtain the word vector of each fault work order; extracting features from the non-text features of each fault work order to obtain the non-text feature vector of each fault work order; and fusing the word vector and non-text feature vector of each fault work order to obtain the full feature vector of each fault work order.

[0011] In some embodiments, the training process of the graph attention neural network model includes: acquiring a multidimensional dataset of sample IT devices in a sample data center; matching and associating the multidimensional dataset to obtain the full sample information of sample fault work orders; vectorizing the full sample information of the sample fault work orders to obtain the full sample feature vector of the sample fault work orders; constructing a sample graph structure corresponding to the sample fault work orders based on the full sample feature vector of the sample fault work orders; determining the sample node feature vector matrix and sample adjacency matrix of the sample graph structure; determining the fault root cause label corresponding to the sample graph structure; training the initial graph attention neural network model using the sample node feature vector matrix and sample adjacency matrix of the sample graph structure as training samples and the fault root cause label corresponding to the sample graph structure as sample labels; and obtaining the graph attention neural network model after training.

[0012] Secondly, the present invention also provides a data center IT equipment fault diagnosis device, comprising: an acquisition unit for acquiring multidimensional datasets of various IT equipment in a data center; a processing unit for matching and associating the multidimensional datasets to obtain full information of each fault work order, and performing vectorization processing on the full information of each fault work order to obtain full feature vectors of each fault work order; a construction unit for constructing a graph structure corresponding to each fault work order based on the full feature vectors of each fault work order, and determining the node feature vector matrix and adjacency matrix of the graph structure; and a fault location unit for inputting the node feature vector matrix and adjacency matrix of the graph structure into a pre-constructed graph attention neural network model to obtain the fault root cause corresponding to the graph structure output by the graph attention neural network model; wherein the graph attention neural network model is trained based on the sample node feature vector matrix and sample adjacency matrix of the sample graph structure, and the fault root cause label corresponding to the sample graph structure.

[0013] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the data center IT equipment fault diagnosis method as described above.

[0014] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the data center IT equipment fault diagnosis method as described above.

[0015] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the data center IT equipment fault diagnosis method as described above.

[0016] The present invention provides a data center IT equipment fault diagnosis method, apparatus, device, and storage medium that acquires multidimensional datasets of each IT device in the data center, matches and associates these datasets to obtain full information of each fault work order, vectorizes them to obtain full feature vectors of each fault work order, constructs a graph structure corresponding to each fault work order based on the full feature vectors of each fault work order, determines the node feature vector matrix and adjacency matrix of the graph structure, and inputs them into a pre-constructed graph attention neural network model to obtain the root cause of the fault corresponding to the graph structure. This method and apparatus have a high level of automation and intelligence, improving the efficiency and accuracy of IT equipment fault root cause localization. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0018] Figure 1 is a flowchart illustrating a data center IT equipment fault diagnosis method provided in an embodiment of the present invention; Figure 2 is a flowchart illustrating vectorized processing of full information of each fault work order provided in an embodiment of the present invention; Figure 3 is a flowchart illustrating the construction of a graph structure corresponding to each fault work order provided in an embodiment of the present invention; Figure 4 is a schematic diagram of a graph structure corresponding to a fault work order provided in an embodiment of the present invention; Figure 5 is a schematic diagram of a graph structure set corresponding to multiple fault work orders provided in an embodiment of the present invention; Figure 6 is a schematic diagram of a graph attention neural network model provided in an embodiment of the present invention; Figure 7 is a schematic diagram of a data center IT equipment fault diagnosis device provided in an embodiment of the present invention; Figure 8 is a schematic diagram of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0020] The terms "first," "second," etc., used in this invention are used to distinguish similar objects and not to describe a specific order or sequence. It should be understood that such terms can be used interchangeably where appropriate so that embodiments of the invention can be implemented in orders other than those illustrated or described herein, and the objects distinguished by "first" and "second" are generally of the same class, without limiting the number of objects; for example, the first object can be one or more.

[0021] Figure 1 is a flowchart illustrating a data center IT equipment fault diagnosis method provided in an embodiment of the present invention. As shown in Figure 1, a data center IT equipment fault diagnosis method is provided, including the following steps: step 110, step 120, step 130, and step 140. These method steps are merely one possible implementation of the present invention.

[0022] Step 110: Obtain the multidimensional dataset of each IT device in the data center.

[0023] A data center is a facility that centrally manages, stores, and processes data. It is typically used to support an enterprise's IT infrastructure and business operations, ensuring data security and availability.

[0024] Optionally, each IT device may include at least a server, storage device, and network device.

[0025] In some embodiments, the multidimensional dataset includes an alarm dataset, a fault ticket dataset, a performance dataset, a complaint dataset, a resource dataset, a business dataset, and a topology dataset.

[0026] The alarm dataset includes fields such as alarm time, device location, and related business; the fault ticket dataset includes fields such as fault ticket occurrence time, device location, and handling measures; the performance dataset includes fields such as device location, collection time, and various performance indicators (e.g., CPU, memory, IO); the complaint dataset includes fields such as device location, related business type, and complaint content; the resource dataset includes fields such as device location and resource information for each device; the business dataset includes fields such as device location and business information; and the topology dataset includes fields such as location topology relationships between various IT devices.

[0027] Step 120: Match and associate the multidimensional dataset to obtain the full information of each fault work order. Then, vectorize the full information of each fault work order to obtain the full feature vector of each fault work order.

[0028] Optionally, data cleaning can be performed on the cube.

[0029] Optionally, the associated fields of the cube can be determined, such as fault ticket ID and device ID.

[0030] Optionally, the multidimensional dataset can be matched and associated based on the associated fields to obtain full information of each fault work order.

[0031] Understandably, by matching and associating multidimensional datasets to obtain full information on each fault work order, a more comprehensive view of fault work orders can be provided, thus providing a data foundation for accurate location of equipment faults.

[0032] In some embodiments, the multidimensional datasets are matched and associated to obtain the full information of each fault work order, including: matching and associating the multidimensional datasets based on the fault work order datasets; matching and associating the multidimensional datasets based on the location information of each IT device; and matching and associating the multidimensional datasets based on the corresponding time information.

[0033] Optionally, for different device types, multi-dimensional data association is performed based on the fault work order dataset. Each dataset can be matched and associated using corresponding location information (such as Internet Protocol IP, device name, and device type). The alarm, performance, and complaint data are associated by selecting the appropriate time granularity (such as selecting alarm and performance data 30 minutes before the fault occurred, and complaint data within 48 hours). If there is no relevant type of data for a device, a null value is selected, thereby obtaining a full information table of fault work orders for each device. The full information table includes information such as alarms, performance, business, and complaints.

[0034] Figure 2 is a schematic diagram of the process for vectorizing the full information of each fault work order according to an embodiment of the present invention. As shown in Figure 2, in some embodiments, the full information of each fault work order is vectorized to obtain the full feature vector of each fault work order, including: extracting features from the full information of each fault work order to obtain the text features and non-text features of each fault work order; inputting the text features of each fault work order into a first vectorization model to obtain the first word vector of each fault work order output by the first vectorization model; inputting the text features of each fault work order into a second vectorization model to obtain the second word vector of each fault work order output by the second vectorization model; weighting and aggregating the first word vector and the second word vector of each fault work order to obtain the word vector of each fault work order; extracting features from the non-text features of each fault work order to obtain the non-text feature vector of each fault work order; and fusing the word vector and non-text feature vector of each fault work order to obtain the full feature vector of each fault work order.

[0035] Optionally, both the first vectorization model and the second vectorization model are Word2Vec models used to generate word embeddings; the first vectorization model is a Continuous Bag of Words (CBOW) model, and the second vectorization model is a Skip-gram model.

[0036] The CBOW model predicts the center word using context words, while the Skip-gram model does the opposite, using the center word to predict its context words.

[0037] Optionally, the first vectorization model is trained based on the sample text features of the sample fault work order and the first word vector label of the sample fault work order.

[0038] Optionally, the second vectorization model is trained based on the sample text features of the sample fault work order and the second word vector label of the sample fault work order.

[0039] Optionally, feature extraction is performed on the full sample information of the sample fault work order to obtain the sample text features of the sample fault work order. The full sample information includes information such as alarms, performance, business, complaints, and resources.

[0040] Optionally, the full information of each fault work order is sorted according to the time of occurrence of each fault work order, and the full information of each fault work order is vectorized in turn to obtain a coded corpus of the full information of each fault work order.

[0041] Step 130: Based on the full feature vector of each fault work order, construct the graph structure corresponding to each fault work order, and determine the node feature vector matrix and adjacency matrix of the graph structure.

[0042] It is understandable that the states of various IT devices in a data center are highly coupled. A failure in one device can affect other devices to varying degrees. By constructing the data of each device into a graph data structure, it is helpful to use graph neural networks to mine the correlation features between devices and thus find the root cause of the failure.

[0043] Figure 3 is a schematic diagram of the process for constructing the graph structure corresponding to each fault work order according to an embodiment of the present invention. As shown in Figure 3, based on the full feature vector of each fault work order, multiple target IT devices involved in each fault work order are determined. Using the multiple target IT devices as nodes, a graph structure corresponding to each fault work order is constructed to obtain the feature vectors of each node, such as server features, switch features, router features, load balancing features, input / output (IO) features, storage features, and memory features. Based on whether the feature vectors of each node exist, the edges of the graph structure are determined, and the number of graph structures is determined based on the number of fault work orders.

[0044] In some embodiments, based on the full feature vector of each fault work order, a graph structure corresponding to each fault work order is constructed, and the node feature vector matrix and adjacency matrix corresponding to the graph structure are determined. This includes: determining multiple target IT devices involved in each fault work order based on the full feature vector of each fault work order; constructing a graph structure corresponding to each fault work order using the multiple target IT devices as nodes; determining the feature vector of each node based on the full feature vector of each fault work order to obtain the node feature vector matrix corresponding to the graph structure; determining whether the feature vectors of any two nodes in each node exist within the target time period; if so, adding an edge to the two nodes; traversing each node to determine all edges to obtain the adjacency matrix corresponding to the graph structure.

[0045] Optionally, multiple target IT devices include servers, switches, routers, load balancers, storage, memory, and I / O devices.

[0046] Optionally, each node has a multi-dimensional feature vector. Each node's feature vector contains only feature data for its own dimension. For example, the feature vector of a memory node only contains relevant data such as memory alarms, memory utilization, total memory, and memory usage, while other dimensions are left empty.

[0047] Figure 4 is a schematic diagram of the graph structure corresponding to the fault work order provided in the embodiment of the present invention. As shown in Figure 4, a graph structure corresponding to the fault work order is provided. The graph has 7 nodes (i.e., server, switch, router, load balancer, storage, memory, and IO device). Each node has its neighbor nodes (i.e., there are connections between nodes). Vi represents the i-th node, and Xi represents the feature information of the i-th node.

[0048] Figure 5 is a schematic diagram of the graph structure set corresponding to multiple fault work orders provided in an embodiment of the present invention. As shown in Figure 5, the graph structures corresponding to different fault work orders are different.

[0049] Optionally, the relationships between nodes in a graph structure can be represented using an adjacency matrix, where each element is a Boolean value or a weight. For a simple graph with n nodes, there is an adjacency matrix: ;in, This represents the edge between nodes V1 and V2.

[0050] Table 1 is the adjacency matrix table corresponding to the graph structure provided in the embodiments of the present invention. As shown in Table 1, the adjacency matrix table of the above graph structure is provided, where V0 represents the server, which is connected to all other nodes, and V5 and V6 are only connected to V0.

[0051] Table 1. Adjacency matrix table corresponding to the graph structure

[0052] Table 2 is the adjacency matrix table corresponding to the graph structure provided in the embodiments of the present invention. As shown in Table 2, each node has F-dimensional features, where "0" indicates that a feature in a certain dimension does not exist.

[0054] Table 2. Node Feature Vector Matrix Corresponding to the Graph Structure

[0055] Step 140: Input the node feature vector matrix and adjacency matrix of the graph structure into the pre-constructed graph attention neural network model to obtain the root cause of the fault corresponding to the graph structure output by the graph attention neural network model.

[0056] The graph attention neural network model is trained based on the sample node feature vector matrix and sample adjacency matrix of the sample graph structure, as well as the root cause labels of the faults corresponding to the sample graph structure.

[0057] Among them, the Graph Attention Network (GAT) model includes multiple graph attention layers based on attention mechanisms. The GAT model uses attention mechanisms to assign different importance weights to the neighbors of each node, enabling the model to focus on more relevant neighbor nodes, thereby improving the effectiveness of information aggregation. The GAT model can handle irregular graph structures without requiring a fixed adjacency matrix, enhancing its flexibility.

[0058] In this embodiment of the invention, by acquiring the multidimensional datasets of each IT device in the data center, matching and associating the multidimensional datasets, the full information of each fault work order is obtained. The data is then vectorized to obtain the full feature vectors of each fault work order. Based on the full feature vectors of each fault work order, a graph structure corresponding to each fault work order is constructed. The node feature vector matrix and adjacency matrix of the graph structure are determined and input into a pre-constructed graph attention neural network model to obtain the root cause of the fault corresponding to the graph structure. This method has a high level of automation and intelligence, improving the efficiency and accuracy of IT device fault root cause localization.

[0059] Figure 6 is a schematic diagram of the graph attention neural network model provided in an embodiment of the present invention. As shown in Figure 6, in some embodiments, the graph attention neural network model includes a graph attention layer based on a multi-head attention mechanism, a feature dimensionality reduction layer, and a classification layer.

[0060] Correspondingly, the node feature vector matrix and adjacency matrix of the graph structure are input into a pre-constructed graph attention neural network model to obtain the root cause of the fault corresponding to the graph structure output by the graph attention neural network model. This includes: inputting the node feature vector matrix and adjacency matrix of the graph structure into the graph attention layer to obtain the high-dimensional feature information of the graph structure output by the graph attention layer; inputting the high-dimensional feature information of the graph structure into the feature reduction layer to obtain the low-dimensional feature information of the graph structure output by the feature reduction layer; and inputting the low-dimensional feature information of the graph structure into the classification layer to obtain the root cause of the fault corresponding to the graph structure output by the classification layer.

[0061] Optionally, the graph attention neural network model includes multiple graph attention layers. Based on these multiple graph attention layers, feature extraction is performed on the node feature vector matrix and adjacency matrix of the graph structure to obtain the high-dimensional feature information of the graph structure output by the graph attention layers.

[0062] Optionally, the classification layer includes the XGBoost classification model. The XGBoost classification model improves classification efficiency through techniques such as parallelization and caching optimization. Through L1 and L2 regularization, it can prevent overfitting and improve generalization ability. It can automatically handle missing data, support multiple loss functions, has strong adaptability, and can provide feature importance scores, which helps to analyze the impact of features on model prediction.

[0063] In some embodiments, the node feature vector matrix and adjacency matrix of the graph structure are input to the graph attention layer to obtain the high-dimensional feature information of the graph structure output by the graph attention layer, including: determining the weight coefficients of each node and its connected nodes based on the graph attention layer, calculating the attention mechanism features of each node, and obtaining the high-dimensional feature information of the graph structure based on the attention mechanism features of each node.

[0064] Optionally, to calculate the deep features of a server node, first calculate the attention coefficient from the node to its connected nodes, then calculate the attention coefficient between the server node and the switch node. It can be represented as: Where W is the linear transformation weight matrix applied to each node, h represents the feature vector of each node, and a is a function for calculating the correlation between two nodes.

[0065] Optionally, the weighting coefficients of server and switch nodes The calculation formula can be expressed as: Where, softmax represents normalization, making the sum of the weight coefficients from the server node to other connected nodes equal to 1, N 服务器 This represents the total number of nodes connected to the server node.

[0066] Alternatively, the LeakyReLU function can be used for calculation, and the formula for calculating the weight coefficients can be expressed as: It should be noted that by calculating the weight coefficients from the server to other nodes (routers, load balancers, I / O devices, storage, memory) using the above method, the attention mechanism characteristics of the server nodes can be obtained, which can be represented as follows: Optionally, the attention mechanism features of all nodes are calculated to obtain the deep node features of the graph structure.

[0067] Optionally, a multi-head attention mechanism can be introduced to stabilize the learning process. That is, when calculating the node weight coefficient, K groups (K=3) of independent attention mechanisms are called and the average value is taken to strengthen the single attention mechanism.

[0068] Optionally, the weight coefficients can be calculated using a multi-head attention mechanism, which can be expressed as: Where k represents the k-th attention mechanism and j represents the j-th node connected to the server.

[0069] In some embodiments, the training process of the graph attention neural network model includes: acquiring a multidimensional dataset of sample IT devices in a sample data center; matching and associating the multidimensional dataset to obtain the full sample information of sample fault work orders; vectorizing the full sample information of sample fault work orders to obtain the full sample feature vector of sample fault work orders; constructing a sample graph structure corresponding to the sample fault work orders based on the full sample feature vector of the sample fault work orders; determining the sample node feature vector matrix and sample adjacency matrix of the sample graph structure; determining the fault root cause label corresponding to the sample graph structure; training the initial graph attention neural network model using the sample node feature vector matrix and sample adjacency matrix of the sample graph structure as training samples and the fault root cause label corresponding to the sample graph structure as sample labels; and obtaining the graph attention neural network model after training.

[0070] Optionally, the sample multidimensional dataset includes at least the sample alarm dataset, sample fault ticket dataset, sample performance dataset, sample complaint dataset, sample resource dataset, sample business dataset, and sample topology dataset.

[0071] Optionally, the sample multidimensional dataset is matched and associated to obtain the full information of the sample fault work orders, including: matching and associating the sample multidimensional dataset based on the sample fault work order dataset; matching and associating the sample multidimensional dataset based on the sample location information of each IT device; and matching and associating the sample multidimensional dataset based on the corresponding sample time information.

[0072] Optionally, the full sample information of the sample fault work order is vectorized to obtain the full sample feature vector of the sample fault work order, including: extracting features from the full sample information of the sample fault work order to obtain the sample text features and sample non-text features of the sample fault work order; inputting the sample text features of the sample fault work order into a first vectorization model to obtain the first sample word vector of the sample fault work order output by the first vectorization model; inputting the sample text features of the sample fault work order into a second vectorization model to obtain the second sample word vector of the sample fault work order output by the second vectorization model; performing weighted aggregation on the first sample word vector and the second sample word vector of the sample fault work order to obtain the sample word vector of the sample fault work order; extracting features from the sample non-text features of the sample fault work order to obtain the sample non-text feature vector of the sample fault work order; and fusing the sample word vector and the sample non-text feature vector of the sample fault work order to obtain the full sample feature vector of the sample fault work order.

[0073] Optionally, based on the full feature vector of the sample fault work order, a sample graph structure corresponding to the sample fault work order is constructed, and the sample node feature vector matrix and sample adjacency matrix of the sample graph structure are determined. This includes: determining multiple sample target IT devices involved in the sample fault work order based on the full feature vector of the sample fault work order; constructing the sample graph structure corresponding to the sample fault work order using multiple sample target IT devices as sample nodes; determining the sample feature vector of each sample node based on the full feature vector of the sample fault work order, thereby obtaining the sample node feature vector matrix corresponding to the sample graph structure; determining whether the sample feature vectors of any two sample nodes in each sample node exist within the sample time period; if so, adding an edge to the two sample nodes; traversing each sample node; determining all edges; and obtaining the sample adjacency matrix corresponding to the sample graph structure.

[0074] The data center IT equipment fault diagnosis device provided in the embodiments of the present invention is described below. The data center IT equipment fault diagnosis device described below and the data center IT equipment fault diagnosis method described above can be referred to in correspondence.

[0075] Figure 7 is a schematic diagram of the data center IT equipment fault diagnosis device provided in an embodiment of the present invention. As shown in Figure 7, the data center IT equipment fault diagnosis device 700 includes: an acquisition unit 710, used to acquire multidimensional datasets of each IT equipment in the data center; a processing unit 720, used to match and associate the multidimensional datasets to obtain the full information of each fault work order, and to perform vectorization processing on the full information of each fault work order to obtain the full feature vector of each fault work order; a construction unit 730, used to construct the graph structure corresponding to each fault work order based on the full feature vector of each fault work order, and to determine the node feature vector matrix and adjacency matrix of the graph structure; and a fault location unit 740, used to input the node feature vector matrix and adjacency matrix of the graph structure into a pre-constructed graph attention neural network model to obtain the fault root cause corresponding to the graph structure output by the graph attention neural network model; wherein, the graph attention neural network model is trained based on the sample node feature vector matrix and sample adjacency matrix of the sample graph structure, and the fault root cause label corresponding to the sample graph structure.

[0076] Optionally, based on the full feature vector of each fault work order, a graph structure corresponding to each fault work order is constructed, and the node feature vector matrix and adjacency matrix corresponding to the graph structure are determined. This includes: determining the multiple target IT devices involved in each fault work order based on the full feature vector of each fault work order, and constructing the graph structure corresponding to each fault work order using the multiple target IT devices as nodes; determining the feature vectors of each node based on the full feature vector of each fault work order, and obtaining the node feature vector matrix corresponding to the graph structure; determining whether the feature vectors of any two nodes in each node exist within the target time period, and if so, adding an edge to the two nodes, traversing each node, determining all edges, and obtaining the adjacency matrix corresponding to the graph structure.

[0077] Optionally, the graph attention neural network model includes a graph attention layer based on a multi-head attention mechanism, a feature reduction layer, and a classification layer. Correspondingly, the node feature vector matrix and adjacency matrix of the graph structure are input into the pre-constructed graph attention neural network model to obtain the root cause of the fault corresponding to the graph structure output by the graph attention neural network model. This includes: inputting the node feature vector matrix and adjacency matrix of the graph structure into the graph attention layer to obtain the high-dimensional feature information of the graph structure output by the graph attention layer; inputting the high-dimensional feature information of the graph structure into the feature reduction layer to obtain the low-dimensional feature information of the graph structure output by the feature reduction layer; and inputting the low-dimensional feature information of the graph structure into the classification layer to obtain the root cause of the fault corresponding to the graph structure output by the classification layer.

[0078] Optionally, the node feature vector matrix and adjacency matrix of the graph structure are input into the graph attention layer to obtain the high-dimensional feature information of the graph structure output by the graph attention layer, including: determining the weight coefficients of each node and connected nodes of the graph structure based on the graph attention layer, calculating the attention mechanism features of each node, and obtaining the high-dimensional feature information of the graph structure based on the attention mechanism features of each node.

[0079] Optionally, the multidimensional dataset includes alarm datasets, fault ticket datasets, performance datasets, complaint datasets, resource datasets, business datasets, and topology datasets.

[0080] Optionally, the multidimensional datasets are matched and associated to obtain the full information of each fault work order, including: matching and associating the multidimensional datasets based on the fault work order datasets; matching and associating the multidimensional datasets based on the location information of each IT device; and matching and associating the multidimensional datasets based on the corresponding time information.

[0081] Optionally, the full information of each fault work order is vectorized to obtain the full feature vector of each fault work order, including: extracting features from the full information of each fault work order to obtain the text features and non-text features of each fault work order; inputting the text features of each fault work order into a first vectorization model to obtain the first word vector of each fault work order output by the first vectorization model; inputting the text features of each fault work order into a second vectorization model to obtain the second word vector of each fault work order output by the second vectorization model; weighting and aggregating the first word vector and the second word vector of each fault work order to obtain the word vector of each fault work order; extracting features from the non-text features of each fault work order to obtain the non-text feature vector of each fault work order; and fusing the word vector and non-text feature vector of each fault work order to obtain the full feature vector of each fault work order.

[0082] Optionally, the training process of the graph attention neural network model includes: acquiring a multidimensional dataset of sample IT devices in a sample data center; matching and associating the multidimensional dataset to obtain the full information of sample fault work orders; vectorizing the full information of sample fault work orders to obtain the full feature vector of sample fault work orders; constructing a sample graph structure corresponding to the sample fault work orders based on the full feature vector of sample fault work orders; determining the sample node feature vector matrix and sample adjacency matrix of the sample graph structure; determining the fault root cause label corresponding to the sample graph structure; training the initial graph attention neural network model using the sample node feature vector matrix and sample adjacency matrix of the sample graph structure as training samples and the fault root cause label corresponding to the sample graph structure as sample labels; and obtaining the graph attention neural network model after training.

[0083] It should be noted that the data center IT equipment fault diagnosis device provided in this embodiment of the invention can implement all the method steps implemented in the above-mentioned data center IT equipment fault diagnosis method embodiment, and can achieve the same technical effect. Here, the parts that are the same as those in the method embodiment and the beneficial effects will not be described in detail.

[0084] Figure 8 illustrates a schematic diagram of the physical structure of an electronic device. As shown in Figure 8, the electronic device may include: a processor 810, a communication interface 820, a memory 830, and a communication bus 840. The processor 810, the communication interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a data center IT equipment fault diagnosis method. This method includes: acquiring a multidimensional dataset of each IT device in the data center; matching and associating the multidimensional dataset to obtain full information of each fault work order; vectorizing the full information of each fault work order to obtain full feature vectors for each fault work order; constructing a graph structure corresponding to each fault work order based on its full feature vectors; determining the node feature vector matrix and adjacency matrix of the graph structure; inputting the node feature vector matrix and adjacency matrix of the graph structure into a pre-constructed graph attention neural network model to obtain the fault root cause corresponding to the graph structure output by the graph attention neural network model; wherein the graph attention neural network model is trained based on the sample node feature vector matrix and sample adjacency matrix of the sample graph structure, and the fault root cause labels corresponding to the sample graph structure.

[0085] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0086] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the data center IT equipment fault diagnosis method provided by the above methods. The method includes: acquiring a multidimensional dataset of each IT equipment in the data center; matching and associating the multidimensional dataset to obtain full information of each fault work order; vectorizing the full information of each fault work order to obtain a full feature vector of each fault work order; constructing a graph structure corresponding to each fault work order based on the full feature vector of each fault work order; determining the node feature vector matrix and adjacency matrix of the graph structure; inputting the node feature vector matrix and adjacency matrix of the graph structure into a pre-constructed graph attention neural network model to obtain the fault root cause corresponding to the graph structure output by the graph attention neural network model; wherein, the graph attention neural network model is trained based on the sample node feature vector matrix and sample adjacency matrix of the sample graph structure, and the fault root cause label corresponding to the sample graph structure.

[0087] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program is implemented to perform the data center IT equipment fault diagnosis method provided by the above methods. The method includes: acquiring a multidimensional dataset of each IT device in the data center; matching and associating the multidimensional dataset to obtain full information of each fault work order; vectorizing the full information of each fault work order to obtain a full feature vector of each fault work order; constructing a graph structure corresponding to each fault work order based on the full feature vector of each fault work order; determining the node feature vector matrix and adjacency matrix of the graph structure; inputting the node feature vector matrix and adjacency matrix of the graph structure into a pre-constructed graph attention neural network model to obtain the fault root cause corresponding to the graph structure output by the graph attention neural network model; wherein, the graph attention neural network model is trained based on the sample node feature vector matrix and sample adjacency matrix of the sample graph structure, and the fault root cause label corresponding to the sample graph structure.

[0088] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0089] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0090] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for diagnosing faults in data center IT equipment, characterized in that, include: Obtain multidimensional datasets of each IT device in the data center; match and associate the multidimensional datasets to obtain full information of each fault work order; and perform vectorization processing on the full information of each fault work order to obtain the full feature vector of each fault work order. Based on the full feature vector of each fault work order, a graph structure corresponding to each fault work order is constructed, and the node feature vector matrix and adjacency matrix of the graph structure are determined. The node feature vector matrix and adjacency matrix of the graph structure are input into a pre-constructed graph attention neural network model to obtain the fault root cause corresponding to the graph structure output by the graph attention neural network model. The graph attention neural network model is trained based on the sample node feature vector matrix and sample adjacency matrix of the sample graph structure, as well as the fault root cause label corresponding to the sample graph structure.

2. The data center IT equipment fault diagnosis method according to claim 1, characterized in that, The step of constructing a graph structure corresponding to each fault work order based on the full feature vector of each fault work order, and determining the node feature vector matrix and adjacency matrix corresponding to the graph structure, includes: determining multiple target IT devices involved in each fault work order based on the full feature vector of each fault work order; constructing a graph structure corresponding to each fault work order using the multiple target IT devices as nodes; determining the feature vector of each node based on the full feature vector of each fault work order to obtain the node feature vector matrix corresponding to the graph structure; determining whether the feature vectors of any two nodes in each node exist within the target time period; if so, adding an edge to the two nodes; traversing each node to determine all edges to obtain the adjacency matrix corresponding to the graph structure.

3. The data center IT equipment fault diagnosis method according to claim 1, characterized in that, The graph attention neural network model includes a graph attention layer based on a multi-head attention mechanism, a feature reduction layer, and a classification layer. Correspondingly, the step of inputting the node feature vector matrix and adjacency matrix of the graph structure into the pre-constructed graph attention neural network model to obtain the root cause of the fault corresponding to the graph structure output by the graph attention neural network model includes: inputting the node feature vector matrix and adjacency matrix of the graph structure into the graph attention layer to obtain the high-dimensional feature information of the graph structure output by the graph attention layer; inputting the high-dimensional feature information of the graph structure into the feature reduction layer to obtain the low-dimensional feature information of the graph structure output by the feature reduction layer; and inputting the low-dimensional feature information of the graph structure into the classification layer to obtain the root cause of the fault corresponding to the graph structure output by the classification layer.

4. The data center IT equipment fault diagnosis method according to claim 1, characterized in that, The step of inputting the node feature vector matrix and adjacency matrix of the graph structure into the graph attention layer to obtain the high-dimensional feature information of the graph structure output by the graph attention layer includes: determining the weight coefficients of each node and its connected nodes in the graph structure based on the graph attention layer, calculating the attention mechanism features of each node, and obtaining the high-dimensional feature information of the graph structure based on the attention mechanism features of each node.

5. The data center IT equipment fault diagnosis method according to claim 1, characterized in that, The multidimensional dataset includes alarm dataset, fault ticket dataset, performance dataset, complaint dataset, resource dataset, business dataset, and topology dataset.

6. The data center IT equipment fault diagnosis method according to claim 5, characterized in that, The multidimensional dataset is matched and associated to obtain the full information of each fault work order, including: matching and associating the multidimensional dataset based on the fault work order dataset; matching and associating the multidimensional dataset according to the location information of each IT device; and matching and associating the multidimensional dataset according to the corresponding time information.

7. The data center IT equipment fault diagnosis method according to claim 1, characterized in that, The full information of each fault work order is vectorized to obtain the full feature vector of each fault work order, including: extracting features from the full information of each fault work order to obtain the text features and non-text features of each fault work order; inputting the text features of each fault work order into a first vectorization model to obtain the first word vector of each fault work order output by the first vectorization model; inputting the text features of each fault work order into a second vectorization model to obtain the second word vector of each fault work order output by the second vectorization model; weighting and aggregating the first word vector and the second word vector of each fault work order to obtain the word vector of each fault work order; extracting features from the non-text features of each fault work order to obtain the non-text feature vector of each fault work order; and fusing the word vector and non-text feature vector of each fault work order to obtain the full feature vector of each fault work order.

8. The data center IT equipment fault diagnosis method according to claim 1, characterized in that, The training process of the graph attention neural network model includes: acquiring a multidimensional dataset of sample IT devices from a sample data center; matching and associating the multidimensional dataset to obtain the full sample information of sample fault work orders; vectorizing the full sample information of the sample fault work orders to obtain the full sample feature vector of the sample fault work orders; constructing a sample graph structure corresponding to the sample fault work orders based on the full sample feature vector of the sample fault work orders; determining the sample node feature vector matrix and sample adjacency matrix of the sample graph structure; determining the fault root cause label corresponding to the sample graph structure; training the initial graph attention neural network model using the sample node feature vector matrix and sample adjacency matrix of the sample graph structure as training samples and the fault root cause label corresponding to the sample graph structure as sample labels; and obtaining the graph attention neural network model after training.

9. A fault diagnosis device for data center IT equipment, characterized in that, include: The acquisition unit is used to acquire multidimensional datasets of various IT devices in the data center; The processing unit is used to match and associate the multidimensional dataset to obtain the full information of each fault work order, and to perform vectorization processing on the full information of each fault work order to obtain the full feature vector of each fault work order; the construction unit is used to construct the graph structure corresponding to each fault work order based on the full feature vector of each fault work order, and to determine the node feature vector matrix and adjacency matrix of the graph structure. The fault localization unit is used to input the node feature vector matrix and adjacency matrix of the graph structure into a pre-constructed graph attention neural network model to obtain the root cause of the fault corresponding to the graph structure output by the graph attention neural network model; wherein, the graph attention neural network model is trained based on the sample node feature vector matrix and sample adjacency matrix of the sample graph structure, as well as the root cause label of the fault corresponding to the sample graph structure.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the data center IT equipment fault diagnosis method as described in any one of claims 1 to 8.

11. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the data center IT equipment fault diagnosis method as described in any one of claims 1 to 8.

12. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the data center IT equipment fault diagnosis method as described in any one of claims 1 to 8.

Images