Method and device for determining root cause of network operation and maintenance fault, equipment and storage medium

Abstract

The embodiment of the invention provides a network operation and maintenance fault root cause determination method and device, equipment and a storage medium, which are applied to the field of network operation and maintenance, and the method comprises the following steps: obtaining network operation and maintenance data; based on the network operation and maintenance data and a feature analysis model, feature extraction is carried out to obtain feature vector data corresponding to the network operation and maintenance data, the feature analysis model comprises a plurality of lightweight controllers, and the lightweight controllers are used for matching a network operation and maintenance fault root cause determination task; and decoding the feature vector data, and determining a fault root cause corresponding to the network operation and maintenance data. According to the method, the relevance and dependency of the network operation and maintenance data are captured, and the accuracy of determining the network operation and maintenance fault root cause is improved.

Description

Technical Field

[0001] This application relates to the field of network operation and maintenance, and in particular to a method, apparatus, device, and storage medium for determining the root cause of network operation and maintenance failures. Background Technology

[0002] Network operation and maintenance fault analysis is a core component of ensuring stable network operation. By monitoring network operation and maintenance fault information and analyzing it using pre-trained network operation and maintenance fault models, network stability, service continuity, and operation and maintenance efficiency can be improved.

[0003] In existing technologies, the root cause of a fault is usually determined by acquiring network operation data in real time and matching the feature vectors corresponding to the real-time network operation data with the fault feature vectors in a pre-stored fault feature knowledge base.

[0004] However, because network operation data has strong correlation and dependency on faults, a single point of failure may trigger a large number of chain alarms, and multi-layered protocols are interdependent. Therefore, determining the root cause of a fault through matching has the problems of low accuracy and low efficiency. Summary of the Invention

[0005] The network operation and maintenance fault root cause determination method, apparatus, equipment and storage medium provided in the embodiments of this application are used to capture the correlation and dependency of network operation and maintenance data, and improve the accuracy of network operation and maintenance fault root cause determination.

[0006] In a first aspect, embodiments of this application provide a method for determining the root cause of network operation and maintenance failures, including:

[0007] Obtain network operation and maintenance data;

[0008] Based on the network operation and maintenance data and the feature analysis model, feature extraction is performed to obtain feature vector data corresponding to the network operation and maintenance data. The feature analysis model includes multiple lightweight controllers, which are used to match the network operation and maintenance fault root cause determination task.

[0009] The feature vector data is decoded to determine the root cause of the fault corresponding to the network operation and maintenance data.

[0010] In one possible implementation, the feature analysis model includes multiple branch networks, each branch network including a lightweight controller and a multi-layer encoder. Based on the network operation and maintenance data and the feature analysis model, feature extraction is performed to obtain feature vector data corresponding to the network operation and maintenance data, including:

[0011] The network operation and maintenance data is preprocessed and its location is embedded to obtain sequential input data;

[0012] The sequence input data is input into the first branch network of multiple branch networks, and the sequence input data is processed by the multi-layer encoder to obtain the first output data;

[0013] The lightweight controller transforms the first output data to obtain the second output data, and uses the second output data as new sequence input data to input into the next branch network. This process continues until the multiple branch networks have completed processing, and the second output data of the last branch network is used as the feature vector data.

[0014] In one possible implementation, the step of processing the sequence input data through the multi-layer encoder to obtain the first output data includes:

[0015] The sequence input data is used as input features and input into the first layer encoder to obtain the output data corresponding to the first layer encoder.

[0016] The output data corresponding to the first layer encoder is used as the input feature of the next layer encoder, and then re-inputted into the next layer encoder. The corresponding output data is then determined again until the output data corresponding to the last layer encoder is obtained.

[0017] The output data corresponding to the last encoder layer is used as the first output data.

[0018] In one possible implementation, the encoder includes a multi-head attention layer and a feedforward neural network layer. The step of inputting the sequence input data as input features into the first encoder layer to obtain the output data corresponding to the first encoder layer includes:

[0019] The sequence input data is input into the multi-head attention layer of the first encoder to obtain the multi-head attention output;

[0020] The multi-head attention output is stabilized, and the stabilization process includes residual connection and normalization.

[0021] The stabilized multi-head attention output is input into the feedforward neural network layer of the first encoder layer to obtain the output of the feedforward neural network layer.

[0022] The output of the feedforward neural network layer is stabilized, and the stabilized output of the feedforward neural network layer is used as the output data corresponding to the first layer encoder.

[0023] In one possible implementation, the sequence input data includes: a query vector, a key vector, and a value vector; the transformation processing of the first output data by the lightweight controller to obtain the second output data includes:

[0024] Based on the first output data and the first linear matrix, a first intermediate vector is determined; based on the first intermediate vector and the content query matrix, a content query vector is determined; and a rotation operation is performed on the first intermediate vector to determine the position query vector.

[0025] The content query vector and the location query vector are combined to determine the query vector;

[0026] Based on the first output data and the second linear matrix, determine the second intermediate vector, and based on the second intermediate vector and the content key matrix, determine the content key vector;

[0027] Perform a rotation operation on the first output data to determine the position key vector, and determine the key vector based on the content key vector and the position key vector;

[0028] The value vector is determined using the second intermediate vector and the third linear matrix.

[0029] In one possible implementation, the sequence input data is input into the multi-head attention layer of the first encoder to obtain the multi-head attention output, including:

[0030] Based on the number of subspaces, the query vector, key vector, and value vector are divided into multiple subspaces, wherein each subspace includes a query vector component, a key vector component, and a value vector component.

[0031] For any one of the plurality of subspaces, calculate the dot product of the query vector component and the key component of the subspace, and obtain the output result of the subspace based on the dot product of the query vector component and the key component;

[0032] Based on the output results of multiple subspaces and the fourth linear matrix, the multi-head attention output is determined, wherein the fourth linear matrix is ​​used to integrate the outputs of multiple subspaces.

[0033] In one possible implementation, the method further includes:

[0034] Based on the root cause of the fault, a corresponding work order is generated and the work order is sent to the operation and maintenance platform;

[0035] Upon receiving the maintenance failure instruction from the maintenance platform, the feature analysis model is retrained based on the work order and network maintenance failure information to update the feature analysis model.

[0036] Secondly, embodiments of this application provide a device for determining the root cause of network operation and maintenance failures, comprising:

[0037] The acquisition module is used to acquire network operation and maintenance data;

[0038] An extraction module is used to extract features based on the network operation and maintenance data and the feature analysis model to obtain feature vector data corresponding to the network operation and maintenance data. The feature analysis model includes multiple lightweight controllers, which are used to match network operation and maintenance tasks.

[0039] The determination module is used to decode the feature vector data and determine the root cause of the fault corresponding to the network operation and maintenance data.

[0040] In one possible implementation, the feature analysis model includes multiple branch networks, each branch network including a lightweight controller and a multi-layer encoder. The extraction module is also used to preprocess and embed the network operation and maintenance data to obtain sequence input data.

[0041] The sequence input data is input into the first branch network of multiple branch networks, and the sequence input data is processed by the multi-layer encoder to obtain the first output data;

[0042] The lightweight controller transforms the first output data to obtain the second output data, and uses the second output data as new sequence input data to input into the next branch network. This process continues until the multiple branch networks have completed processing, and the second output data of the last branch network is used as the feature vector data.

[0043] In one possible implementation, the extraction module is further configured to input the sequence input data as input features into the first layer encoder to obtain the output data corresponding to the first layer encoder.

[0044] The output data corresponding to the first layer encoder is used as the input feature of the next layer encoder, and then re-inputted into the next layer encoder. The corresponding output data is then determined again until the output data corresponding to the last layer encoder is obtained.

[0045] The output data corresponding to the last encoder layer is used as the first output data.

[0046] In one possible implementation, the encoder includes a multi-head attention layer and a feedforward neural network layer. The extraction module is also used to input the sequence input data into the multi-head attention layer of the first encoder to obtain a multi-head attention output.

[0047] The multi-head attention output is stabilized, and the stabilization process includes residual connection and normalization.

[0048] The stabilized multi-head attention output is input into the feedforward neural network layer of the first encoder layer to obtain the output of the feedforward neural network layer.

[0049] The output of the feedforward neural network layer is stabilized, and the stabilized output of the feedforward neural network layer is used as the output data corresponding to the first layer encoder.

[0050] In one possible implementation, the sequence input data includes: a query vector, a key vector, and a value vector; the extraction module is further configured to determine a first intermediate vector based on the first output data and the first linear matrix, determine a content query vector based on the first intermediate vector and the content query matrix, and perform a rotation operation on the first intermediate vector to determine a position query vector;

[0051] The content query vector and the location query vector are combined to determine the query vector;

[0052] Based on the first output data and the second linear matrix, determine the second intermediate vector, and based on the second intermediate vector and the content key matrix, determine the content key vector;

[0053] Perform a rotation operation on the first output data to determine the position key vector, and determine the key vector based on the content key vector and the position key vector;

[0054] The value vector is determined using the second intermediate vector and the third linear matrix.

[0055] In one possible implementation, the extraction module is further configured to divide the query vector, key vector, and value vector into multiple subspaces according to the number of subspaces, wherein each subspace includes a query vector component, a key vector component, and a value vector component.

[0056] For any one of the plurality of subspaces, calculate the dot product of the query vector component and the key component of the subspace, and obtain the output result of the subspace based on the dot product of the query vector component and the key component;

[0057] Based on the output results of multiple subspaces and the fourth linear matrix, the multi-head attention output is determined, wherein the fourth linear matrix is ​​used to integrate the outputs of multiple subspaces.

[0058] In one possible implementation, the device further includes: a training module, used to generate a corresponding work order based on the root cause of the fault, and send the work order to the operation and maintenance platform; after receiving the operation and maintenance failure instruction from the operation and maintenance platform, the feature analysis model is retrained based on the work order and network operation and maintenance fault information to update the feature analysis model.

[0059] Thirdly, embodiments of this application provide a device for determining the root cause of network operation and maintenance failures, including: a memory and a processor;

[0060] The memory stores computer-executed instructions;

[0061] The processor executes computer execution instructions stored in the memory, causing the processor to perform the first aspect and / or various possible implementations of the first aspect as described above.

[0062] Fourthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the first aspect and / or various possible implementations of the first aspect.

[0063] Fifthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the first aspect and / or various possible implementations of the first aspect.

[0064] The network operation and maintenance fault root cause determination method, apparatus, device, and storage medium provided in this application embodiment obtain network operation and maintenance data such as user address, network status, fault time, and GPU location to provide evidence for determining the fault root cause. Based on the network operation and maintenance data and feature analysis model, feature extraction is performed to obtain feature vector data corresponding to the network operation and maintenance data, capturing the correlation and dependency of the network operation and maintenance data, and improving the accuracy and reliability of the determined fault root cause. The feature vector data is decoded to determine the fault root cause corresponding to the network operation and maintenance data, transforming the abstract fault feature vector into a natural language root cause description, realizing accurate analysis of the fault root cause of network fault information, assisting operation and maintenance personnel in handling more fault information every day, freeing up manpower and productivity, applicable to 24-hour network operation and maintenance assistance in handling operation and maintenance faults, and improving user experience. Attached Figure Description

[0065] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0066] Figure 1 A flowchart illustrating a method for determining the root cause of network operation and maintenance failures provided in this application embodiment. Figure 1 ;

[0067] Figure 2 A flowchart illustrating a method for determining the root cause of network operation and maintenance failures provided in this application embodiment. Figure 2 ;

[0068] Figure 3This is a schematic diagram of a multi-head attention mechanism structure provided in an embodiment of this application;

[0069] Figure 4 A schematic diagram of a device for determining the root cause of network operation and maintenance failures provided in an embodiment of this application;

[0070] Figure 5 This is a schematic diagram of a device for determining the root cause of network operation and maintenance failures, provided in an embodiment of this application.

[0071] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0072] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0073] "Multiple" refers to two or more, and other quantifiers are similar. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following objects have an "or" relationship.

[0074] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that embodiments of the invention described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms “comprising” and “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, products, or apparatus.

[0075] It should be noted that, in the embodiments of this application, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design scheme described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design schemes. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0076] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0077] First, let me explain the terms used in this application:

[0078] Rotary Position Embedding (RoPE) is a positional encoding technique used to assign positional information to each element in an input sequence, helping the model capture the relative positional relationships within the sequence.

[0079] Network operation and maintenance fault analysis is a core component of ensuring stable network operation. By monitoring network operation and maintenance fault information and analyzing it using pre-trained network operation and maintenance fault models, network stability, service continuity, and operation and maintenance efficiency can be improved.

[0080] In existing technologies, the root cause of a fault is usually determined by acquiring network operation data in real time and matching the feature vectors corresponding to the real-time network operation data with the fault feature vectors in a pre-stored fault feature knowledge base.

[0081] However, because network operation data has strong correlation and dependency on faults, a single point of failure may trigger a large number of chain alarms, and multi-layered protocols are interdependent. Therefore, determining the root cause of a fault through matching has the problems of low accuracy and low efficiency.

[0082] The method for determining the root cause of network operation and maintenance failures provided in this application extracts features from network operation and maintenance data using a feature analysis model. This model includes multiple lightweight controllers, which are used to match the task of determining the root cause of network operation and maintenance failures and decode the extracted feature vector data to obtain the root cause of the failure. This method solves the problems of low accuracy and low efficiency in matching and determining the root cause of failures in existing technologies. By capturing the correlation and dependency of network operation and maintenance data, it improves the accuracy of determining the root cause of network operation and maintenance failures.

[0083] The technical solution of this application and how the technical solution of this application solves the above-mentioned technical problems are described in detail below with specific embodiments. These specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of this application will now be described with reference to the accompanying drawings.

[0084] Figure 1 A flowchart illustrating a method for determining the root cause of network operation and maintenance failures provided in this application embodiment. Figure 1 ,like Figure 1 As shown, the method includes:

[0085] S101. Obtain network operation and maintenance data;

[0086] Network operation and maintenance data can include, for example, user addresses, network status, failure times, GPU locations, etc.

[0087] Collect network operation and maintenance data recorded within the network operation and maintenance system. The network operation and maintenance data can be in CSV format and contains a large number of fields, such as user address, network status, fault cause, fault time, GPU location, etc.

[0088] S102. Based on the network operation and maintenance data and feature analysis model, feature extraction is performed to obtain feature vector data corresponding to the network operation and maintenance data;

[0089] The feature analysis model includes multiple lightweight controllers, which are used to match the task of determining the root cause of network operation and maintenance failures. Feature vector data refers to the data that extracts key fault features from network operation and maintenance data.

[0090] Feature analysis models are derived by fine-tuning large models. Large models are typically pre-trained on a very large dataset. This dataset covers a wide range of topics and tasks, aiming to allow the model to learn general data representations. After pre-training, these models are able to extract high-level, abstract feature representations from the input data. These features are useful for many downstream tasks because they capture the basic structure and patterns of the input data, allowing for adjustments to some or all of the model's parameters as needed when the model encounters significant differences from the training corpus during inference.

[0091] Feature analysis models enhance data fit by adapting lightweight controllers and model structures, such as adding adapters to all fully connected layers. This approach is more effective for feedforward training when fine-tuning with smaller datasets and can reduce resource utilization.

[0092] The network operation and maintenance data is subjected to feature extraction by a feature analysis model to capture the correlation and dependency of the network operation and maintenance data, and to obtain the feature vector data corresponding to the network operation and maintenance data.

[0093] S103. Decode the feature vector data to determine the root cause of the fault corresponding to the network operation and maintenance data.

[0094] The feature vector data is decoded to transform the abstract fault feature vector into a root cause description in natural language, that is, to determine the root cause of the fault corresponding to the network operation and maintenance data.

[0095] In one possible implementation, the method further includes: generating a corresponding work order based on the root cause of the fault and sending the work order to the operation and maintenance platform; after receiving the operation and maintenance failure instruction from the operation and maintenance platform, retraining the feature analysis model based on the work order and network operation and maintenance fault information to update the feature analysis model.

[0096] The work order is used to instruct maintenance personnel to perform network maintenance based on the information in the work order. This information may include, for example, the root cause of the fault, the fault area, and the solution. The maintenance failure instruction is used to trigger the training of the feature analysis model, i.e., fine-tuning of the feature analysis model.

[0097] Based on the root cause of the fault, the fault area and solution are determined, and a corresponding work order is generated and sent to the operation and maintenance platform. Operation and maintenance personnel then perform network maintenance according to the work order. Feedback is provided through the operation and maintenance platform after a network maintenance failure. When maintenance fails, it indicates that the current feature analysis model cannot meet the current maintenance requirements; therefore, the feature analysis model needs to be trained to adapt to these requirements. The lightweight controller in the feature analysis model includes model parameters. During training, the hyperparameters in the lightweight controller are adjusted, including the learning rate and momentum. The learning rate can also be dynamically adjusted based on changes in loss and gradient distribution to improve the model's convergence speed and avoid getting trapped in local optima. This improves the model's convergence speed, avoids getting trapped in local optima, and enhances the accuracy of the network operation and maintenance fault detection model.

[0098] For example: If the loss does not change significantly over a period of time during training, the learning rate is reduced.

[0099] After an operational failure, the feature analysis model is retrained to achieve supervised self-learning and version management, which improves the overall performance of the pre-trained model, enhances the accuracy of the network operation and maintenance fault detection model, improves the accuracy of feature extraction by the feature analysis model, and improves the accuracy of root cause analysis.

[0100] The network operation and maintenance fault detection model training method provided in this application provides evidence for determining the root cause of faults by acquiring network operation and maintenance data such as user address, network status, fault time, and GPU location. Based on the network operation and maintenance data and feature analysis model, feature extraction is performed to obtain feature vector data corresponding to the network operation and maintenance data, capturing the correlation and dependency of the network operation and maintenance data, and improving the accuracy and reliability of the determined root cause of faults. The feature vector data is decoded to determine the root cause of faults corresponding to the network operation and maintenance data, transforming the abstract fault feature vector into a root cause description in natural language, realizing accurate analysis of the root cause of network fault information, assisting operation and maintenance personnel in handling more fault information every day, freeing up manpower and productivity, applicable to 24-hour network operation and maintenance assistance in handling operation and maintenance faults, and improving user experience.

[0101] Figure 2 A flowchart illustrating a network operation and maintenance fault detection model training method provided in this application embodiment. Figure 2 In this embodiment, the feature analysis model includes multiple branch networks, each branch network including a lightweight controller and a multi-layer encoder. Figure 1 Based on the embodiments, a detailed explanation is provided on the feature extraction performed on the network operation and maintenance data and the feature analysis model to obtain the feature vector data corresponding to the network operation and maintenance data, such as... Figure 2 As shown, the method includes:

[0102] S201. Preprocess and embed the location of the network operation and maintenance data to obtain the sequence input data;

[0103] Preprocessing includes data cleaning and missing value imputation of network operation and maintenance fault data. Location embedding processing refers to embedding locations to capture location information in the data sequence.

[0104] The network operations and maintenance (O&M) data undergoes data cleaning and missing value imputation preprocessing. The cleaned data is then converted into a streaming format and transformed into a vector format. Next, location embedding is applied to the vector-formatted O&M data to capture positional information within the data sequence. This helps the model distinguish between simultaneous independent events and capture long-distance dependencies.

[0105] S202. Input the sequence input data into the first branch network of multiple branch networks, and use the sequence input data as input features to input into the first layer encoder to obtain the output data corresponding to the first layer encoder.

[0106] The number of encoders in each branch network can be the same or different, and this application does not impose any restrictions on this.

[0107] S203. Use the output data corresponding to the first layer encoder as the input feature of the next layer encoder, re-input it into the next layer encoder, and redetermine the corresponding output data until the output data corresponding to the last layer encoder is obtained.

[0108] In one possible implementation, the encoder includes a multi-head attention layer and a feedforward neural network layer. The following detailed explanation is provided regarding the process of taking sequential input data as input features and feeding it into the first encoder layer to obtain the corresponding output data:

[0109] The sequence input data is input into the multi-head attention layer of the first encoder to obtain the multi-head attention output; the multi-head attention output is stabilized; the stabilized multi-head attention output is input into the feedforward neural network layer of the first encoder to obtain the feedforward neural network layer output; the feedforward neural network layer output is stabilized, and the stabilized feedforward neural network layer output is used as the corresponding output data of the first encoder.

[0110] The stabilization process includes residual connection and normalization.

[0111] The sequence input data is fed into the multi-head attention layer of the first encoder to obtain the multi-head attention output. Residual connections are made to this output to preserve input information and improve gradient propagation. The output is then normalized to stabilize it. The stabilized multi-head attention output is then fed into the feedforward neural network layer of the first encoder to obtain its output, enhancing the model's expressive power. The feedforward neural network output is stabilized again, and residual connections are made to preserve input information and improve gradient propagation. This output is then normalized to stabilize it, and the stabilized output is used as the input features for the next encoder layer. In essence, the output data from the first encoder layer enters the second encoder layer for further data processing, until the final encoder layer is reached.

[0112] S204. Use the output data corresponding to the last encoder layer as the first output data;

[0113] S205. Based on the first output data and the first linear matrix, determine the first intermediate vector; based on the first intermediate vector and the content query matrix, determine the content query vector; and perform a rotation operation on the first intermediate vector to determine the position query vector.

[0114] A query can be understood as a question expressing the information you want to know. The first intermediate vector is calculated using the following formula:

[0115]

[0116] in, The first intermediate vector, Let be the first linear matrix. This is the first output data.

[0117] Based on the first intermediate vector and the content query matrix, determine the content query vector:

[0118]

[0119] in, For content query vectors; This is a content query matrix.

[0120] Perform a rotation operation on the first intermediate vector to determine the position query vector:

[0121] The location query vector is determined using the RoPE method:

[0122]

[0123] in, This is a location query vector.

[0124] S206. Merge the content query vector and the location query vector to determine the query vector;

[0125] The expression is as follows:

[0126]

[0127] in, This is the query vector.

[0128] S207. Determine the second intermediate vector based on the first output data and the second linear matrix, and determine the content key vector based on the second intermediate vector and the content key matrix.

[0129] The merging of key vectors and value vectors can be understood as the answer to the problem, and the key vector is determined using the following formula.

[0130] Based on the first output data and the second linear matrix, determine the second intermediate vector:

[0131]

[0132] in, This is the second intermediate vector; It is the second linear matrix.

[0133] Based on the second intermediate vector and the content key matrix, determine the content key vector:

[0134]

[0135] in, For content key matrix; This is a content key vector.

[0136] S208. Perform a rotation operation on the first output data, determine the position key vector, and determine the key vector based on the content key vector and the position key vector;

[0137] Perform a rotation operation on the first output data to determine the position key vector:

[0138]

[0139] in, It is the fifth linear matrix; This is the position key vector.

[0140] Based on the content key vector and the position key vector, determine the key vector:

[0141]

[0142] in, This is the key vector.

[0143] S209. Determine the value vector using the second intermediate vector and the third linear matrix.

[0144] The expression is as follows:

[0145]

[0146] in, It is the third linear matrix; It is a value vector.

[0147] In one possible implementation, the process of inputting the sequence input data into the multi-head attention layer of the first encoder to obtain the multi-head attention output is described in detail, including:

[0148] Based on the number of subspaces, the query vector, key vector, and value vector are divided into multiple subspaces. For any one of the multiple subspaces, the dot product of the query vector component and the key component of the subspace is calculated, and the output result of the subspace is obtained based on the dot product of the query vector component and the key component. Based on the output results of the multiple subspaces and the fourth linear matrix, the multi-head attention output is determined.

[0149] Each subspace includes a query vector component, a key vector component, and a value vector component; the fourth linear matrix is ​​used to integrate the outputs of multiple subspaces.

[0150] The query vector, key vector, and value vector are divided into multiple subspaces on an equal basis according to the number of subspaces. Figure 3This is a schematic diagram of a multi-head attention mechanism structure provided in an embodiment of this application. Figure 3 As shown, it includes 8 subspaces (8 heads). The query vector, key vector, and value vector are projected into multiple subspaces, and each subspace includes query vector components, key vector components, and value vector components.

[0151] For any one of the multiple subspaces, calculate the dot product of the query vector component and the key component of the subspace, scale the dot product, and then determine the attention weights using the softmax function.

[0152] The attention weights are calculated using the following formula:

[0153]

[0154] in, Indicates the first The query component of each subspace; Indicates the first The key components of each subspace; Indicates the first Value components of each subspace; express Dimension size; for The transpose of .

[0155] The output of the subspace is obtained based on the attention weights, and the first multi-head attention output is determined based on the outputs of multiple subspaces and the first linear matrix.

[0156] The first linear matrix is ​​used to integrate the outputs of multiple subspaces.

[0157] The first multi-head attention output is calculated using the following formula:

[0158]

[0159] in, For multi-head attention output; For the first The output of each subspace, It is the fourth linear matrix.

[0160] S209. The second output data is used as new sequence input data and input into the next branch network until the multiple branch networks are processed. The second output data of the last branch network is used as the feature vector data.

[0161] In one possible implementation, the specific process of determining the query vector based on input features is described in detail, including:

[0162] The network operation and maintenance fault detection model training method provided in this application preprocesses and embeds location data into network operation and maintenance data to obtain sequential input data, which helps the model distinguish between simultaneous independent events and capture long-distance dependencies. The sequential input data is input into the first branch network of multiple branch networks. This sequential input data is then used as input features and input into the first encoder layer to obtain the output data corresponding to the first encoder layer. The output data of the first encoder layer is then used as input features for the next encoder layer, and the corresponding output data is re-determined until the output data of the last encoder layer is obtained. The output data of the last encoder layer is then used as the first output data, which integrates abstract features and data. Based on the first output data, query vectors, key vectors, and value vectors are determined, enabling the model to better capture contextual dependencies in the text. Furthermore, RoPE encoding introduces location information, enhancing the model's ability to understand long texts. The second output data is then used as new sequential input data and input into the next branch network. This process continues until all multiple branch networks have completed processing. Finally, the second output data of the last branch network is used as the feature vector data, thus achieving feature extraction from the network operation and maintenance data.

[0163] Figure 4 This application provides a schematic diagram of the structure of a device for determining the root cause of network operation and maintenance failures, as shown in the embodiments of this application. Figure 4 As shown, the network operation and maintenance fault root cause determination device 40 provided in this embodiment includes:

[0164] Module 401 is used to acquire network operation and maintenance data;

[0165] The extraction module 402 is used to extract features based on the network operation and maintenance data and the feature analysis model to obtain feature vector data corresponding to the network operation and maintenance data. The feature analysis model includes multiple lightweight controllers, which are used to match network operation and maintenance tasks.

[0166] The determination module 403 is used to decode the feature vector data and determine the root cause of the fault corresponding to the network operation and maintenance data.

[0167] In one possible implementation, the feature analysis model includes multiple branch networks, each branch network including a lightweight controller and a multi-layer encoder. The extraction module 402 is also used to preprocess and position embedding the network operation and maintenance data to obtain sequence input data.

[0168] The sequence input data is input into the first branch network of multiple branch networks, and the sequence input data is processed by the multi-layer encoder to obtain the first output data;

[0169] The lightweight controller transforms the first output data to obtain the second output data, and uses the second output data as new sequence input data to input into the next branch network. This process continues until the multiple branch networks have completed processing, and the second output data of the last branch network is used as the feature vector data.

[0170] In one possible implementation, the extraction module 402 is further configured to input the sequence input data as input features into the first layer encoder to obtain the output data corresponding to the first layer encoder.

[0171] The output data corresponding to the first layer encoder is used as the input feature of the next layer encoder, and then re-inputted into the next layer encoder. The corresponding output data is then determined again until the output data corresponding to the last layer encoder is obtained.

[0172] The output data corresponding to the last encoder layer is used as the first output data.

[0173] In one possible implementation, the encoder includes a multi-head attention layer and a feedforward neural network layer. The extraction module 402 is also used to input the sequence input data into the multi-head attention layer of the first encoder to obtain the multi-head attention output.

[0174] The multi-head attention output is stabilized, and the stabilization process includes residual connection and normalization.

[0175] The stabilized multi-head attention output is input into the feedforward neural network layer of the first encoder layer to obtain the output of the feedforward neural network layer.

[0176] The output of the feedforward neural network layer is stabilized, and the stabilized output of the feedforward neural network layer is used as the output data corresponding to the first layer encoder.

[0177] In one possible implementation, the sequence input data includes: a query vector, a key vector, and a value vector; the extraction module 402 is further configured to determine a first intermediate vector based on the first output data and the first linear matrix, determine a content query vector based on the first intermediate vector and the content query matrix, and perform a rotation operation on the first intermediate vector to determine a position query vector;

[0178] The content query vector and the location query vector are combined to determine the query vector;

[0179] Based on the first output data and the second linear matrix, determine the second intermediate vector, and based on the second intermediate vector and the content key matrix, determine the content key vector;

[0180] Perform a rotation operation on the first output data to determine the position key vector, and determine the key vector based on the content key vector and the position key vector;

[0181] The value vector is determined using the second intermediate vector and the third linear matrix.

[0182] In one possible implementation, the extraction module 402 is further configured to divide the query vector, key vector and value vector into multiple subspaces according to the number of subspaces, wherein each subspace includes a query vector component, a key vector component and a value vector component.

[0183] For any one of the plurality of subspaces, calculate the dot product of the query vector component and the key component of the subspace, and obtain the output result of the subspace based on the dot product of the query vector component and the key component;

[0184] Based on the output results of multiple subspaces and the fourth linear matrix, the multi-head attention output is determined, wherein the fourth linear matrix is ​​used to integrate the outputs of multiple subspaces.

[0185] In one possible implementation, the device further includes: a training module, used to generate a corresponding work order based on the root cause of the fault, and send the work order to the operation and maintenance platform; after receiving the operation and maintenance failure instruction from the operation and maintenance platform, the feature analysis model is retrained based on the work order and network operation and maintenance fault information to update the feature analysis model.

[0186] The network operation and maintenance fault root cause determination device provided in this embodiment can execute the training method provided in the above training method embodiment. Its implementation principle and technical effect are similar, and will not be described in detail here.

[0187] Figure 5 This is a schematic diagram of a device for determining the root cause of network operation and maintenance failures, provided as an embodiment of this application. Figure 5 As shown, the network operation and maintenance fault detection model training device 50 provided in this embodiment includes at least one processor 501 and a memory 502. Optionally, the device 50 also includes a communication component 503. The processor 501, memory 502, and communication component 503 are connected via a bus 504.

[0188] In a specific implementation, at least one processor 501 executes computer execution instructions stored in memory 502, causing at least one processor 501 to perform the above-described method.

[0189] The specific implementation process of processor 501 can be found in the above training method embodiment, and its implementation principle and technical effect are similar, so it will not be repeated here.

[0190] In the above embodiments, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly implemented by a hardware processor, or implemented by a combination of hardware and software modules within the processor.

[0191] The memory may include random access memory (RAM) and may also include non-volatile memory (NVM), such as at least one disk storage device.

[0192] The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, the buses shown in the accompanying drawings are not limited to a single bus or a single type of bus.

[0193] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described method.

[0194] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the above-described method.

[0195] The aforementioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The readable storage medium can be any available medium accessible to a general-purpose or special-purpose computer.

[0196] An exemplary readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and the readable storage medium can exist as discrete components in the device.

[0197] The division of units is merely a logical functional division; 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 indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0198] 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.

[0199] In addition, the functional units in the various embodiments of the present invention 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.

[0200] If a function is implemented as 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 this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0201] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0202] Finally, it should be noted that other embodiments of the invention will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not disclosed herein, and is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of the invention is limited only by the appended claims.

Claims

1. A method for determining the root cause of network operation and maintenance failures, characterized in that, include: Obtain network operation and maintenance data; Based on the network operation and maintenance data and the feature analysis model, feature extraction is performed to obtain feature vector data corresponding to the network operation and maintenance data. The feature analysis model includes multiple lightweight controllers, which are used to match the network operation and maintenance fault root cause determination task. The feature vector data is decoded to determine the root cause of the fault corresponding to the network operation and maintenance data.

2. The method according to claim 1, characterized in that, The feature analysis model includes multiple branch networks, each branch network including a lightweight controller and a multi-layer encoder. Based on the network operation and maintenance data and the feature analysis model, feature extraction is performed to obtain feature vector data corresponding to the network operation and maintenance data, including: The network operation and maintenance data is preprocessed and its location is embedded to obtain sequential input data; The sequence input data is input into the first branch network of multiple branch networks, and the sequence input data is processed by the multi-layer encoder to obtain the first output data; The lightweight controller transforms the first output data to obtain the second output data, and uses the second output data as new sequence input data to input into the next branch network. This process continues until the multiple branch networks have completed processing, and the second output data of the last branch network is used as the feature vector data.

3. The method according to claim 2, characterized in that, The step of processing the sequence input data through the multi-layer encoder to obtain the first output data includes: The sequence input data is used as input features and input into the first layer encoder to obtain the output data corresponding to the first layer encoder. The output data corresponding to the first layer encoder is used as the input feature of the next layer encoder, and then re-inputted into the next layer encoder. The corresponding output data is then determined again until the output data corresponding to the last layer encoder is obtained. The output data corresponding to the last encoder layer is used as the first output data.

4. The method according to claim 3, characterized in that, The encoder includes a multi-head attention layer and a feedforward neural network layer. The step of taking the sequence input data as input features and inputting it into the first encoder layer to obtain the output data corresponding to the first encoder layer includes: The sequence input data is input into the multi-head attention layer of the first encoder to obtain the multi-head attention output; The multi-head attention output is stabilized, and the stabilization process includes residual connection and normalization. The stabilized multi-head attention output is input into the feedforward neural network layer of the first encoder layer to obtain the output of the feedforward neural network layer. The output of the feedforward neural network layer is stabilized, and the stabilized output of the feedforward neural network layer is used as the output data corresponding to the first layer encoder.

5. The method according to claim 4, characterized in that, The sequence input data includes: a query vector, a key vector, and a value vector; the transformation processing of the first output data by the lightweight controller to obtain the second output data includes: Based on the first output data and the first linear matrix, a first intermediate vector is determined; based on the first intermediate vector and the content query matrix, a content query vector is determined; and a rotation operation is performed on the first intermediate vector to determine the position query vector. The content query vector and the location query vector are combined to determine the query vector; Based on the first output data and the second linear matrix, determine the second intermediate vector, and based on the second intermediate vector and the content key matrix, determine the content key vector; Perform a rotation operation on the first output data to determine the position key vector, and determine the key vector based on the content key vector and the position key vector; The value vector is determined using the second intermediate vector and the third linear matrix.

6. The method according to claim 5, characterized in that, The step of inputting the sequence input data into the multi-head attention layer of the first encoder to obtain the multi-head attention output includes: Based on the number of subspaces, the query vector, key vector, and value vector are divided into multiple subspaces, wherein each subspace includes a query vector component, a key vector component, and a value vector component. For any one of the plurality of subspaces, calculate the dot product of the query vector component and the key component of the subspace, and obtain the output result of the subspace based on the dot product of the query vector component and the key component; Based on the output results of multiple subspaces and the fourth linear matrix, the multi-head attention output is determined, wherein the fourth linear matrix is ​​used to integrate the outputs of multiple subspaces.

7. The method according to claim 1, characterized in that, The method further includes: Based on the root cause of the fault, a corresponding work order is generated and the work order is sent to the operation and maintenance platform; Upon receiving the maintenance failure instruction from the maintenance platform, the feature analysis model is retrained based on the work order and network maintenance failure information to update the feature analysis model.

8. A device for determining the root cause of network operation and maintenance failures, characterized in that, include: The acquisition module is used to acquire network operation and maintenance data; An extraction module is used to extract features based on the network operation and maintenance data and the feature analysis model to obtain feature vector data corresponding to the network operation and maintenance data. The feature analysis model includes multiple lightweight controllers, which are used to match network operation and maintenance tasks. The determination module is used to decode the feature vector data and determine the root cause of the fault corresponding to the network operation and maintenance data.

9. A device for determining the root cause of network operation and maintenance failures, characterized in that, include: Memory, processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-7.

11. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.

Images