A network fault prediction method, device and electronic equipment

Abstract

This invention discloses a network fault prediction method, apparatus, and electronic device, comprising: preprocessing network fault data, converting feedback data from faulty network nodes into sequence sample data to generate time series data, where each time series data represents all fault types occurring at the current node over time; constructing a momentum-driven course comparison dual hypergraph gated convolutional network model, the network model including an embedding module, a representation learning module for a gated dual hypergraph convolutional network, a course comparison learning module based on a momentum-driven optimization strategy, and a prediction module, wherein the momentum-driven course comparison dual hypergraph gated convolutional network model uses the input time series data to predict faults in the network fault data. Through the above technical solution, this invention can solve the problem of existing technologies being unable to accurately predict anomalies, achieving accurate network fault prediction.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a network fault prediction method, and more particularly to a network fault prediction method, apparatus and electronic device based on momentum-driven curriculum contrastive dual hypergraph convolutional networks. Background Technology

[0002] With the widespread application of 5G technology, broadband networks are shouldering increasingly heavy responsibilities, while the requirements for broadband network quality are also constantly increasing. Timely and accurate prediction of network fault data can issue alerts to potential abnormal events, thereby greatly reducing the risk of network failures or even unexpected outages.

[0003] Traditional methods for predicting network failures have relied on expert experience, making it difficult to proactively predict network failures. Furthermore, due to the complexity and randomness of network failures, it is challenging to effectively predict them using traditional methods.

[0004] In recent years, with the development of graph deep learning, using graph deep learning methods for network fault prediction has become a trend. However, most methods cannot fully consider the higher-order structural relationships of network nodes and the network topology, which brings difficulties to network fault prediction in real-world scenarios. Furthermore, existing technologies have not fully explored the mutual influence relationships between faults. Summary of the Invention

[0005] The purpose of this invention is to provide a network fault prediction method based on momentum-driven course comparison dual hypergraph gated convolutional networks. This method aims to solve the problem that traditional prediction methods cannot capture the impact of different network faults on other faults due to the complexity of network faults, thus failing to accurately predict faults. Therefore, this invention provides a technical solution that can accurately predict network faults.

[0006] To address the aforementioned technical problems, this invention provides a network fault prediction method, the method comprising:

[0007] S1. Preprocess the network fault data, convert the feedback data of the faulty network node into sequence sample data, and generate fault time series data. The network fault data includes fault type, network node information and time node information. Each fault time series data represents all fault types that occur at the current node over time. The fault types include, but are not limited to, port fault, network card fault, router fault, network outage, power outage and network cable fault.

[0008] S2. Construct a momentum-driven course comparison dual hypergraph gated convolutional network model. The network model includes an embedding module, a representation learning module based on a gated dual hypergraph convolutional network, a course comparison learning module based on a momentum-driven optimization strategy, and a prediction module. The momentum-driven course comparison dual hypergraph gated convolutional network model uses the input time series data to predict faults in the network fault data.

[0009] Preferably, the construction of the momentum-driven course comparison dual hypergraph gated convolutional network model specifically includes:

[0010] S21. Construct an embedding module. Taking the fault time series data as input, initialize the features of the fault nodes and transform the time series data into an embedding vector s. Then, construct a hypergraph structure g based on the embedded time series. U，j Each fault data point is represented by the following feature:

[0011] G U ={g U，base g U，1 …, g U，k};

[0012] g U，j ={U, ε U，j}, ε U，j Supergraph g U，j In the superedge, the supergraph g U，j All hyperedges in the network share the same set of faulty network nodes U. For a faulty network node u∈U, in g U，j Introduce a hyperedge ε U，j Connect {u|u∈U, (u,i)∈E j}, that is, in the set of faulty network nodes U, it represents all nodes that are directly connected to the faulty network node u through the interaction type Ej;

[0013] S22. Construct a representation learning module for a hypergraph convolutional network based on a gated mechanism to obtain a node representation X based on a faulty node hypergraph;

[0014] S23, Regarding the hypergraph g U，j Perform a graph diffusion enhancement operation to obtain an enhanced diffusion-enhanced supergraph g′. U，j To extract complex high-order interaction relationships between faulty nodes, the graph diffusion operation process is formulated as follows:

[0015]

[0016] in It is a parameter that controls the diffusion rate of network nodes, g′ U，j It is a super map g U，jThe new diffusion-enhanced hypergraph after the graph diffusion transformation;

[0017] After propagation at layer l, the READOUT function is used to process the hypergraph g. U，j and enhanced diffusion hypergraph g′ U，j Node features learned in and Perform embedding operations and The final output z is obtained by using a multilayer perceptron for dimensionality reduction of the downstream task. U and z U′ The formula is as follows:

[0018]

[0019]

[0020]

[0021]

[0022] Among them, z U Represents the hypergraph g U，j Embedded representation of nodes, z U′ Represents the diffusion hypergraph g′ U，j Embedded representation of nodes;

[0023] S24. Construct a momentum-driven comparative learning module for courses:

[0024] Based on the sampling strategy of course learning, in g′ U，j Find the negative sampling embedding z′ of c faulty network nodes with different difficulties. c And define a scoring function S(·), embedding different negative samples into z′ c Mapping to numerical fraction S(z′) c The difficulty is measured by setting the scoring function S(·) to sim(z). v , z′ c To measure the difficulty of negative samples:

[0025] S(z′ c ) = sim(z v , z′ c )=(z v -z′ c ) T Σ .-1 (z v -z′ c );

[0026] Among them, z v Indicates something different from z′ cFaulty network nodes are being sampled and embedded;

[0027] The single negative sample embedding z′ is obtained during the entire contrastive learning process. c The fraction S(z′) c After that, a step function δ(t) is used to arrange the introduction of negative samples into the training process, where the step function δ(t) specifies the size of the negative samples in each step t. The negative samples in step t consist of δ(t) of the lowest scores. T represents the total number of training steps.

[0028] δ(t)=(t / T) φ ·K;

[0029] Where φ is a smoothing parameter to control the pace function guiding the training speed, and during contrastive learning, according to sim(z) v , z′ c Negative samples are sorted from lowest to highest score.

[0030] Maximize the hypergraph g U，j and diffusion-enhanced hypergraph g′ U，j The mutual information between them is used to estimate the contrastive loss using noise. The final loss function of the contrastive learning module can be expressed as follows:

[0031]

[0032] Among them, the similarity measurement function sim(z) v , z′ v ) is used to calculate each positive sample pair {z v , z′ v} and positive and negative sample pairs {z v , z′ c The similarity of identical nodes in the array is given by τ, where τ represents the temperature parameter. This indicates that the embedding of C negative samples is guided during the training process using a scoring function;

[0033] In the comparative learning process, a momentum update mechanism is used to update all parameters. The momentum-driven update iteration process of the model parameters is as follows:

[0034]

[0035] Where μ is an adjustable parameter controlling the momentum of training information extraction, and γ t and They represent the hypergraph g respectively. U，j and enhanced diffusion hypergraph g′ U，j Learn the parameter set during iterative optimization training t rounds;

[0036] S25. Construct a prediction module based on the initial embeddings of each sequence.g And the current faulty hypergraph network node representation X U，j The inner product is used to calculate the fraction Ω of all nodes in the hypergraph. j Then, the softmax function is used to obtain the output of predicting that the node is a possible failure in the next time period:

[0037] Ω j =s g T X U，j ;

[0038]

[0039] in, This indicates a prediction of whether a network failure will occur in the next period of time.

[0040] Then, using existing nodes in the fault hypergraph network as edges for positive sample sampling, a prediction loss function is designed by maximizing the edge probability of positive samples and minimizing the edge probability of negative samples. The loss function is defined as:

[0041]

[0042] Where σ is the sigmoid activation function, λ is the weight parameter to balance the importance of positive and negative samples, P(u) defines the distribution of candidate fault nodes u, n is the number of negative samples, the existing edges in the multipart graph are taken as positive samples, for each positive sample edge (u, i), n negative edges with nodes u and i are randomly selected as negative sample edges;

[0043] Finally, the dual-hypergraph representation learning module and the momentum-driven optimization strategy-based course comparison learning module are unified into a comprehensive learning framework, and the final prediction learning objective loss is defined as:

[0044] l = l r +ζl NCE ;

[0045] Here, ζ is a variable factor that controls the contrastive learning task.

[0046] Preferably, the representation learning module for constructing a hypergraph convolutional network based on a gating mechanism to obtain the node representation X based on the fault node hypergraph specifically includes:

[0047] S221. Introduce an incidence matrix H in the hypergraph to describe the relationship between nodes and hyperedges in the faulty hypergraph network. Given a hypergraph g... U，j ={U, ε U，j}, where j∈{base, 1, ..., k} represents the different interaction types between faulty network nodes, and k is the number of interaction types between faulty network nodes. Let g be defined.U，j The given correlation matrix is:

[0048]

[0049] in, ε U，j Indicates in the hypergraph g U，j The set of hyperedges, where j∈{base, 1, ..., k}, is based on the hypergraph g. I，j Define the cross-correlation matrix

[0050] Let diagonal matrix and Let represent the node degree matrix and the hyperedge degree matrix, respectively. B U，j (e, e) = ∑ u∈U H U，j (u, e);

[0051] S222, Using the hypergraph convolution operator, normalized hypergraph convolution operation is used to learn the embedding of each hypergraph, g U，j The hypergraph convolution is represented as:

[0052]

[0053] Where σ represents the nonlinear activation function - ReLU function, This represents the user characteristics of layer l. It is an identity matrix. It is a learnable transformation matrix, F l and F l+1 Indicates the embedding dimensions of the l-th layer and the (l+1)-th layer;

[0054] S223, For the hypergraph g U，j Through hypergraph convolution operations, from each hypergraph g U，j By learning node features and considering different types of interactions between nodes in the faulty network, the node representations of the faulty network are obtained:

[0055] X = {X U，base X U，1 , ..., X U，k}

[0056] Furthermore, the method also includes:

[0057] Design a gated information transfer mechanism within a hypergraph, based on the hypergraph g. U，j The feature information X of the faulty node learned by the construction method U，base Passed to every hypergraph node X U，j In this context, the propagation process is as follows:

[0058]

[0059]

[0060] Where β is the gating weight, The hypergraph g represents the hypergraph that has undergone information propagation and fusion. U，j The faulty hypergraph network node representation, thereby realizing the hypergraph g U，j Information transfer between different interaction types between nodes in a faulty hypergraph and b β These are learnable parameters.

[0061] Secondly, in order to solve the technical problem of the present invention, embodiments of the present invention also provide a network fault prediction device, the device comprising:

[0062] The data preprocessing module is used to preprocess network fault data, convert feedback data from faulty network nodes into sequence sample data, and generate fault time series data. The network fault data includes fault type, network node information, and time node information. Each fault time series data represents all fault types that occur at the current node over time. The fault types include, but are not limited to, port fault, network card fault, router fault, network outage, power outage, and network cable fault.

[0063] The prediction module is used to construct a momentum-driven course comparison dual hypergraph gated convolutional network model. The network model includes an embedding module, a representation learning module for a gated dual hypergraph convolutional network, a course comparison learning module based on a momentum-driven optimization strategy, and a prediction module. The momentum-driven course comparison dual hypergraph gated convolutional network model uses the input time series data to predict faults in the network fault data.

[0064] Thirdly, in order to solve the technical problems of the present invention, embodiments of the present invention also provide an electronic device, including a processor and a memory, wherein the memory is used to store a computer program, the computer program includes program instructions, and the processor is configured to call the program instructions to execute the method described above.

[0065] Fourthly, in order to solve the technical problem of the present invention, embodiments of the present invention also provide a computer-readable storage medium storing a computer program, the computer program including program instructions, which, when executed by a processor, cause the processor to perform the method described above.

[0066] The embodiments of the present invention, by adopting the above technical solutions, have the following effects: preprocessing network fault data, constructing a momentum-driven course comparison dual hypergraph gated convolutional network model to predict faults in the network fault data, and using dual hypergraph gated convolution and momentum-driven course comparison learning methods, achieving accurate prediction before network faults occur, enabling accurate analysis of the causes of network faults and their complex correlations, thereby achieving accurate prediction of network faults and improving the accuracy of network fault prediction. Attached Figure Description

[0067] The preferred embodiments will now be described in a clear and easy-to-understand manner, in conjunction with the accompanying drawings, to further explain the above-mentioned characteristics, technical features, advantages, and implementation methods of the present invention.

[0068] Figure 1 This is a flowchart of the network fault prediction method based on momentum-driven course comparison dual hypergraph convolutional network according to an embodiment of the present invention;

[0069] Figure 2 This is a structural diagram of another embodiment of the present invention: network fault prediction based on momentum-driven course comparison dual hypergraph convolutional network.

[0070] Figure 3 This is a schematic diagram of the structure of an electronic device that uses a momentum-driven course comparison dual hypergraph convolutional network for network fault prediction, as provided in an embodiment of the present invention. Detailed Implementation

[0071] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the specific implementation methods of the present invention will be described below with reference to the accompanying drawings. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings and other implementation methods can be obtained based on these drawings without any creative effort.

[0072] To keep the drawings concise, each figure only schematically shows the parts relevant to the invention, and these do not represent the actual structure of the product. Furthermore, to facilitate understanding, in some figures, only one of components with the same structure or function is schematically depicted, or only one is labeled. In this document, "one" not only means "only one," but can also mean "more than one."

[0073] The following detailed description of the implementation of the technical solution of the present invention will be based on some specific embodiments.

[0074] like Figure 1As shown in the figure, a network fault prediction method for a momentum-driven curriculum comparison dual hypergraph gated convolutional network according to an embodiment of the present invention includes the following steps:

[0075] S1. Preprocess the network fault data, convert the feedback data of the network node device that caused the fault into sequence sample data, and generate fault time series data. The network fault data includes fault type, network node device information and time node information. Each fault time series data represents all fault types that occur in the current device over time. The fault types include, but are not limited to, port fault, network card fault, router fault, network outage, power outage and network cable fault.

[0076] S2. Construct a momentum-driven course comparison dual hypergraph gated convolutional network model. The network model includes an embedding module, a representation learning module based on a gated dual hypergraph convolutional network, a course comparison learning module based on a momentum-driven optimization strategy, and a prediction module. The momentum-driven course comparison dual hypergraph gated convolutional network model uses the input fault time series data to predict faults in the network fault data.

[0077] In this embodiment of the invention, the collected network fault data is preprocessed, the feedback data of the faulty network node device is converted into sequence sample data, fault time series data is generated, and the data sequence of all fault types in the time series is obtained.

[0078] Specifically, constructing a momentum-driven course comparison dual hypergraph gated convolutional network model includes:

[0079] S21. Construct an embedding module that takes fault time-series data as input, initializes the features of fault nodes, and transforms the time-series data into embedding vectors s; then, constructs a hypergraph structure g based on the embedded time-series data. U For each fault data point, j is represented by the following feature:

[0080] G U ={g U，base g U，1 …, g U，k};

[0081] g U，j ={U, ε U，j}, ε U，j Supergraph g U，j In the superedge, the supergraph g U，j All hyperedges in the network share the same set of faulty network nodes U. For a faulty network node u∈U, in g U，j Introduce a hyperedge ε U，j Connect {u|u∈U, (u,i)∈E j}, that is, in the faulty network node set U, it represents the interaction type E. j All nodes directly connected to the faulty network node u;

[0082] S22. Construct a representation learning module for a hypergraph convolutional network based on a gated mechanism to obtain a node representation X based on a faulty node hypergraph, specifically including:

[0083] S221. Introduce an incidence matrix H in the hypergraph to describe the relationship between nodes and hyperedges in the faulty hypergraph network. Given a hypergraph g... U，j ={U, ε U，j}, where j∈{base, 1, ..., k} represents the different interaction types between faulty network nodes, and k is the number of interaction types between faulty network nodes. Let g be defined. U，j The given correlation matrix is:

[0084]

[0085] in, ε U，j Indicates in the hypergraph g U，j The set of hyperedges, where j∈{base, 1, ..., k}, is based on the hypergraph g. I，j Define the cross-correlation matrix Let diagonal matrix and Let represent the node degree matrix and the hyperedge degree matrix, respectively. B U，j (e, e) = ∑ u∈U H U，j (u, e);

[0086] S222, Using the hypergraph convolution operator, normalized hypergraph convolution operation is used to learn the embedding of each hypergraph, g U，j The hypergraph convolution is represented as:

[0087]

[0088] Where σ represents the nonlinear activation function - ReLU function, This represents the user characteristics of layer l. It is an identity matrix. It is a learnable transformation matrix, F l and F l+1 Indicates the embedding dimensions of the l-th layer and the (l+1)-th layer;

[0089] S223, For the hypergraph g U，j Through hypergraph convolution operations, from each hypergraph g U，j By learning node features and considering different types of interactions between nodes in the faulty network, the node representations of the faulty network are obtained:

[0090] X = {X U，base X U，1 , ..., X U，k};

[0091] S23, Regarding the hypergraph g U，j Perform a graph diffusion enhancement operation to obtain an enhanced diffusion-enhanced supergraph g′. U，j To extract complex high-order interaction relationships between faulty nodes, the graph diffusion operation process is formulated as follows:

[0092]

[0093] in It is a parameter that controls the diffusion rate of network nodes, g′ U，j It is a super map g U，j The new diffusion-enhanced hypergraph after the graph diffusion transformation;

[0094] After propagation at layer l, the READOUT function is used to process the hypergraph g. U，j and enhanced diffusion hypergraph g′ U，j Node features learned in and Perform embedding operations and The final output z is obtained by using a multilayer perceptron for dimensionality reduction of the downstream task. U and z U′ The formula is as follows:

[0095]

[0096]

[0097]

[0098]

[0099] Among them, z U Represents the hypergraph g U，j Embedded representation of nodes; z U′ Represents the diffusion hypergraph g′ U，j Embedded representation of nodes.

[0100] S24. Construct a momentum-driven comparative learning module for courses:

[0101] Based on the sampling strategy of course learning, in g′ U，j Find the negative sampling embedding z′ of c faulty network nodes with different difficulties. c And define a scoring function S(·), embedding different negative samples into z′ cMapping to numerical fraction S(z′) c The difficulty is measured by setting the scoring function S(·) to sim(z). v , z′ c To measure the difficulty of negative samples:

[0102] S(z′ c ) = sim(z v , z′ c )=(z v -z′ c ) T Σ .-1 (z v -z′ c );

[0103] Among them, z v Indicates something different from z′ c Faulty network nodes are being sampled and embedded;

[0104] The single negative sample embedding z′ is obtained during the entire contrastive learning process. c The fraction S(z′) c After that, a step function δ(t) is used to arrange the introduction of negative samples into the training process, where the step function δ(t) specifies the size of the negative samples in each step t. The negative samples in step t consist of δ(t) of the lowest scores. T represents the total number of training steps.

[0105] δ(t)=(t / T) φ ·K;

[0106] Where φ is a smoothing parameter to control the pace function guiding the training speed, and during contrastive learning, according to sim(z) v , z′ c Negative samples are sorted from lowest to highest score.

[0107] Maximize the hypergraph g U，j and diffusion-enhanced hypergraph g′ U，j The mutual information between them is used to estimate the contrastive loss using noise. The final loss function of the contrastive learning module can be expressed as follows:

[0108]

[0109] Among them, the similarity measurement function sim(z) v , z′ v ) is used to calculate each positive sample pair {z v , z′ v} and positive and negative sample pairs {z v , z′ c The similarity of identical nodes in the array is given by τ, where τ represents the temperature parameter. This indicates that the embedding of C negative samples is guided during the training process using a scoring function;

[0110] In the comparative learning process, a momentum update mechanism is used to update all parameters. The momentum-driven update iteration process of the model parameters is as follows:

[0111]

[0112] Where μ is an adjustable parameter controlling the momentum of training information extraction, and γ t and They represent the hypergraph g respectively. U，j and enhanced diffusion hypergraph g′ U，j Learn the parameter set during iterative optimization training t rounds;

[0113] The input to this contrastive learning module is a hypergraph g. U，j The other is the node embedding representation, and the diffusion hypergraph g′. U，j The embedded representation of a node. For g U，j z is obtained by positive sampling of the node's embedding representation. v For the diffusion supergraph g′ U，j The node's embedding representation is positively sampled to obtain z′ v Meanwhile, the diffusion supergraph g′ U，j The node obtains z′ using a negative sampling strategy for course learning based on the scoring function. c Unlike traditional contrastive learning, my proposed contrastive learning loss function also introduces a negative sampling z′ based on the course learning strategy. c .

[0114] This invention presents a momentum-driven contrastive learning framework. First, the framework sorts negative samples based on their scores. Then, it uses a scoring function and a pace function to weigh the differences in difficulty among the negative samples, thereby introducing different negative samples into the contrastive learning training. Simultaneously, a momentum-driven optimization strategy is used to enhance the differences in training between the two views of the contrastive learning framework (i.e., the original hypergraph network and the diffusion-enhanced hypergraph network), further extracting the complex higher-order nature of the hypergraph structure in learning fault network feature representations.

[0115] First, based on the sampling strategy learned from the course, in the diffusion-enhanced hypergraph g′ U，j We can find the embedding z′ of negative sampling for c faulty network nodes of different difficulty. c Therefore, a scoring function S(·) is defined, which embeds different negative samples into z′. c Mapping to numerical fraction S(z′) c To measure this difficulty, the scoring function S(·) is set as sim(z) v , z′c This is used to measure the difficulty of negative samples.

[0116] S(z′ c ) = sim(z v , z′ c )=(z v -z′ c ) T Σ .-1 (z v -z′ c );

[0117] Among them, z v It also indicates that it is different from z′ c The faulty network node is being sampled and embedded.

[0118] Next, during the entire contrastive learning process, a single negative sample embedding z′ is obtained. c The fraction S(z′) c After that, a step function δ(t) is used to schedule the introduction of negative samples into the training process. The step function Sδ(t) specifies the size of the samples that can be negatively sampled in each step t. The negative samples in step t consist of the δ(t) lowest-scoring negative samples. Negative samples are sampled uniformly in batches, and T represents the total number of training steps:

[0119] δ(t)=(t / T) φ ·K;

[0120] Where φ is a smoothing parameter used to control the speed of the training process guided by the pacing function. During contrastive learning, according to sim(z)... v , z′ c Negative samples are sorted from lowest to highest score to facilitate their introduction into training.

[0121] Finally, maximize the hypergraph g. U，j and diffusion-enhanced hypergraph g′ U，j Mutual information is generated between them. Noise is used to estimate the contrastive loss. The final loss function of the contrastive learning module can be expressed as follows:

[0122]

[0123] Among them, the similarity measurement function sim(z) v , z′ v ) is used to compute each positive sample pair {zv, z′} v} and positive and negative sample pairs {z v , z′ c The similarity of identical nodes in the array is given by τ, where τ represents the temperature parameter. This indicates that the embedding of C negative samples is guided during the training process using a scoring function.

[0124] Meanwhile, to avoid the potential collapse of training that relies too heavily on negative samples, and considering the two different views within the contrastive learning framework, this embodiment of the invention employs a momentum update mechanism to update all parameters during the contrastive learning process. The momentum-driven update iteration process for the model parameters is as follows:

[0125]

[0126] Where μ is an adjustable parameter controlling the momentum of training information extraction, and γ t and These represent the original hypergraph network G. U and diffusion-enhanced hypergraph G' U The parameter set is learned during iterative optimization training t rounds.

[0127] S25. Construct the prediction module:

[0128] Construct a prediction module based on the initial embedding sg of each sequence and the current faulty hypergraph network node representation X. U，j The inner product is used to calculate the fraction Ω of all nodes in the hypergraph. j Then, the softmax function is used to obtain the output of predicting that the node is a possible failure in the next time period:

[0129] Ω j =s g T X U，j ;

[0130]

[0131] in, This indicates a prediction of whether a network failure will occur in the next period of time.

[0132] Then, using existing nodes in the fault hypergraph network as edges for positive sample sampling, a prediction loss function is designed by maximizing the edge probability of positive samples and minimizing the edge probability of negative samples. The loss function is defined as:

[0133]

[0134] Where σ is the sigmoid activation function, λ is the weight parameter to balance the importance of positive and negative samples, P(u) defines the distribution of candidate fault nodes u, n is the number of negative samples, the existing edges in the multipart graph are taken as positive samples, for each positive sample edge (u, i), n negative edges with nodes u and i are randomly selected as negative sample edges;

[0135] Finally, the dual-hypergraph representation learning module and the momentum-driven optimization strategy-based course comparison learning module are unified into a comprehensive learning framework, and the final prediction learning objective loss is defined as:

[0136] l = l r +ζl NCE ;

[0137] Here, ζ is a variable factor that controls the contrastive learning task.

[0138] Even if existing nodes in the faulty hypergraph network are used as edges for positive sample sampling, the prediction loss function is designed by maximizing the edge probability of positive samples and minimizing the edge probability of negative samples. The loss function is defined as follows:

[0139]

[0140] Where σ is the sigmoid activation function, λ is the weight parameter to balance the importance of positive and negative samples, P(u) defines the distribution of candidate fault nodes u, and n is the number of negative samples. The existing edges in the multipart graph are taken as positive samples. For each positive sample edge (u, i), n negative edges with nodes u and i are randomly selected as negative sample edges.

[0141] Finally, the dual-hypergraph representation learning module and the momentum-driven optimization-based course comparison learning module are unified into a unified learning framework. Formally, the final prediction learning objective loss is defined as:

[0142] l = l r +ζl NCE ;

[0143] Here, ζ is a variable factor that controls the contrastive learning task.

[0144] Preferably, the network fault prediction method of this embodiment of the invention further includes:

[0145] Design a gated information transfer mechanism within a hypergraph, based on the hypergraph g. U，j The feature information X of the faulty node learned by the construction method U，base Passed to every hypergraph node X U，j In this context, the propagation process is as follows:

[0146]

[0147]

[0148] Where β is the gating weight, The hypergraph g represents the hypergraph that has undergone information propagation and fusion. U，j The faulty hypergraph network node representation, thereby realizing the hypergraph gU，j Information transfer between different interaction types between nodes in a faulty hypergraph and b β These are learnable parameters.

[0149] See Figure 3 , Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application, such as... Figure 3 As shown, the electronic device 1000 includes a processor 1001 and a memory 1002, which are interconnected via a communication bus 1003. The communication bus 1003 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus 1003 can be divided into an address bus, a data bus, a control bus, etc. For ease of illustration, Figure 3 The bus is represented by a single thick line, but this does not indicate that there is only one bus or one type of bus. Memory 1002 stores computer programs, which include program instructions. Processor 1001 is configured to invoke these program instructions, which include instructions for execution. Figure 1 It includes some or all of the steps in the methods.

[0150] The processor 1001 may be a general-purpose central processing unit (CPU), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits used to control the execution of the above-mentioned program.

[0151] The memory 1002 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), 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 optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. The memory may exist independently and be connected to the processor via a bus. The memory may also be integrated with the processor.

[0152] The electronic device 1000 may further include a communication module 1004 and a display 1005. The communication module 1004 can communicate with the optical tracking device. The communication module 1004 can be a wireless communication module (e.g., a WiFi module, a Bluetooth module, etc.) or a wired communication module.

[0153] In addition, the electronic device 1000 may also include general components such as communication interfaces (e.g., USB interfaces, microphone interfaces, etc.) and antennas, which will not be described in detail here.

[0154] This invention also provides a computer-readable storage medium storing a computer program for electronic data interchange, which causes a computer to perform some or all of the steps of any of the network fault prediction methods described in the above method embodiments.

[0155] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, as some steps may be performed in other orders or simultaneously according to this application. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0156] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0157] In the several embodiments provided in this application, it should be understood that the disclosed apparatus can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical or other forms.

[0158] 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 units can be selected to achieve the purpose of this embodiment according to actual needs.

[0159] Furthermore, the functional units in the various embodiments of the application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software program module.

[0160] If the integrated unit is implemented as a software program module and sold or used as an independent product, it can be stored in a computer-readable storage device (CMD). Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory 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 this application. The aforementioned memory includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0161] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage device, which may include: a flash drive, a read-only memory, a random access memory, a magnetic disk, or an optical disk, etc.

[0162] The embodiments of this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The description of the above embodiments is only for the purpose of helping to understand the method and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

[0163] It should be noted that the above embodiments can be freely combined as needed. The above are merely preferred embodiments of the present invention. It should be pointed out that for those skilled in the art, several improvements and modifications can be made without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A network fault prediction method, characterized in that, The method includes: S1. Preprocess the network fault data, convert the feedback data of the faulty network node into sequence sample data, and generate fault time series data. The network fault data includes fault type, network node information and time node information. Each fault time series data represents all fault types that occur at the current node over time. The fault types include, but are not limited to, port fault, network card fault, router fault, network outage, power outage and network cable fault. S2. Construct a momentum-driven course comparison dual hypergraph gated convolutional network model. The network model includes an embedding module, a representation learning module for the gated dual hypergraph convolutional network, a course comparison learning module based on a momentum-driven optimization strategy, and a prediction module. The momentum-driven course comparison dual hypergraph gated convolutional network model uses the input time-series data to predict faults in the network fault data. Specifically, constructing the momentum-driven course comparison dual hypergraph gated convolutional network model includes: S21. Construct an embedding module. Taking the fault time series data as input, initialize the features of the fault nodes and transform the time series data into an embedding vector s. Then, construct a hypergraph structure g based on the embedded time series. U,j Each fault data point is represented by the following feature: G U ={g U,base ,g U,1 ...,g U,k }; g U,j ={U,ε U,j }, ε U,j Supergraph g U,j In the superedge, the supergraph g U,j All hyperedges in the network share the same set of faulty network nodes U. For a faulty network node u∈U, in g U,j Introduce a hyperedge ε U,j Connect {u|u∈U,(u,i)∈E j }, that is, in the faulty network node set U, it represents the interaction type E. j All nodes u that are directly connected to the faulty network node i; S22. Construct a representation learning module for a hypergraph convolutional network based on a gated mechanism to obtain a node representation X based on a faulty node hypergraph; S23, Regarding the hypergraph g U,j Perform a graph diffusion enhancement operation to obtain an enhanced diffusion-enhanced supergraph g′. U,j To extract complex high-order interaction relationships between faulty nodes, the graph diffusion operation process is formulated as follows: in It is a parameter that controls the diffusion rate of network nodes, g′ U,j It is a super map g U,j The new diffusion-enhanced hypergraph after the graph diffusion transformation; After propagation at layer l, the READOUT function is used to process the hypergraph g. U,j and enhanced diffusion hypergraph g′ U,j Node features learned in and Perform embedding operations and The final output z is obtained by using a multilayer perceptron for dimensionality reduction of the downstream task. U and z U′ The formula is as follows: Among them, z U Represents the hypergraph g U,j Embedded representation of nodes, z U′ Represents the diffusion hypergraph g′ U,j Embedded representation of nodes; S24. Construct a momentum-driven comparative learning module for courses: Based on the sampling strategy of course learning, in g′ U,j Find the negative sampling embedding z′ of c faulty network nodes with different difficulties. c And define a scoring function S(·), embedding different negative samples into z′ c Mapping to numerical fraction S(z′) c The difficulty is measured by setting the scoring function S(·) to sim(z). v ,z′ c To measure the difficulty of negative samples: S(z′ c )=sim(z v ,With' c )=(of v -With' c ) T Σ .-1 (With v -With' c ); Among them, z v Indicates something different from z′ c Faulty network nodes are being sampled and embedded; Obtaining a single negative sample embedding z′ throughout the entire contrastive learning process c The fraction S(z′) c After that, a step function δ(t) is used to schedule the introduction of negative samples into the training process, where the step function δ(t) specifies the size of the negative samples in each step t. The negative samples in step t consist of δ(t) of the lowest scores. T represents the total number of training steps. δ(t)=(t / T) φ ·K; Where φ is a smoothing parameter to control the pace function guiding the training speed, and during contrastive learning, according to sim(z) v ,z′ c Negative samples are sorted from lowest to highest score. Maximize the hypergraph g U,j and diffusion-enhanced hypergraph g′ U,j The mutual information between them is used to estimate the contrastive loss using noise. The final loss function of the contrastive learning module can be expressed as follows: Among them, the similarity measurement function sim(z) v ,z v ′) is used to calculate each positive sample pair {z v ,z v ′} and positive and negative sample pairs {z v ,z c The similarity of identical nodes in '}, where τ represents the temperature parameter. This indicates that the embedding of C negative samples during the training process is guided by a scoring function; In the comparative learning process, a momentum update mechanism is used to update all parameters. The momentum-driven update iteration process of the model parameters is as follows: Where 'a' is an adjustable parameter that controls the momentum of the training information extraction process. and They represent the hypergraph g respectively. U,j and enhanced diffusion hypergraph g′ U,j Learn the parameter set during iterative optimization training t rounds; S25. Construct a prediction module, for each sequence's initial embedding s g And the current faulty hypergraph network node representation X U,j Perform inner product calculations to compute the fraction Ω of all nodes in the hypergraph. j Then, the softmax function is used to obtain the output of predicting that the node is a possible failure in the next time period: Oh j =s g T X U,j ; in, This indicates a prediction of whether a network failure will occur in the next period of time. Then, existing nodes in the fault hypergraph network and the diffusion-enhanced fault hypergraph are used as edges for positive sample sampling, respectively. The prediction loss function is designed by maximizing the edge probability of positive samples and minimizing the edge probability of negative samples. The loss function is defined as: Where σ is the sigmoid activation function, λ is the weight parameter to balance the importance of positive and negative samples, P(u) defines the distribution of candidate fault nodes u, n is the number of negative samples, the existing edges in the multipart graph are taken as positive samples, for each positive sample edge (u,i), n negative edges with nodes u and i are randomly selected as negative sample edges; Finally, the dual-hypergraph representation learning module and the momentum-driven optimization strategy-based course comparison learning module are unified into a comprehensive learning framework, and the final prediction learning objective loss is defined as: l=l r +ζl NCE ; Here, ζ is a variable factor that controls the contrastive learning task.

2. The network fault prediction method as described in claim 1, characterized in that, The representation learning module for constructing a hypergraph convolutional network based on a gated mechanism to obtain the node representation X based on the fault node hypergraph specifically includes: S221. Introduce an incidence matrix H in the hypergraph to describe the relationship between nodes and hyperedges in the faulty hypergraph network. Given a hypergraph g... U,j ={U,ε U,j }, where j∈{base,1,...,k} represents the different interaction types between faulty network nodes, and k is the number of interaction types between faulty network nodes. Let g be defined as... U,j The given correlation matrix is: in, ε U,j Indicates in the hypergraph g U,j The set of hyperedges, where j∈{base,1,...,k}, is based on the hypergraph g. I,j Define the cross-correlation matrix Let diagonal matrix and Let represent the node degree matrix and the hyperedge degree matrix, respectively. B U,j (e,e)=∑ u∈U H U,j (u,e); S222, Using the hypergraph convolution operator, normalized hypergraph convolution operation is used to learn the embedding of each hypergraph, g U,j The hypergraph convolution is represented as: Where σ represents the nonlinear activation function - ReLU function, This represents the user characteristics of layer l. It is an identity matrix. It is a learnable transformation matrix, F l and F l+1 Indicates the embedding dimensions of the l-th layer and the (l+1)-th layer; S223, For the hypergraph g U,j Through hypergraph convolution operations, from each hypergraph g U,j By learning node features and considering different types of interactions between nodes in the faulty network, the node representations of the faulty network are obtained: X={X U,base ,X U,1 ,...,X U,k }。 3. The network fault prediction method as described in claim 2, characterized in that, The method further includes: Design a gated information transfer mechanism within a hypergraph, based on the hypergraph g. U,j The feature information X of the faulty node learned by the construction method U,base Passed to every hypergraph node X U,j In this process, the propagation is as follows: Where β is the gating weight, The hypergraph g represents the hypergraph that has undergone information propagation and fusion. U,j The faulty hypergraph network node representation, thereby realizing the hypergraph g U,j Information transfer between different interaction types between nodes in a faulty hypergraph and b β These are learnable parameters.

4. A network fault prediction device, characterized in that, The device includes: The data preprocessing module is used to preprocess network fault data, convert feedback data from faulty network nodes into sequence sample data, and generate fault time series data. The network fault data includes fault type, network node information, and time node information. Each fault time series data represents all fault types that occur at the current node over time. The fault types include, but are not limited to, port fault, network card fault, router fault, network outage, power outage, and network cable fault. The prediction module is used to construct a momentum-driven course comparison hypergraph gated convolutional network model. The network model includes an embedding module, a representation learning module for the gated hypergraph convolutional network, a course comparison learning module based on a momentum-driven optimization strategy, and a prediction module. The momentum-driven course comparison hypergraph gated convolutional network model uses the input time-series data to predict faults in the network fault data. Specifically, constructing the momentum-driven course comparison hypergraph gated convolutional network model includes: An embedding module is constructed by taking fault time-series data as input, initializing the features of fault nodes, and transforming the time-series data into embedding vectors s; then, a hypergraph structure g is constructed based on the embedded time-series data. U,j Each fault data point is represented by the following feature: G U ={g U,base ,g U,1 ...,g U,k }; g U,j ={U,ε U,j }, ε U,j Supergraph g U,j In the superedge, the supergraph g U,j All hyperedges in the network share the same set of faulty network nodes U. For a faulty network node u∈U, in g U,j Introduce a hyperedge ε U,j Connect {u|u∈U,(u,i)∈E j }, that is, in the faulty network node set U, it represents the interaction type E. j All nodes u that are directly connected to the faulty network node i; Construct a representation learning module for a hypergraph convolutional network based on a gated mechanism to obtain a node representation X based on a faulty node hypergraph; For hypergraph g U,j Perform a graph diffusion enhancement operation to obtain an enhanced diffusion-enhanced supergraph g′. U,j To extract complex high-order interaction relationships between faulty nodes, the graph diffusion operation process is formulated as follows: in It is a parameter that controls the diffusion rate of network nodes, g′ U,j It is a super map g U,j The new diffusion-enhanced hypergraph after the graph diffusion transformation; After propagation at layer l, the READOUT function is used to process the hypergraph g. U,j and enhanced diffusion hypergraph g′ U,j Node features learned in and Perform embedding operations and The final output z is obtained by using a multilayer perceptron for dimensionality reduction of the downstream task. U and z U′ The formula is as follows: Among them, z U Represents the hypergraph g U,j Embedded representation of nodes, z U′ Represents the diffusion hypergraph g′ U,j Embedded representation of nodes; Constructing a momentum-driven comparative learning module for courses: Based on the sampling strategy of course learning, in g′ U,j Find the negative sampling embedding z′ of c faulty network nodes with different difficulties. c And define a scoring function S(·), embedding different negative samples into z′ c Mapping to numerical fraction S(z′) c The difficulty is measured by setting the scoring function S(·) to sim(z). v ,z′ c To measure the difficulty of negative samples: S(z′ c )=sim(z v ,With' c )=(of v -With' c ) T Σ .-1 (With v -With' c ); Among them, z v Indicates something different from z′ c Faulty network nodes are being sampled and embedded; Obtaining a single negative sample embedding z′ throughout the entire contrastive learning process c The fraction S(z′) c After that, a step function δ(t) is used to schedule the introduction of negative samples into the training process, where the step function δ(t) specifies the size of the negative samples in each step t. The negative samples in step t consist of δ(t) of the lowest scores. T represents the total number of training steps. δ(t)=(t / T) φ ·K; Where φ is a smoothing parameter to control the pace function guiding the training speed, and during contrastive learning, according to sim(z) v ,z′ c Negative samples are sorted from lowest to highest score. Maximize the hypergraph g U,j and diffusion-enhanced hypergraph g′ U,j The mutual information between them is used to estimate the contrastive loss using noise. The final loss function of the contrastive learning module can be expressed as follows: Among them, the similarity measurement function sim(z) v ,z v ′) is used to calculate each positive sample pair {z v ,z v ′} and positive and negative sample pairs {z v ,z c The similarity of identical nodes in '}, where τ represents the temperature parameter. This indicates that the embedding of C negative samples during the training process is guided by a scoring function; In the comparative learning process, a momentum update mechanism is used to update all parameters. The momentum-driven update iteration process of the model parameters is as follows: Where 'a' is an adjustable parameter that controls the momentum of the training information extraction process. and They represent the hypergraph g respectively. U,j and enhanced diffusion hypergraph g′ U,j Learn the parameter set during iterative optimization training t rounds; Construct a prediction module for the initial embeddings of each sequence. g And the current faulty hypergraph network node representation X U,j Perform inner product calculations to compute the fraction Ω of all nodes in the hypergraph. j Then, the softmax function is used to obtain the output of predicting that the node is a possible failure in the next time period: Oh j =s g T X U,j ; in, This indicates a prediction of whether a network failure will occur in the next period of time. Then, existing nodes in the fault hypergraph network and the diffusion-enhanced fault hypergraph are used as edges for positive sample sampling, respectively. The prediction loss function is designed by maximizing the edge probability of positive samples and minimizing the edge probability of negative samples. The loss function is defined as: Where σ is the sigmoid activation function, λ is the weight parameter to balance the importance of positive and negative samples, P(u) defines the distribution of candidate fault nodes u, n is the number of negative samples, the existing edges in the multipart graph are taken as positive samples, for each positive sample edge (u,i), n negative edges with nodes u and i are randomly selected as negative sample edges; Finally, the dual-hypergraph representation learning module and the momentum-driven optimization strategy-based course comparison learning module are unified into a comprehensive learning framework, and the final prediction learning objective loss is defined as: l=l r +ζl NCE ; Here, ζ is a variable factor that controls the contrastive learning task.

5. An electronic device, characterized in that, The device includes a processor and a memory, the memory being used to store a computer program, the computer program including program instructions, and the processor being configured to invoke the program instructions to perform the method as described in any one of claims 1 to 3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, the computer program including program instructions that, when executed by a processor, cause the processor to perform the method as described in any one of claims 1 to 3.

Images