Fault detection method and device, computer storable medium
By using a combination of convolutional encoders and sequence models in virtualized servers, the challenge of fault detection in complex structures is solved, enabling fast and accurate fault location and detection.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA TELECOM CORP LTD
- Filing Date
- 2022-11-07
- Publication Date
- 2026-06-12
AI Technical Summary
In virtualized servers, the complex internal structure makes fault detection difficult to debug and test quickly, and existing technologies cannot quickly and accurately determine the time and type of fault occurrence.
By obtaining the location identifier and data category of log data, a convolutional encoder is used to generate convolutional codes, which are then combined with a sequence model to determine fault location information. A combination of convolutional encoder and sequence model is used for fault detection.
It enables rapid and accurate fault detection, improving the data processing efficiency and detection performance for fault location.
Smart Images
Figure CN115543676B_ABST
Abstract
Description
Technical Field
[0001] This disclosure relates to the fields of big data or artificial intelligence technology, and in particular to fault detection methods and devices, and computer-storable media. Background Technology
[0002] The development of new technologies such as cloud computing, blockchain, and 6G mobile communication has driven massive new demands for servers, which are evolving towards centralized and integrated management. Virtualized server resources bring more efficient resource utilization, but also present significant challenges for fault detection. Overly complex internal structures make it difficult for developers to quickly debug and test. Therefore, how to quickly determine the time and type of faults through intelligent technologies is a highly significant research direction for the future. Summary of the Invention
[0003] This disclosure presents a solution that enables rapid and accurate fault detection.
[0004] According to a first aspect of this disclosure, a fault detection method is provided, comprising: acquiring multiple log data entries, wherein each log data entry has a location identifier and a data category, the location identifier of the multiple log data entries reflecting the temporal relationship of the multiple log data entries; performing convolutional code encoding operation using a convolutional encoder corresponding to the data category corresponding to each log data entry, based on the location identifier of each log data entry, to obtain a convolutional code for each log data entry; and determining fault location information using a sequence model based on the convolutional codes of the multiple log data entries.
[0005] In some embodiments, performing convolutional code encoding operations based on the location identifier of each log data entry and using a convolutional encoder corresponding to the data category of each log data entry includes: encoding the location identifier of each log data entry to obtain the location code; determining the corresponding convolutional encoder based on the data category of each log data entry; and generating the convolutional code using the determined convolutional encoder based on the location code of each log data entry.
[0006] In some embodiments, determining fault location information based on the convolutional codes of multiple log data using a sequence model includes: grouping the multiple log data to obtain multiple sets of log data; merging the convolutional codes corresponding to each set of log data to obtain a word embedding matrix corresponding to each set of log data; and determining fault location information based on the word embedding matrices corresponding to the multiple sets of log data using the sequence model.
[0007] In some embodiments, the sequence model includes an encoder, a decoder, and a softmax layer. Determining fault location information using the sequence model based on word embedding matrices corresponding to multiple sets of log data includes: determining the hidden layer semantic vector of each word embedding matrix using the encoder; determining the weight of the hidden layer semantic vector of each word embedding matrix using an attention mechanism, wherein the weight represents the importance of the data category to fault detection; and determining the fault location information using the decoder and the softmax layer based on the hidden layer semantic vectors of the multiple word embedding matrices and their corresponding weights.
[0008] In some embodiments, determining the weights of the hidden layer semantic vectors of each word embedding matrix using an attention mechanism includes: calculating the information coefficients of the data categories for fault detection for each group of log data using the feature selection library of sklearn, wherein the information coefficients characterize the importance of the data categories of the corresponding group of log data for fault detection; and determining the weights of the hidden layer semantic vectors of each word embedding matrix based on the information coefficients corresponding to each word embedding matrix.
[0009] In some embodiments, determining fault location information based on the hidden layer semantic vectors of the plurality of word embedding matrices and their corresponding weights, using the decoder and the softmax layer, includes: multiplying the hidden layer semantic vector corresponding to each word embedding matrix by the weights; and inputting the product obtained by multiplication sequentially into the decoder and the softmax layer to obtain fault location information.
[0010] In some embodiments, the fault detection method further includes: determining the location identifier of each log data based on the timestamp information of multiple log data.
[0011] According to a second aspect of this disclosure, a fault detection device is provided, comprising: an acquisition module configured to acquire multiple log data entries, wherein each log data entry has a location identifier and a data category, and the location identifier of the multiple log data entries reflects the temporal relationship of the multiple log data entries; an execution module configured to perform a convolutional code encoding operation based on the location identifier of each log data entry, using a convolutional encoder corresponding to the data category of each log data entry, to obtain a convolutional code for each log data entry; and a determination module configured to determine fault location information based on the convolutional codes of the multiple log data entries using a sequence model.
[0012] According to a third aspect of this disclosure, a fault detection apparatus is provided, comprising: a memory; and a processor coupled to the memory, the processor being configured to execute the fault detection method described in any of the above embodiments based on instructions stored in the memory.
[0013] According to a fourth aspect of this disclosure, a computer-storeable medium is provided that stores computer program instructions thereon, which, when executed by a processor, implement the fault detection method described in any of the above embodiments.
[0014] In the above embodiments, rapid and accurate fault detection can be achieved. Attached Figure Description
[0015] The accompanying drawings, which form part of this specification, illustrate embodiments of this disclosure and, together with the specification, serve to explain the principles of this disclosure.
[0016] This disclosure will become clearer with reference to the accompanying drawings and the following detailed description, wherein:
[0017] Figure 1 This is a flowchart illustrating a fault detection method according to some embodiments of the present disclosure;
[0018] Figure 2 This is a schematic diagram illustrating the process of obtaining a word embedding matrix according to some embodiments of the present disclosure;
[0019] Figure 3 This is a schematic diagram illustrating the determination of fault location information according to some embodiments of the present disclosure;
[0020] Figure 4 This is a block diagram illustrating a fault detection apparatus according to some embodiments of the present disclosure;
[0021] Figure 5 This is a block diagram illustrating a fault detection apparatus according to other embodiments of the present disclosure;
[0022] Figure 6 This is a block diagram illustrating a computer system for implementing some embodiments of the present disclosure. Detailed Implementation
[0023] Various exemplary embodiments of the present disclosure will now be described in detail with reference to the accompanying drawings. It should be noted that, unless otherwise specifically stated, the relative arrangement, numerical expressions, and values of the components and steps set forth in these embodiments do not limit the scope of the present disclosure.
[0024] At the same time, it should be understood that, for ease of description, the dimensions of the various parts shown in the accompanying drawings are not drawn according to actual scale.
[0025] The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this disclosure or its application or use.
[0026] Techniques, methods, and equipment known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and equipment should be considered part of the specification.
[0027] In all examples shown and discussed herein, any specific values should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.
[0028] It should be noted that similar labels and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be discussed further in subsequent figures.
[0029] Figure 1 This is a flowchart illustrating a fault detection method according to some embodiments of the present disclosure.
[0030] like Figure 1 As shown, the fault detection method includes steps S110-S130.
[0031] In step S110, multiple log data entries are acquired. Each log data entry has a location identifier and a data category. The location identifiers of the multiple log data entries reflect their temporal sequence. In some embodiments, the location identifier of each log data entry can be determined based on the timestamp information of the multiple log data entries.
[0032] In some embodiments, multiple log entries are generated by the same type of calling program. For example, multiple log entries may be port call data, and the data categories or log types of port call data include system information acquisition, network communication, file operations, service behavior, registry behavior, and operation process behavior. Log data is text data generated by print output code embedded in a program; it records key information such as variables and execution status during program runtime. Log data is a series of time-based server information.
[0033] In step S120, based on the location identifier of each log data, a convolutional code encoding operation is performed using a convolutional encoder corresponding to the data category of each log data to obtain the convolutional code of each log data.
[0034] Typically, vectorized text information can be obtained directly from existing corpora or trained on a network as unknown parameters. This disclosure is applied to server log data scenarios where the associated corpora are relatively small, requiring custom model training. Without prior information, one-hot encoding is often used. While simple to implement, this encoding is prone to the curse of dimensionality and fails to capture the correlation and sequence information between different types of text information.
[0035] Convolutional codes are widely used channel coding in the communications industry. As a non-linear block code, they combine reliability and efficiency in encoding. Based on this, this disclosure innovatively proposes a method that combines convolution and word embedding: the information bits of the convolutional code record the position information of the text as input parameters to determine the convolutional code, and different binary operation convolutional kernels (convolutional encoders) are constructed according to the input data category for encoding. In subsequent processes, convolutional codes of log data of different data categories are output in turn.
[0036] Different Hamming distances between convolutional codes represent the differences between different codes, and the greater the difference in convolutional kernels, the greater the difference in the output after encoding the same sequence, i.e., a larger Hamming distance. Based on this, this disclosure innovatively designs convolutional kernel values for different classes based on Hamming distance. If log data belongs to the same class, then the convolutional kernels are consistent. If log data belongs to different classes, then the convolutional kernels are different, and the greater the class difference, the greater the difference in convolutional kernels. This is the basic innovative idea of this disclosure for word embedding encoding based on convolutional codes. In the above embodiments, based on channel coding in communication, the convolutional code of each log data is determined, making text data such as log data numerical or vectorized, which is convenient for neural networks such as sequence models to understand, process, and learn features. Because convolutional codes encode information bits based on positional information, neural networks can capture sequence information more quickly, reducing the difficulty of network training convergence.
[0037] In some embodiments, step S120 can be implemented in the following manner.
[0038] First, the location identifier of each log data entry is encoded to obtain a location code. A timestamp is generated the instant the log data is created. The timestamp information not only records the time but also the order in which the log information was generated. Based on this, this disclosure innovatively references information bits in convolutional codes for encoding.
[0039] For example, taking 1024 log data entries as an example, the log entry with sequence 1 is encoded as 0000000001 (nine zeros and one one), the log entry with sequence 2 is encoded as 0000000010, and the log entry with sequence 512 is encoded as 0100000000. And so on, the location encoding information can be obtained. Compared to one-hot encoding, which is 1024-dimensional, the information bit portion of the encoding disclosed herein is only 10-dimensional, significantly reducing data dimensionality and improving the data processing efficiency for fault detection or location.
[0040] Then, based on the data category of each log data, the corresponding convolutional encoder is determined.
[0041] Finally, based on the position encoding of each log data entry, the convolutional code is generated using the determined convolutional encoder.
[0042] Convolutional codes add check bits by performing convolution operations on the input parameters of a convolutional encoder, increasing the Hamming distance between different coded sequences to address bit errors that occur during channel transmission. If the erroneous symbol is smaller than the Hamming distance, a transmission error can be determined. Based on this, this disclosure innovatively combines log data types with convolutional coding circuits, using different convolutional coding circuits according to the log data type, and expressing or displaying differences between log information by setting different convolution kernels. Log data of the same type uses the same coding circuit for matrix operations.
[0043] Figure 2 This is a schematic diagram illustrating the process of obtaining a word embedding matrix according to some embodiments of the present disclosure.
[0044] like Figure 2 As shown, taking log data of two different data categories as an example, two convolutional coding circuits (convolutional encoders) g1 and g2 are pre-designed, with their convolutional expressions represented, for example, as 1011 and 1111, respectively. The greater the difference in data categories between the log data, the greater the difference in the binary numbers of the corresponding convolutional encoder's convolutional expression. First, based on the location information of the log data (such as location identifiers), the input parameters of the convolutional encoder are constructed, thus obtaining the corresponding location code 1011. Then, based on the data category of the log data, the convolutional encoder corresponding to the data category is selected. For example, Figure 2 The log data corresponds to convolutional encoder g2. Position code 1011 is input into convolutional encoder g2 for convolution operation, and finally the corresponding convolutional code 110100011 is output from switch C2. For multiple log data, the corresponding convolutional code is output by pressing the switch in turn, and the multiple convolutional codes form a word embedding matrix.
[0045] Different encoding circuits are used depending on the type of log data. This allows the neural network to better capture the differences between log data. Furthermore, since the output of convolutional codes after passing through a convolutional encoder has excellent properties such as orthogonality and randomness, it is more suitable for training the neural network. By setting different convolutional kernels for Argenomic encoding based on the type of log data, the Hamming distance in the convolutional code is correlated with the differences in log data types. This enhances the neural network's understanding of the word embedding matrix and improves the accuracy of fault detection.
[0046] return Figure 1In step S130, fault location information is determined using a sequence model based on the convolutional codes of multiple log data. In some embodiments, server log activity information from the Tianchi platform (open source) can be used as training data to train the sequence model. For example, the sequence model is a Transformer model.
[0047] In some embodiments, fault location information can be determined by using a sequence model based on the convolutional codes of multiple log data in the following manner.
[0048] First, the multiple log data entries are grouped to obtain multiple groups of log data.
[0049] Then, the convolutional codes corresponding to each group of log data are merged to obtain the word embedding matrix corresponding to each group of log data. For example, the convolutional codes corresponding to multiple log data in each group of log data are merged (e.g., sequentially concatenated) to obtain the word embedding matrix.
[0050] Finally, based on the word embedding matrix corresponding to multiple sets of log data, the fault location information is determined using the sequence model.
[0051] In some embodiments, the sequence model includes an encoder, a decoder, and a softmax layer (normalization layer). In this case, fault location information can be determined using the sequence model based on word embedding matrices corresponding to multiple sets of log data in the following manner.
[0052] First, based on multiple word embedding matrices, the encoder is used to determine the hidden layer semantic vector of each word embedding matrix.
[0053] Then, using an attention mechanism, the weights of the hidden layer semantic vectors of each word embedding matrix are determined, where the weights characterize the importance of the data category to fault detection. In some embodiments, the information coefficients of the data category for fault detection for each group of log data are first calculated using the sklearn feature selection library; then, based on the information coefficients corresponding to each word embedding matrix, the weights of the hidden layer semantic vectors of each word embedding matrix are determined. The information coefficients characterize the importance of the data category of the corresponding group of log data to fault detection.
[0054] This disclosure calculates the information coefficients for fault detection of different types of log data using the sklearn feature selection library. It sorts and assigns values to the log data types based on the gain value of fault detection, calculates these values, and innovatively uses them as parameters for the attention mechanism. Each time the neural network receives a word vector input from the previously processed log data, it first sends it to the sklearn library in Python to calculate the maximum information coefficient, and uses the calculated value as a parameter for the attention mechanism, i.e., the weight of the log data.
[0055] The classic mutual information (IC) is a method for evaluating the correlation between qualitative independent and dependent variables. Its value effectively reflects the predictive ability of a factor for the target value; a higher IC indicates a stronger predictive ability for fault detection in that period. The formula for calculating mutual information is as follows: In the above embodiments, the maximum mutual information method is used to adjust the attention mechanism of the neural network on different log data, giving greater weight to noteworthy data, which helps the neural network to better understand the meaning of the data and improve the accuracy of fault detection or fault location.
[0056] Finally, based on the hidden layer semantic vectors of the multiple word embedding matrices and their corresponding weights, the fault location information is determined using the decoder and the softmax layer. In some embodiments, the hidden layer semantic vector corresponding to each word embedding matrix is multiplied by its weight; the product is then sequentially input into the decoder and the softmax layer to obtain the fault location information.
[0057] Figure 3 This is a schematic diagram illustrating the determination of fault location information according to some embodiments of the present disclosure.
[0058] like Figure 3 As shown, each word embedding matrix is input as a sentence vector into the encoder, resulting in the hidden layer semantic vector output by the encoder. Taking sentence vectors 1, 2, 3, and 4 as inputs to the encoder as an example, the encoder outputs hidden layer semantic vectors C1, C2, C3, and C4. The sequence model, as a neural network model, provides an attention mechanism. This attention mechanism determines the information coefficients for fault detection in each group of log data (sentences) 1, 2, 3, and 4. Each group of log data belongs to the same data category. The information coefficients characterize the importance of the data category of the corresponding group of log data to fault detection. The information coefficients of each group of log data are used as weights for the hidden layer semantic vectors corresponding to each group of log data. The product of these weights and the hidden layer semantic vectors is then passed through the data decoder and the softmax layer to obtain the outputs of sentence vectors 1, 2, 3, and 4, respectively, which serve as fault location information.
[0059] Figure 3 The neural network model shown employs a seq2seq architecture, which is common to sequence models. Compared to RNNs, it can freely control the lengths of the input and output sequences. This disclosure processes log data generated by a network or server, using log information to determine the type of fault. Since the amount of log data generated within a single time period is variable, a seq2seq architecture is used, ensuring that the output at each time step is constrained by the entire input sequence. The essence of the seq2seq architecture can be viewed as a conditional language model (i.e., a conditional probability model): P(Y│X)=P(y_1│x)P(y_2│y_1,x)…
[0060] In some embodiments, the encoder can be implemented using a recurrent neural network, where the hidden layer semantic vector is the last hidden state of the recurrent neural network. This hidden state is passed to the decoder after passing through a fully connected layer. The input to the encoder is a vector obtained by convolution and pooling the word vector matrix, which is fed into the encoder in fixed-length groups of 30 log data entries.
[0061] In some embodiments, the decoder may also employ a recurrent neural network. During training, the decoder accepts the hidden state provided by the encoder and the output vector of the convolutional layer as input. During prediction, since there is no target sequence, it accepts the output of the decoder from the previous time step and the hidden state provided by the encoder as input, and outputs the probability distribution predicted at the next time step.
[0062] In the above embodiments, log data enables the location of abnormal requests, the tracing of program execution logic, and the execution of more granular fault diagnosis. Location identifiers reflect the temporal information of the log data, allowing sequence models to capture this information. Combined with data categories, this enables rapid and accurate fault detection or location, improving fault detection performance and facilitating cloud-network convergence.
[0063] Figure 4 This is a block diagram illustrating a fault detection apparatus according to some embodiments of the present disclosure.
[0064] like Figure 4 As shown, the fault detection device 4 includes an acquisition module 41, an execution module 42, and a determination module 43.
[0065] The acquisition module 41 is configured to acquire multiple log data entries, each of which has a location identifier and a data category. The location identifiers of the multiple log data entries reflect their temporal relationship. For example, when performing an operation... Figure 1 The step S110 is shown.
[0066] Execution module 42 is configured to perform convolutional code encoding operations based on the position identifier of each log data entry, using a convolutional encoder corresponding to the data category of each log data entry, to obtain the convolutional code for each log data entry. For example, it performs the following... Figure 1 The step S120 shown.
[0067] Module 43 is configured to determine fault location information based on the convolutional codes of multiple log data entries, using a sequence model, for example, by executing... Figure 1 The step S130 shown.
[0068] Figure 5 This is a block diagram illustrating a fault detection apparatus according to other embodiments of the present disclosure.
[0069] like Figure 5 As shown, the fault detection device 5 includes a memory 51 and a processor 52 coupled to the memory 51. The memory 51 is used to store instructions for executing embodiments of the fault detection method. The processor 52 is configured to execute fault detection methods in any of the embodiments of this disclosure based on the instructions stored in the memory 51.
[0070] Figure 6 This is a block diagram illustrating a computer system for implementing some embodiments of the present disclosure.
[0071] like Figure 6 As shown, the computer system 60 can be represented in the form of a general computing device. The computer system 60 includes a memory 610, a processor 620, and a bus 600 connecting different system components.
[0072] The memory 610 may include, for example, system memory, non-volatile storage media, etc. The system memory may store, for example, an operating system, application programs, a boot loader, and other programs. The system memory may include volatile storage media, such as random access memory (RAM) and / or cache memory. The non-volatile storage media may store, for example, instructions for performing at least one embodiment of a fault detection method. Non-volatile storage media include, but are not limited to, disk storage, optical storage, flash memory, etc.
[0073] The processor 620 can be implemented using a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete hardware components such as discrete gates or transistors. Accordingly, each module, such as the decision module and the determination module, can be implemented by executing instructions in the central processing unit (CPU) memory to perform the corresponding steps, or by implementing dedicated circuitry to perform the corresponding steps.
[0074] Bus 600 can use any of the various bus architectures. For example, bus architectures include, but are not limited to, Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MCA) bus, and Peripheral Component Interconnect (PCI) bus.
[0075] The computer system 60 may also include an input / output interface 630, a network interface 640, and a storage interface 650. These interfaces 630, 640, and 650, as well as the memory 610 and processor 620, can be connected via a bus 600. The input / output interface 630 provides a connection interface for input / output devices such as a monitor, mouse, and keyboard. The network interface 640 provides a connection interface for various networked devices. The storage interface 650 provides a connection interface for external storage devices such as floppy disks, USB flash drives, and SD cards.
[0076] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus, and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations thereof, can be implemented by computer-readable program instructions.
[0077] These computer-readable program instructions are provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable device to produce a machine, such that execution of the instructions by the processor produces means for implementing the functions specified in one or more boxes of the flowchart and / or block diagram.
[0078] These computer-readable program instructions may also be stored in a computer-readable storage medium. These instructions cause a computer to work in a particular manner to produce an article of manufacture, including instructions that implement the functions specified in one or more boxes in a flowchart and / or block diagram.
[0079] This disclosure may take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects.
[0080] The fault detection method and apparatus, and the computer storage medium described in the above embodiments can achieve rapid and accurate fault detection.
[0081] The fault detection method and apparatus, and the computer-storable medium according to this disclosure have been described in detail above. To avoid obscuring the concept of this disclosure, some details known in the art have not been described. Those skilled in the art can fully understand how to implement the technical solutions disclosed herein based on the above description.
Claims
1. A fault detection method, comprising: Multiple log data entries are obtained, each log data entry having a location identifier and a data category, and the location identifiers of the multiple log data entries reflect the temporal relationship of the multiple log data entries; Based on the location identifier of each log data, a convolutional code encoding operation is performed using a convolutional encoder corresponding to the data category of each log data to obtain the convolutional code of each log data. This includes: encoding the location identifier of each log data to obtain the location code; determining the corresponding convolutional encoder based on the data category of each log data; and generating the convolutional code using the determined convolutional encoder based on the location code of each log data. Based on the convolutional codes of multiple log data, a sequence model is used to determine fault location information, including: grouping the multiple log data to obtain multiple sets of log data; merging the convolutional codes corresponding to each set of log data to obtain a word embedding matrix corresponding to each set of log data; and using the sequence model based on the word embedding matrices corresponding to the multiple sets of log data.
2. The fault detection method according to claim 1, wherein, The sequence model includes an encoder, a decoder, and a softmax layer. Based on the word embedding matrix corresponding to multiple sets of log data, the sequence model is used to determine fault location information, including: Based on multiple word embedding matrices, the encoder is used to determine the hidden layer semantic vector of each word embedding matrix; Using an attention mechanism, the weights of the hidden layer semantic vectors in each word embedding matrix are determined, where the weights characterize the importance of data categories to fault detection; Based on the hidden layer semantic vectors of the multiple word embedding matrices and their corresponding weights, the fault location information is determined using the decoder and the softmax layer.
3. The fault detection method according to claim 2, wherein, Using an attention mechanism, the weights of the hidden layer semantic vectors in each word embedding matrix are determined as follows: Using the feature selection library of sklearn, the information coefficient of the data category for fault detection of each group of log data is calculated. The information coefficient represents the importance of the data category of the corresponding group of log data for fault detection. The weights of the hidden layer semantic vectors of each word embedding matrix are determined based on the information coefficients corresponding to each word embedding matrix.
4. The fault detection method according to claim 3, wherein, Based on the hidden layer semantic vectors of the multiple word embedding matrices and their corresponding weights, the fault location information is determined using the decoder and the softmax layer, including: Multiply the hidden layer semantic vector corresponding to each word embedding matrix by the weight; The product obtained by multiplying is then input into the decoder and the softmax layer in sequence to obtain fault location information.
5. The fault detection method according to claim 1 further includes: The location identifier of each log data is determined based on the timestamp information of multiple log data.
6. A fault detection device, comprising: The acquisition module is configured to acquire multiple log data entries, each of which has a location identifier and a data category. The location identifiers of the multiple log data entries reflect the temporal relationship of the multiple log data entries. The execution module is configured to perform convolutional code encoding operations based on the position identifier of each log data entry, using a convolutional encoder corresponding to the data category of each log data entry, to obtain the convolutional code for each log data entry. This includes: encoding the position identifier of each log data entry to obtain the position code; determining the corresponding convolutional encoder based on the data category of each log data entry; and generating the convolutional code using the determined convolutional encoder based on the position code of each log data entry. The determination module is configured to determine fault location information based on the convolutional codes of multiple log data and using a sequence model, including: grouping the multiple log data to obtain multiple sets of log data; merging the convolutional codes corresponding to each set of log data to obtain a word embedding matrix corresponding to each set of log data; and determining the fault location information based on the word embedding matrices corresponding to the multiple sets of log data and using the sequence model.
7. A fault detection device, comprising: Memory; as well as A processor coupled to the memory, the processor being configured to execute the fault detection method as described in any one of claims 1 to 5 based on instructions stored in the memory.
8. A computer-storeable medium having stored thereon computer program instructions that, when executed by a processor, implement the fault detection method as described in any one of claims 1 to 5.