Fault positioning method and electronic equipment

Abstract

The embodiment of the invention provides a fault positioning method and electronic equipment. The method is applied to the electronic equipment and comprises the following steps: taking an entity in a cloud network as an entity node; according to the layer domain structure of the cloud network, taking different domains as domain nodes; obtaining a first entity node fault probability of the entity node, and obtaining a domain node fault probability of the domain node; using the domain node fault probability to bias the first entity node fault probability to obtain a second entity node fault probability; and positioning the failed entity node according to the fault probability of the second entity node. According to the method provided by the embodiment of the invention, fault domains are classified by taking different entities as entity nodes and taking different domains as domain nodes; the fault probability of the entity node is biased by using the fault probability of the domain node, so that the final domain biased entity fault probability is obtained, and the fault positioning accuracy can be improved.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a fault location method and electronic equipment. Background Art

[0002] Cloud networking is a new type of telecommunications network infrastructure architecture that uses new IT technologies such as virtualization, containers, and microservices to concentrate network resources and functions on the "cloud", thereby providing high-speed, flexible, secure, and low-cost network services.

[0003] With the development of the Internet and the continuous changes and enrichment of different business scenarios such as data centers, the Internet of Things, government and enterprise organizations, mobile users, and artificial intelligence, the traditional network infrastructure can no longer meet the needs of high-speed, flexible, secure, and low-cost network services. Therefore, with the help of new IT technologies such as virtualization, containers, and microservices, by concentrating network resources and functions on the "cloud", clouding the infrastructure of telecommunications networks has become a transformation trend for major operators.

[0004] Entities in cloud networks include network elements, device ports, cloud services, and other different elements. The overall architecture of cloud networks exhibits cross-layer and cross-domain characteristics in both vertical and horizontal directions. Although network cloudification brings many benefits, the cross-layer and cross-domain characteristics of cloud networks also bring new challenges to fault location in telecommunication networks. Therefore, a set of fault location methods for cloud networks is urgently needed to achieve efficient cross-layer and cross-domain fault location. Summary of the invention

[0005] In order to solve the problem of how to realize cross-layer and cross-domain fault location, the present application provides a fault location method and an electronic device, and the present application also provides a computer-readable storage medium.

[0006] The present application embodiment adopts the following technical solutions:

[0007] In a first aspect, the present application provides a fault location method, the method comprising:

[0008] The entities in the cloud network are regarded as physical nodes;

[0009] According to the layer domain structure of the cloud network, different domains are used as domain nodes;

[0010] Obtaining a first entity node failure probability of the entity node, and obtaining a domain node failure probability of the domain node;

[0011] Using the domain node failure probability to bias the first entity node failure probability to obtain a second entity node failure probability;

[0012] The failed entity node is located according to the second entity node failure probability.

[0013] According to the method of the embodiment of the present application, different entities are used as entity nodes and different domains are used as domain nodes to classify fault domains; the fault probability of the entity node is biased by using the fault probability of the domain node, and then the final domain-biased entity fault probability is obtained, which can improve the accuracy of fault location.

[0014] In an implementation manner of the first aspect, the acquiring a first entity node failure probability of the entity node and the acquiring a domain node failure probability of the domain node include:

[0015] Obtaining an entity node vectorized representation of the entity node, and obtaining a domain node vectorized representation of the domain node;

[0016] The entity node vectorized representation and the domain node vectorized representation are respectively used as feature inputs to construct corresponding classifiers to obtain the first entity node failure probability and the domain node failure probability.

[0017] In an implementation manner of the first aspect, the obtaining of the entity node vectorized representation of the entity node and the obtaining of the domain node vectorized representation of the domain node include:

[0018] Based on the entity nodes, construct an entity undirected graph, and extract initial features of the entity nodes;

[0019] Based on the domain nodes, construct a layer domain directed graph;

[0020] Based on the initial features of the entity nodes, performing representation learning on the entity nodes in the entity undirected graph to generate vectorized representations of the entity nodes;

[0021] Based on the vectorized representation of the entity node, representation learning is performed on the domain nodes in the layer domain directed graph to generate the vectorized representation of the domain node.

[0022] According to the above implementation method, the cloud network is jointly represented based on the entity undirected graph and the layer-domain directed graph, which improves the accuracy of the cloud network representation, solves the problems of difficulty in fault location and computational efficiency caused by cross-layer and cross-domain, and can achieve more accurate and faster fault location.

[0023] In an implementation manner of the first aspect, in the layer-domain directed graph, the edge direction between domain nodes in the same layer is bidirectional, and the edge direction between domain nodes in different layers is from the upper layer domain to the lower layer domain.

[0024] According to the above implementation method, drawing on the prior knowledge of "first locating the domain in a large range and then locating the entity in a fine-grained manner" and "top-down" traceability and positioning, a layer-domain directed graph is constructed according to the layer-domain structure of the cloud network, and a hierarchical message passing mechanism is proposed to characterize the domain nodes, which can reduce the probability of false positives and improve the accuracy of fault location.

[0025] In an implementation of the first aspect, performing representation learning on the entity nodes in the entity undirected graph based on the initial features of the entity nodes to generate vectorized representations of the entity nodes includes:

[0026] Representation learning is performed on entity nodes in the entity undirected graph based on a spectral domain graph transformer model.

[0027] In an implementation of the first aspect, the performing representation learning on the entity nodes in the entity undirected graph based on the spectral domain graph transformer model includes:

[0028] Use graph filtering to decompose the input features of each entity node into multiple sub-features of different frequencies;

[0029] Feature transformation is performed through a cross-attention mechanism between the input features of the entity node and the sub-features of different frequencies.

[0030] According to the above implementation method, the graph filter and the transformer are combined, and the graph structure and node features are fused into the transformer. While ensuring the model representation capability, the computational complexity is reduced to linear complexity, thereby improving the computational efficiency.

[0031] In an implementation of the first aspect, based on the vectorized representation of the entity node, performing representation learning on the domain node in the layer domain directed graph to generate the vectorized representation of the domain node includes:

[0032] The entity node vectorized representation is transformed using a multi-layer perceptron, and, according to the correspondence between the domain nodes in the layer-domain directed graph and the entity nodes in the entity undirected graph, the entity node vectorized representation of the entity node corresponding to the domain node is summed and pooled or averaged to obtain an initial vectorized representation of the domain node;

[0033] On the basis of the initial vectorized representation, the domain node representation is updated layer by layer from top to bottom according to the hierarchical structure in the layer-domain directed graph to obtain the domain node vectorized representation.

[0034] In an implementation of the first aspect, the method of taking the entity node vectorized representation and the domain node vectorized representation as feature inputs respectively and constructing corresponding classifiers to obtain the first entity node failure probability and the domain node failure probability includes:

[0035] Taking the vectorized representation of the entity node as feature input, constructing an entity node classifier based on a multi-layer perceptron, and calculating the failure probability of each entity node;

[0036] and / or,

[0037] The vectorized representation of the domain node is used as a feature input, a domain node classifier based on a multi-layer perceptron is constructed, and the failure probability of each domain node is calculated.

[0038] In an implementation of the first aspect, the vectorized representation of the entity node is used as a feature input, an entity node classifier based on a multilayer perceptron is constructed, and the failure probability of each entity node is calculated, wherein:

[0039] In the process of training the entity node classifier, the binary cross entropy between the predicted entity node failure probability and the entity node failure label is used as the loss function, and the parameters to be trained in the entity node classifier are optimized by gradient descent.

[0040] According to the above implementation, if the failure probability of a non-physical node is high, but the failure probability of its domain is very low, in this case, due to the bias of the domain failure probability, the final failure probability of the physical node is still low, thereby reducing the frequency of false positives.

[0041] In an implementation of the first aspect, using the domain node failure probability to bias the first entity node failure probability to obtain a second entity node failure probability includes:

[0042] The first entity node failure probability is multiplied by the domain node failure probability, and the obtained product is the second entity node failure probability.

[0043] In a second aspect, the present application provides an electronic device, comprising a memory for storing computer program instructions and a processor for executing computer program instructions, wherein when the computer program instructions are executed by the processor, the electronic device is triggered to execute the method steps described in the first aspect.

[0044] In a third aspect, the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, which, when executed on a computer, enables the computer to execute the method described in the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 The figure shows an example diagram of the overall layer domain structure of a cloud network according to an embodiment of the present application;

[0046] Figure 2 FIG. 1 is a flowchart of a fault location method based on graph learning according to an embodiment of the present application;

[0047] Figure 3 The figure is a schematic diagram of a fault location process according to an embodiment of the present application;

[0048] Figure 4 Shown is a system architecture diagram according to an embodiment of the present application;

[0049] Figure 5 The figure is a schematic diagram of a fault location process according to an embodiment of the present application;

[0050] Figure 6 Shown is a physical undirected graph according to an embodiment of the present application;

[0051] Figure 7 Shown is a layer domain directed graph according to an embodiment of the present application;

[0052] Figure 8 The figure is a schematic diagram of the SGT model calculation process according to an embodiment of the present application;

[0053] Fig. 9 The figure is a schematic diagram of a domain node representation process according to an embodiment of the present application;

[0054] Fig.10 The figure is a schematic diagram of a failure probability calculation process according to an embodiment of the present application;

[0055] Fig.11 Schematic diagram of the structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in combination with the specific embodiments of the present application and the corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present application.

[0057] The terms used in the implementation section of this application are only used to explain the specific embodiments of this application and are not intended to limit this application.

[0058] After the telecommunications network is cloudified, its overall architecture presents cross-layer and cross-domain characteristics in both vertical and horizontal directions.

[0059] Figure 1 The figure shows an example diagram of the overall layer domain structure of a cloud network according to an embodiment of the present application.

[0060] like Figure 1 As shown, in the vertical direction, the cloud network architecture includes different layers including a network element management layer 110 , a virtualized network element layer 120 , a virtual resource layer 130 , a virtualization layer 140 , and a general hardware layer 150 .

[0061] In each layer, different domains are also involved. For example, the virtual resource layer 130 includes three domains: virtual computing resources 131, virtual storage resources 132, and virtual network resources 133; the network element network management layer 110 includes three domains: element management system (EMS) 1, EMS2, and EMS3; the virtualized network element layer 120 includes three domains: virtual network function unit (VNF) 1, VNF2, and VNF3; the general hardware layer 150 includes three domains: hardware computing resources 151, hardware storage resources 152, and hardware network resources 153.

[0062] The purpose of cross-layer and cross-domain fault location is to locate faults in entities in different domains and layers in the cloud network, that is, to locate the faulty entities that cause cloud network anomalies.

[0063] One solution to achieve fault location is manual location. Based on personal experience, the operation and maintenance engineer first checks the KPI, log, alarm and other data on each device in the cloud network to diagnose the operating status of the device. Then, based on the macro-layer domain structure of the network and the topological relationship between devices at the micro level, according to the business logic of "from upper layer to lower layer, and from top to bottom by layer", the operation and maintenance engineer analyzes each device to determine the root fault device.

[0064] However, in the process of cross-layer and cross-domain fault location, entities from different technical fields are involved, such as Figure 1 The entities in the virtualized network element layer 120 belong to the core network technology field; the entities in the virtual resource layer 130 and the virtualization layer 140 belong to the cloud computing technology field; and the entities in the general hardware layer 150 belong to the IT technology field. This requires a high level of knowledge reserve for operation and maintenance engineers, making it difficult for operation and maintenance engineers in a single field to solve the fault location problem in this case. This increases the manpower cost of operation and maintenance engineers. At the same time, due to the increased complexity of faults caused by cross-layer and cross-domain, the efficiency of manual location is low.

[0065] Another solution to achieve fault location is a fault location solution based on graph learning, which abstracts the fault location problem of telecommunication networks into a problem of graph node representation and classification.

[0066] Figure 2Shown is a flowchart of fault location based on graph learning according to an embodiment of the present application.

[0067] The fault location process based on graph learning mainly includes Figure 2 The following steps are shown:

[0068] S210, telecommunication network graph construction: The telecommunication network is represented as a graph, where nodes represent different entities in the telecommunication network, such as network elements, device ports, and cloud services, and edges represent interactions such as physical topological connections and communication relationships between two entities. The initial attribute feature vector of the node can be constructed from KPI, log, alarm and other data;

[0069] S220, node representation learning: using a graph learning model, such as a graph neural network, a graph attention model (graph Transformer, a graph Transformer is a model that uses the Transformer architecture for graph data representation), etc., to perform node representation learning and obtain a node representation;

[0070] S230, fault node classification: taking the node representation as feature input, using classifiers such as logistic regression and multi-layer perceptron to output the fault probability of each node.

[0071] Although the above solution can reduce the workload of manual operations for fault location, it does not incorporate prior knowledge of cross-layer and cross-domain location logic in the business. In scenarios such as network fault location where the amount of data is insufficient, directly representing and classifying nodes is like "looking for a needle in a haystack", and the location effect is average.

[0072] Furthermore, in the above schemes, most of the existing graph neural network models are based on the aggregation of local neighbor node features, which has the optimization problems of over-smoothing and over-squeezing, resulting in limited representation capabilities; and the existing graph Transformer model learns the graph as a fully connected graph, and calculates the attention scores between all nodes, with a computational complexity of O(N 2 ), where N is the number of nodes. The computational efficiency is low and it is not suitable for learning large-scale graphs. In addition, the N×N optimization space can easily lead to overfitting.

[0073] Furthermore, unified management of cross-layer and cross-domain entities will inevitably make the scale of entities in the network very large. On the one hand, this itself further increases the difficulty of fault location; on the other hand, it also requires related analysis algorithms to have high computational complexity, further increasing the difficulty of implementing fault location.

[0074] In response to the above problems, an embodiment of the present application proposes a fault location method that incorporates prior knowledge of the cloud network layer domain structure and traceability location logic, and can effectively address the problem of increased fault complexity brought about by cross-layer and cross-domain network cloudification.

[0075] Figure 3 FIG. 4 is a schematic diagram of a fault location process according to an embodiment of the present application.

[0076] Electronic equipment execution Figure 3 The following process is shown to achieve fault location.

[0077] S301, taking an entity in a cloud network as an entity node, calculating a failure probability of the entity node (a first entity node failure probability).

[0078] S302 , according to the layer domain structure of the cloud network, different domains are used as domain nodes, and the failure probability of the domain nodes (domain node failure probability) is calculated.

[0079] S303: Use the failure probability of the domain node to bias the failure probability of the entity node to obtain a domain-biased entity node failure probability (a second entity node failure probability).

[0080] Specifically, in S303, for each entity node, the failure probability of the domain node corresponding to the domain in which it is located is used to bias the failure probability of its entity node (first entity node failure probability), and the domain-biased entity node failure probability (second entity node failure probability) is output.

[0081] S304: locate the failed physical node according to the physical node failure probability of the domain bias.

[0082] Specifically, in one embodiment, an entity node whose failure probability exceeds a first preset threshold (e.g., 85%) is a failed entity node, and a domain node whose failure probability exceeds a second preset threshold (the second threshold may be the same as the first threshold) is a failed domain node. The failure may occur in multiple entity nodes and / or multiple domain nodes. When the failure probability of all entity nodes is lower than the first preset threshold (and / or the failure probability of all domain nodes is lower than the second preset threshold), it is deemed that no failure has occurred.

[0083] In another embodiment, the entity node with the highest failure probability among the entity nodes whose failure probability exceeds a third preset threshold (for example, 85%) is the entity node where the failure occurs, and the domain node with the highest failure probability among the domain nodes whose failure probability exceeds a fourth preset threshold (the fourth threshold may be the same as the third threshold) is the domain node where the failure occurs.

[0084] According to the method of the embodiment of the present application, different entities are used as entity nodes and different domains are used as domain nodes to classify fault domains; the fault probability of the entity node is biased by using the fault probability of the domain node, and then the final domain-biased entity fault probability is obtained, which can improve the accuracy of fault location.

[0085] Furthermore, in one embodiment, an entity undirected graph and a layer domain directed graph are constructed for joint representation to generate vectorized representations of entity nodes and domain nodes. The entity node vectorized representation and the domain node representation are used as feature inputs respectively, and corresponding classifiers are constructed to obtain the first entity node failure probability and the domain node failure probability.

[0086] Figure 4 FIG. 4 is a system architecture diagram according to an embodiment of the present application.

[0087] In one embodiment, a system for implementing the fault location method of the present application includes: Figure 4 The graph construction module 310, graph representation module 320 and fault classification module 330 are shown. The graph construction module 310 includes an entity undirected graph construction unit 311 and a layer domain directed graph construction unit 312; the graph representation module 320 includes an entity node vectorization representation unit 321 and a domain node representation unit 322.

[0088] Figure 5 FIG. 4 is a schematic diagram of a fault location process according to an embodiment of the present application.

[0089] Figure 4 The system shown executes Figure 5 The following process is shown to achieve fault location.

[0090] S400, acquiring characteristic data of the cloud network, including different entities in the cloud network, interaction relationships between entities, data generated by the entities, and a layer domain structure of the cloud network.

[0091] S411, the entity undirected graph construction unit 311 constructs an entity undirected graph with entities in the cloud network as entity nodes, and extracts initial features of the entity nodes.

[0092] S411 models the relationship between different entities in the cloud network at the micro level. Since the mutual influence relationship between entities at the micro level is relatively vague, the relationship between entities is modeled as an undirected graph, that is, an entity undirected graph is constructed with different entities as nodes.

[0093] In one embodiment, S411 includes: abstracting nodes and edges in the entity undirected graph; and constructing an initial attribute vector of the node.

[0094] Specifically, in the entity undirected graph, each node can correspond to an entity in the telecom cloud network, such as a VNF network element, a cloud service, a container, a virtual machine, a host, a switch, etc., or a port of the above entity. The edges between the nodes in the graph can be constructed based on the physical topological connection, communication relationship, and other interactive relationships between the two entities. In addition, the alarm, KPI, Operation, and other data generated on the entity can be encoded by one-hot encoding, BoW vector, etc., so as to construct the initial attribute vector of the entity node.

[0095] Figure 6 The figure shows a physical undirected graph according to an embodiment of the present application.

[0096] like Figure 6 As shown, in an application scenario, the cloud network includes 21 physical nodes G1-G21. The association relationship between the physical nodes is represented by the connection lines between the nodes.

[0097] S412, the layer-domain directed graph construction unit 312 constructs a layer-domain directed graph according to the layer-domain structure of the cloud network and with different domains as domain nodes.

[0098] Drawing on the prior knowledge of "first locating the domain in a large scale, then locating the entity in a fine-grained manner" and "top-down" traceability and positioning, S412 models the relationship between different domains in the cloud network from a macro level. Since the interaction between domains at the macro level has a clear direction, the relationship between domains is modeled as a directed graph, that is, a layer domain directed graph is constructed with different domains as nodes.

[0099] Specifically, in the layer-domain directed graph, each node may correspond to a domain in the cloud network, such as virtual computing resources, virtual storage resources, virtual network resources, physical computing resources, and the like.

[0100] In S412, based on the actual cloud network, the relationship between the entity nodes in the entity undirected graph and the domain nodes in the layer domain directed graph is constructed, that is, it is clear which domain the entity corresponding to the entity node belongs to. Specifically, based on the inclusion relationship between domains and entities in the actual network environment, the corresponding relationship between the domain nodes in the layer domain directed graph and the entity nodes in the entity undirected graph is constructed, and each domain node corresponds to a subgraph in the entity undirected graph.

[0101] Furthermore, in S412, edges between different domain nodes are constructed based on actual business relationships. That is, if there is an edge between the entity nodes corresponding to two domain nodes, then there is also an edge between the two domain nodes. At the same time, when determining the edge direction between domain nodes, based on the prior knowledge of "top-down analysis and positioning by layer" in actual business, the edge direction between domain nodes in different layers is from the upper domain to the lower domain, and the edge direction between domain nodes in the same layer is bidirectional (i.e., pointing to each other).

[0102] According to the method of the embodiment of the present application, by drawing on the prior knowledge of "first locating the domain on a large scale and then locating the entity on a fine-grained basis" and "top-down" tracing and positioning, a layer-domain directed graph is constructed according to the layer-domain structure of the cloud network, and a hierarchical message passing mechanism based on the hierarchical structure is proposed to characterize the domain nodes, which can reduce the probability of false positives and improve the accuracy of fault location.

[0103] According to the method of the embodiment of the present application, the cloud network is jointly characterized based on the entity undirected graph and the layer-domain directed graph, which improves the accuracy of the cloud network characterization, solves the problems of difficulty in fault location and computational efficiency caused by cross-layer and cross-domain, and can achieve more accurate and faster fault location.

[0104] Figure 7 FIG. 4 shows a layer domain directed graph according to an embodiment of the present application.

[0105] like Figure 7 As shown, the cloud network includes layer 601, layer 602, and layer 603. Layer 601 includes domain 610, layer 602 includes domain 621 and domain 622, and layer 603 includes domain 631, domain 632, and domain 633. There are bidirectional edges between domain 621 and domain 622 in the same layer. There is an edge pointing to domain 621 between domain 610 and domain 621, there is an edge pointing to domain 622 between domain 610 and domain 622, there is an edge pointing to domain 631 between domain 621 and domain 631, there is an edge pointing to domain 632 between domain 621 and domain 632, and there is an edge pointing to domain 633 between domain 622 and domain 633.

[0106] like Figure 6 As shown, entity nodes G1, G2, and G3 belong to domain 610 of layer 601. Entity nodes G4, G5, G6, and G7 belong to domain 621 of layer 602. Entity nodes G8, G9, G10, and G11 belong to domain 622 of layer 602. Entity nodes G12, G13, G14, and G15 belong to domain 631 of layer 603. Entity nodes G16, G17, and G18 belong to domain 632 of layer 603. Entity nodes G19, G20, and G21 belong to domain 633 of layer 603.

[0107] Further, in Figure 5 In the illustrated embodiment, S400, S411, and S412 are executed each time fault location is performed. In another embodiment, characteristic data of the cloud network used to perform S411 and S412 are obtained before fault location, S411 and S412 are executed, and the entity undirected graph and layer domain directed graph of the cloud network are saved. When fault location is performed, the saved entity undirected graph and layer domain directed graph are called, characteristic data of the cloud network used for fault location is obtained, and subsequent fault location operations are performed.

[0108] S421 , the entity node vectorized representation unit 321 performs representation learning on the entity nodes in the entity undirected graph based on the initial features of the entity nodes, and generates entity node vectorized representations of the entity nodes.

[0109] In S421, representation learning is performed on the entity undirected graph based on a spectral graph transformer (SGT) model.

[0110] Figure 8 The figure is a schematic diagram of the SGT model calculation process according to an embodiment of the present application.

[0111] like Figure 8 As shown, in each layer calculation process of the SGT model, S710 is first executed.

[0112] S710, using graph spectral filtering to decompose the input feature of each entity node into multiple sub-features of different frequencies.

[0113] Specifically, in one embodiment, during the calculation process of each layer, let the matrix Represents the entity node features of the layer input.

[0114] Each row of X i,: Represents node v i The eigenvector of each column X :,j Represents the value distribution of the jth feature dimension on different nodes, N is the number of nodes in the graph, and H is the number of dimensions of the node features.

[0115] Further, in one embodiment, for the eigendecomposition in S710, the specific calculation process is as follows:

[0116] (1) Each column feature X of the node feature matrix X :,j As a graph signal, it is decomposed into components of different frequencies using filters based on polynomials such as Chebyshev and Bernstein. k,j , taking the filter based on K-order Bernstein polynomial as an example, the calculation method is shown in the following formula 1:

[0117]

[0118] In Formula 1, k represents the kth Bernstein basis function, indicating the frequency order corresponding to the feature, I is the N×N unit matrix, and L is the Laplace matrix calculated based on the adjacency matrix of the entity undirected graph.

[0119] (2) Rearrange the components of different frequencies corresponding to all column features to obtain the feature matrix Z of different frequencies k , refer to the following formula 2:

[0120]

[0121] (3) Extract sub-features of different frequencies corresponding to each node from the feature matrix of different frequencies Refer to the following formula 3:

[0122]

[0123] Furthermore, in S710, in the process of decomposing sub-features, the Laplacian matrix corresponding to the graph adjacency matrix is ​​required. Therefore, in the above steps, both the topological information and the feature information, that is, the structural information of the graph, have been considered at the same time. In the feature decomposition stage, it has been integrated into the sub-features of different frequencies of each node.

[0124] After S710 , S720 is executed.

[0125] S720, performing feature transformation through a cross-attention mechanism between the input feature of the entity node and the sub-features of different frequencies, and obtaining a feature representation of the entity node after transformation.

[0126] In S720, after the graph filtering is completed, for feature transformation, a cross-attention mechanism based on focusing on sub-features of different frequencies is used to transform and learn the node features.

[0127] Specifically, in one embodiment, for each node v i , the input feature X of each layer i,: As the Q of Transformer, the sub-features of different frequencies obtained by decomposition As K and V of Transformer, adaptively learn node v i The degree of attention paid to different frequency sub-features allows each node to focus on the frequency band it needs to focus on, and the weighted sum of different frequency sub-features is used as the output feature of the corresponding node after transformation.

[0128] Based on the above process of S720, the attention range of each node changes from all nodes in the graph to its own sub-features of different frequencies, and the computational complexity of each layer changes from O(N 2 ) is reduced to O(KN), K<<N (O(KN), K<<N is a linear complexity). At the same time, the optimization space of each layer is also O(KN), which can reduce the possibility of overfitting of the model.

[0129] Combining the two stages of feature decomposition and feature transformation of S710 and S720 constitutes the core functional block of the SGT model. By continuously stacking this functional block, the module can go deeper with low computational complexity and low overfitting.

[0130] According to the method of the embodiment of the present application, graph filtering and Transformer are combined, and the graph structure and node features are fused into Transformer. While ensuring the model representation capability, the computational complexity is reduced to linear complexity, thereby improving the computational efficiency.

[0131] S422, the domain node representation unit 322 performs representation learning on the domain nodes in the layer domain directed graph based on the entity node vectorized representation of the entity node, and generates a domain node vectorized representation of the domain node.

[0132] In S422, domain nodes in the layer-domain directed graph are represented based on a message passing mechanism of the hierarchical structure.

[0133] Fig. 9 Shown is a schematic diagram of a domain node representation process according to an embodiment of the present application.

[0134] like Fig. 9 As shown, when performing domain node representation, S810 is first executed.

[0135] S810, using a multi-layer perceptron (MLP) to transform the vectorized representation of the entity nodes in the entity undirected graph, and according to the correspondence between the domain nodes in the layer-domain directed graph and the entity nodes in the entity undirected graph, summing pooling or mean pooling is performed on the entity node vectorized representation of the entity nodes corresponding to the domain nodes to obtain the domain node initial vectorized representation of the domain node, that is:

[0136]

[0137] In formula 4: v i is a domain node in the layer domain directed graph, is the domain node v i The initial vectorization representation of M(v i ) is the domain node v i The corresponding entity node set, Entity node v output by the SGT model j The entity node vector representation of MLP is a multi-layer perceptron neural network.

[0138] S820, based on the initial vectorized representation of the domain node obtained in S810, the domain node representation is updated layer by layer from top to bottom in a manner similar to a recurrent neural network (RNN) according to the hierarchical structure in the layer-domain directed graph to obtain a vectorized representation of the domain node.

[0139] That is, in the process of updating the domain node representation, the representation of each domain node is only related to the domain nodes pointing to it in the previous layer and the same layer, and the representation of the domain nodes in the kth layer is updated after the representation of the domain nodes in the k-1th layer is updated.

[0140] Specifically, to update the domain node v at the kth layer i As an example, the following steps are mainly included:

[0141] (1) Construct a pointer to v in the k-1th layer i The domain node set S k-1 (v i ) and points to v in the kth layer i The domain node set S k (v i ), and use these domain nodes as v i Adjacent domain nodes are used to update domain node v i Vectorized representation of .

[0142] (2) Use a shared linear transformation applied to v i and its neighboring domain nodes, and calculate v i and the attention score between adjacent domain nodes.

[0143] During the calculation process, for the domain nodes in the k-1th layer, since their domain node representations have been updated, their updated representations are used; for the domain nodes in the kth layer, their initial vectorized representations are used, that is:

[0144]

[0145]

[0146] In formulas 5 and 6: e ij is the domain node v i For adjacent domain node v j ’s attention score, W′ and W are the parameter matrices of the linear transformation, CONCAT(a,b) represents the concatenation operation of vector a and vector b. is the domain node v j The updated vectorized representation.

[0147] (3) Use the softmax function for all e ij Normalize:

[0148]

[0149] (4) Use the normalized weight coefficients above to calculate S(v i ) is weighted summed as the domain node v i Updated vectorized representation:

[0150]

[0151] S430, the fault classification module 330 uses the entity node vectorized representation and the domain node vectorized representation as feature inputs, respectively, to construct corresponding classifiers to obtain the first entity node failure probability and the domain node failure probability.

[0152] S440, the fault classification module 330 uses the fault probability of the domain node to bias the first entity node fault probability to obtain a domain-biased second entity node fault probability. (See S304)

[0153] Specifically, Fig.10 FIG. 4 is a schematic diagram of a failure probability calculation process according to an embodiment of the present application.

[0154] The fault classification module 330 performs the following steps: Fig.10 The following steps are shown in the flow.

[0155] S910, calculating a failure probability of a first entity node.

[0156] Specifically, in one embodiment, the vectorized representation of the entity node output by the SGT model in S421 is used as input, an MLP-based entity node classifier is constructed, and the failure probability of each entity node (the first entity node failure probability) is calculated:

[0157]

[0158] In formula 9: For entity node v i The probability of failure of the first entity node, For entity node v i The feature representation of .

[0159] S920, calculating domain node failure probability.

[0160] The domain node representation output by S422 is used as input to build an MLP-based domain node classifier to calculate the failure probability of each domain node:

[0161]

[0162] In formula 10: is the domain node v i The probability of failure, is the domain node v i The feature representation of .

[0163] S930, integrating the first entity node failure probability and the domain node failure probability to generate a second entity node failure probability. (Refer to S303)

[0164] In one embodiment, the bias is performed by multiplication:

[0165]

[0166] In formula 11: For entity node v i The failure probability after bias (failure probability of the second entity node), D(v i ) represents the entity node v i The corresponding domain node.

[0167] In the physical node classifier training process of S910, the binary cross entropy between the predicted physical node failure probability and the physical node failure label is used as the loss function, and the parameters to be trained (learned) in the entire physical node classifier model are optimized by gradient descent. In this way, if the failure probability of a non-physical node is high, but the failure probability of its domain is very low, in this case, due to the bias of the domain failure probability, the final failure probability of the physical node is still low, thereby reducing the frequency of false positives.

[0168] S440: Locate the failed entity node and / or domain node according to the second entity node failure probability and / or the domain node failure probability.

[0169] (Refer to S304)

[0170] Furthermore, in one embodiment, when a fault occurs in the cloud network, S400 and subsequent steps are triggered to locate the fault.

[0171] In another embodiment, when it is not confirmed whether there is a fault in the cloud network, S400 and subsequent steps are triggered to determine whether there is a fault, and to locate the fault if there is a fault.

[0172] In another embodiment, S400 and subsequent steps are triggered periodically to determine whether a fault exists, and to locate the fault if a fault exists.

[0173] In the description of the embodiments of the present application, for the convenience of description, the device is described by dividing it into various modules according to its functions. The division of each module is merely a division of logical functions. When implementing the embodiments of the present application, the functions of each module can be implemented in the same or one or more software and / or hardware.

[0174] Specifically, the device proposed in the embodiment of the present application can be fully or partially integrated into a physical entity during actual implementation, or it can be physically separated. And these modules can all be implemented in the form of software calling through processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software calling through processing elements, and some modules can be implemented in the form of hardware. For example, the detection module can be a separately established processing element, or it can be integrated in a chip of an electronic device. The implementation of other modules is similar. In addition, all or part of these modules can be integrated together, or they can be implemented independently. In the implementation process, each step of the above method or each of the above modules can be completed by an integrated logic circuit of hardware in a processor element or instructions in the form of software.

[0175] For example, the above modules may be one or more integrated circuits configured to implement the above methods, such as one or more application specific integrated circuits (ASIC), or one or more digital signal processors (DSP), or one or more field programmable gate arrays (FPGA), etc. For another example, these modules may be integrated together and implemented in the form of a system-on-a-chip (SOC).

[0176] An embodiment of the present application further provides an electronic device.

[0177] Fig.11 Schematic diagram of the structure of an electronic device according to an embodiment of the present application.

[0178] like Fig.11 As shown, the electronic device 1000 includes a memory 1002 for storing computer program instructions and a processor 1001 for executing the program instructions, wherein when the computer program instructions are executed by the processor 1001, the electronic device 1000 is triggered to execute the method steps described in the embodiments of the present application.

[0179] Specifically, in one embodiment of the present application, the above-mentioned one or more computer programs are stored in the above-mentioned memory 1002, and the above-mentioned one or more computer programs include instructions. When the above-mentioned instructions are executed by the above-mentioned electronic device 1000, the above-mentioned electronic device 1000 executes the method steps described in the embodiment of the present application.

[0180] It is to be understood that the structural description of the electronic device 1000 in the embodiment of the present application does not constitute a specific limitation on the electronic device 1000. In other embodiments of the present application, the electronic device 1000 may include other components besides the processor 1001 and the memory 1002.

[0181] The processor 1001 may be a device on a chip (SOC), and the processor 1001 may include a central processing unit (CPU), and may further include other types of processors.

[0182] The processor involved in the processor 1001 may include, for example, a CPU, a DSP, a microcontroller or a digital signal processor, and may also include a GPU, an embedded neural network processor (Neural-network Process Units, NPU) and an image signal processor (Image Signal Processing, ISP). The processor may also include necessary hardware accelerators or logic processing hardware circuits, such as ASICs, or one or more integrated circuits for controlling the execution of the program of the technical solution of the present application. In addition, the processor may have the function of operating one or more software programs, and the software programs may be stored in a storage medium.

[0183] The processor 1001 may include one or more processing units. For example, the processor may include an application processor (AP), a modem processor, a graphics processor (GPU), an image signal processor (ISP), a controller, a video codec, a digital signal processor (DSP), a baseband processor, and / or a neural-network processing unit (NPU), etc. Among them, different processing units may be independent components or integrated into one or more processors. In some embodiments, the electronic device 1000 may also include one or more processors 1001. Among them, the controller may generate an operation control signal according to the instruction opcode and the timing signal to complete the control of fetching and executing instructions.

[0184] In some embodiments, the processor 1001 may include one or more interfaces. The interface may include an inter-integrated circuit (I2C) interface, an integrated circuit sound (I2S) interface, a pulse code modulation (PCM) interface, a universal asynchronous receiver / transmitter (UART) interface, a mobile industry processor interface (MIPI), a general-purpose input / output (GPIO) interface, a SIM card interface, and / or a USB interface, etc. Among them, the USB interface is an interface that complies with the USB standard specification, and specifically can be a Mini USB interface, a Micro USB interface, a USB Type C interface, etc. The USB interface can be used to connect a charger to charge an electronic device, and can also be used to transmit data between an electronic device and a peripheral device.

[0185] The electronic device 1000 may also include an external memory interface, which is used to connect an external memory card, such as a Micro SD card, to expand the storage capacity of the electronic device. The external memory card communicates with the processor 1001 through the external memory interface to implement a data storage function. For example, files such as music and videos are stored in the external memory card.

[0186] The memory 1002 may include a code storage area and a data storage area. The code storage area may store an operating system. The data storage area may store data created during the use of the electronic device 1000, etc. In addition, the memory 1002 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more disk storage components, flash memory components, universal flash storage (UFS), etc.

[0187] The memory 1002 may be a read-only memory (ROM), other types of static storage devices that can store static information and instructions, a random access memory (RAM) or other types of dynamic storage devices that can store information and instructions, or an electrically erasable programmable read-only memory (EEPROM), a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), a magnetic disk storage medium or other magnetic storage device, or may be any computer-readable medium that can be used to carry or store desired program code in the form of instructions or data structures and can be accessed by a computer.

[0188] The processor 1001 and the memory 1002 may be integrated into one processing device, or more commonly, they are independent components.

[0189] Furthermore, the devices, apparatuses, and modules described in the embodiments of the present application may be implemented by computer chips or entities, or by products having certain functions.

[0190] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, devices, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0191] In the several embodiments provided in the present application, if any function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application.

[0192] Specifically, in one embodiment of the present application, a computer-readable storage medium is further provided, in which a computer program is stored. When the computer program is run on a computer, the computer executes the method provided in the embodiment of the present application.

[0193] An embodiment of the present application further provides a computer program product, which includes a computer program. When the computer program is run on a computer, the computer executes the method provided by the embodiment of the present application.

[0194] The embodiments described in this application are described with reference to the flowcharts and / or block diagrams of the methods, devices (apparatus), and computer program products according to the embodiments of the application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0195] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0196] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0197] It should also be noted that in the embodiments of the present application, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent the existence of A alone, the existence of A and B at the same time, and the existence of B alone. Among them, A and B can be singular or plural. The character " / " generally indicates that the previous and subsequent associated objects are in an "or" relationship. "At least one of the following" and similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b and c can be represented by: a, b, c, a and b, a and c, b and c, or a and b and c, where a, b, c can be single or multiple.

[0198] In the embodiments of the present application, the terms "include", "comprises" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of further restrictions, an element defined by the sentence "includes a ..." does not exclude the presence of other identical elements in the process, method, commodity or device including the element.

[0199] The present application may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0200] Each embodiment in this application is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0201] Those skilled in the art will appreciate that the various units and algorithm steps described in the embodiments of the present application can be implemented by a combination of electronic hardware, computer software, and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0202] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, apparatuses and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0203] The above is only a specific implementation of the present application. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. The protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A fault location method, characterized in that: The method comprises: The entities in the cloud network are regarded as physical nodes; According to the layer domain structure of the cloud network, different domains are used as domain nodes; Obtaining a first entity node failure probability of the entity node, and obtaining a domain node failure probability of the domain node; Using the domain node failure probability to bias the first entity node failure probability to obtain a second entity node failure probability; The failed entity node is located according to the second entity node failure probability.

2. The method according to claim 1, characterized in that The obtaining of the first entity node failure probability of the entity node and the obtaining of the domain node failure probability of the domain node include: Obtaining an entity node vectorized representation of the entity node, and obtaining a domain node vectorized representation of the domain node; The entity node vectorized representation and the domain node vectorized representation are respectively used as feature inputs to construct corresponding classifiers to obtain the first entity node failure probability and the domain node failure probability.

3. The method according to claim 2, characterized in that The obtaining of the entity node vectorized representation of the entity node and the obtaining of the domain node vectorized representation of the domain node include: Based on the entity nodes, construct an entity undirected graph, and extract initial features of the entity nodes; Based on the domain nodes, construct a layer domain directed graph; Based on the initial features of the entity nodes, performing representation learning on the entity nodes in the entity undirected graph to generate vectorized representations of the entity nodes; Based on the vectorized representation of the entity node, representation learning is performed on the domain nodes in the layer domain directed graph to generate the vectorized representation of the domain node.

4. The method according to claim 3, characterized in that In the layer-domain directed graph, the edge directions between domain nodes in the same layer are bidirectional, and the edge directions between domain nodes in different layers are from the upper layer domain to the lower layer domain.

5. The method according to claim 3 or 4, characterized in that: The performing representation learning on the entity nodes in the entity undirected graph based on the initial features of the entity nodes to generate vectorized representations of the entity nodes includes: Representation learning is performed on entity nodes in the entity undirected graph based on a spectral domain graph transformer model.

6. The method according to claim 5, characterized in that The performing representation learning on the entity nodes in the entity undirected graph based on the spectral domain graph transformer model comprises: Use graph filtering to decompose the input features of each entity node into multiple sub-features of different frequencies; Feature transformation is performed through a cross-attention mechanism between the input features of the entity node and the sub-features of different frequencies.

7. The method according to any one of claims 3 to 6, characterized in that: The step of performing representation learning on the domain nodes in the layer domain directed graph based on the vectorized representation of the entity nodes to generate the vectorized representation of the domain nodes includes: The entity node vectorized representation is transformed using a multi-layer perceptron, and, according to the correspondence between the domain nodes in the layer-domain directed graph and the entity nodes in the entity undirected graph, the entity node vectorized representation of the entity node corresponding to the domain node is summed and pooled or averaged to obtain an initial vectorized representation of the domain node; On the basis of the initial vectorized representation, the domain node representation is updated layer by layer from top to bottom according to the hierarchical structure in the layer-domain directed graph to obtain the domain node vectorized representation.

8. The method according to any one of claims 2 to 7, characterized in that: The method of taking the entity node vectorized representation and the domain node vectorized representation as feature inputs respectively and constructing corresponding classifiers to obtain the first entity node failure probability and the domain node failure probability includes: Taking the vectorized representation of the entity node as feature input, constructing an entity node classifier based on a multi-layer perceptron, and calculating the failure probability of each entity node; and / or, The vectorized representation of the domain node is used as a feature input, a domain node classifier based on a multi-layer perceptron is constructed, and the failure probability of each domain node is calculated.

9. The method according to claim 8, characterized in that The vectorized representation of the entity node is used as a feature input, an entity node classifier based on a multi-layer perceptron is constructed, and the failure probability of each entity node is calculated, wherein: In the process of training the entity node classifier, the binary cross entropy between the predicted entity node failure probability and the entity node failure label is used as the loss function, and the parameters to be trained in the entity node classifier are optimized by gradient descent.

10. The method according to any one of claims 1 to 9, characterized in that The using the domain node failure probability to bias the first entity node failure probability to obtain a second entity node failure probability includes: The first entity node failure probability is multiplied by the domain node failure probability, and the obtained product is the second entity node failure probability.

11. An electronic device, characterized in that: The electronic device comprises a memory for storing computer program instructions and a processor for executing the computer program instructions, wherein when the computer program instructions are executed by the processor, the electronic device is triggered to execute the method steps as claimed in any one of claims 1 to 10.

12. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, which, when executed on a computer, enables the computer to execute the method according to any one of claims 1 to 10.

Images