Operation and maintenance fault prediction method and device based on large model, equipment and medium

Abstract

The invention discloses an operation and maintenance fault prediction method and device based on a large model, equipment and a medium, and relates to the technical field of artificial intelligence, and the method comprises the steps: obtaining initial alarm data based on a preset data grade condition and a preset time condition, and determining a target instruction template corresponding to the initial alarm data; performing data preprocessing on the initial alarm data according to the target instruction template and a target large model to obtain corresponding target alarm data; and constructing a target operation and maintenance fault prediction model based on a preset collaborative neural network and historical alarm data so as to output a target operation and maintenance fault prediction result corresponding to the target alarm data by using the target operation and maintenance fault prediction model, and sending the target operation and maintenance fault prediction result based on a preset notification mode. In this way, in combination with the data processing capability of the large model and the adaptive learning characteristic of the collaborative neural network, the efficiency and accuracy of operation and maintenance fault prediction are significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a large model-based operation and maintenance fault prediction method, device, equipment and medium. Background Art

[0002] With the continuous expansion of the scale and increasing complexity of information systems, operation and maintenance management is facing unprecedented severe challenges. Timely prediction of operation and maintenance failures has become a key means to improve operation and maintenance efficiency and ensure service continuity, and can provide the operation and maintenance team with sufficient time windows to take preventive measures, thereby effectively reducing the occurrence of failures. In addition, accurate fault prediction can ensure that users continue to receive stable and high-quality services, thereby significantly improving user satisfaction. However, the current field of operation and maintenance fault prediction still faces many difficulties, especially when dealing with high-dimensional, nonlinear, and time-series operation and maintenance fault data, traditional methods often find it difficult to achieve ideal results. And due to the problem of unbalanced data samples, it is difficult for the prediction model to accurately capture the precursor characteristics of the failure, so that it is impossible to issue an early warning in time, and further aggravate the risk and cost of operation and maintenance management.

[0003] To sum up, how to achieve efficient and accurate prediction of operation and maintenance failures is a technical problem that needs to be solved urgently. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a method, device, equipment and medium for predicting operation and maintenance faults based on a large model, which can achieve efficient and accurate prediction of operation and maintenance faults. The specific scheme is as follows:

[0005] In a first aspect, the present application provides an operation and maintenance fault prediction method based on a large model, comprising:

[0006] Acquire initial warning data based on preset data level conditions and preset time conditions, and determine a target instruction template corresponding to the initial warning data;

[0007] Preprocessing the initial warning data according to the target instruction template and the target large model to obtain corresponding target warning data;

[0008] A target operation and maintenance fault prediction model is constructed based on a preset collaborative neural network and historical alarm data, so as to use the target operation and maintenance fault prediction model to output a target operation and maintenance fault prediction result corresponding to the target alarm data, and send the target operation and maintenance fault prediction result based on a preset notification method.

[0009] Optionally, the obtaining of initial alarm data based on a preset data level condition and a preset time condition includes:

[0010] Determine the data levels corresponding to the alarm data in the preset data repository, and determine the alarm data to be extracted that meets the preset data level conditions from the preset data repository, so as to determine the initial alarm data based on the alarm data to be extracted and the preset time conditions in the preset data repository.

[0011] Optionally, the determining the initial alarm data based on the alarm data to be extracted and the preset time conditions in the preset data repository includes:

[0012] Determine a preset data extraction period, and obtain the alarm data to be extracted in the preset data repository within the current time period every preset data extraction period to obtain corresponding data to be processed;

[0013] Or, obtain the alarm data to be extracted in the preset data repository in real time to obtain corresponding data to be processed;

[0014] Filter the data to be processed based on preset filtering conditions to obtain the initial alarm data, and determine the target performance indicators corresponding to the initial alarm data.

[0015] Optionally, the determining the target instruction template corresponding to the initial alarm data includes:

[0016] Divide the initial alarm data based on preset area conditions, and determine the target instruction template corresponding to the divided initial alarm data according to preset data format conditions;

[0017] Correspondingly, the data preprocessing the initial alarm data according to the target instruction template and the target large model to obtain corresponding target alarm data includes:

[0018] Input the divided initial alarm data into a large model based on a preset neural network architecture, and use the large model based on the preset neural network architecture to output the corresponding target alarm data according to the target instruction template.

[0019] Optionally, the constructing the target operation and maintenance fault prediction model based on a preset collaborative neural network and historical alarm data includes:

[0020] Construct a preset embedding layer of the target operation and maintenance fault prediction model based on a preset multi-layer perceptron and a preset rectified linear unit, so as to use the preset embedding layer to encode the input binary data into a target floating point number;

[0021] Construct a preset memory layer of the target operation and maintenance fault prediction model according to a preset unbiased multi-layer perceptron, and construct a preset unflattened layer of the target operation and maintenance fault prediction model according to the first target dimension;

[0022] Construct a preset neural network layer of the target operation and maintenance fault prediction model based on a preset activation function, a preset backpropagation algorithm, a preset normalization technique, and a preset number of iterations, and construct a preset flattening layer of the target operation and maintenance fault prediction model according to a second target dimension;

[0023] Construct a preset output layer of the target operation and maintenance fault prediction model based on the preset unbiased multi-layer perceptron, so as to construct the target operation and maintenance fault prediction model according to the preset embedding layer, the preset memory layer, the preset non-flattening layer, the preset neural network layer, the preset flattening layer, the preset output layer, and the historical alarm data.

[0024] Optionally, constructing the target operation and maintenance fault prediction model according to the preset embedding layer, the preset memory layer, the preset non-flattening layer, the preset neural network layer, the preset flattening layer, the preset output layer, and the historical alarm data includes:

[0025] Construct an initial operation and maintenance fault prediction model according to the preset embedding layer, the preset memory layer, the preset non-flattening layer, the preset neural network layer, the preset flattening layer, and the preset output layer;

[0026] Determine a target gradient based on a preset cross-entropy loss function, a preset error backpropagation algorithm, the historical alarm data, and the initial operation and maintenance fault prediction model;

[0027] Use a preset optimizer to update the model parameters corresponding to the initial operation and maintenance fault prediction model according to a preset correction weight decay technique and the target gradient to obtain the target operation and maintenance fault prediction model.

[0028] Optionally, sending the target operation and maintenance fault prediction result based on a preset notification method includes:

[0029] Receive in real time the target operation and maintenance fault prediction result corresponding to the target alarm data output by the target operation and maintenance fault prediction model based on a preset data interface;

[0030] Use the preset data interface to send the target operation and maintenance fault prediction result according to a preset hierarchical notification strategy and the preset notification method, and receive a target sending result corresponding to the target operation and maintenance fault prediction result, so as to process the target sending result based on a preset error handling mechanism.

[0031] In a second aspect, the present application provides an operation and maintenance fault prediction device based on a large model, including:

[0032] A target instruction template determination module, configured to obtain initial alarm data based on a preset data level condition and a preset time condition, and determine a target instruction template corresponding to the initial alarm data;

[0033] A target alarm data determination module, configured to perform data preprocessing on the initial alarm data according to the target instruction template and a target large model to obtain corresponding target alarm data;

[0034] A target operation and maintenance fault prediction result output module, configured to construct a target operation and maintenance fault prediction model based on a preset collaborative neural network and historical alarm data, so as to output a target operation and maintenance fault prediction result corresponding to the target alarm data by using the target operation and maintenance fault prediction model, and send the target operation and maintenance fault prediction result based on a preset notification method.

[0035] In a third aspect, the present application provides an electronic device, including:

[0036] A memory, configured to store a computer program;

[0037] A processor, configured to execute the computer program to implement the foregoing operation and maintenance fault prediction method based on a large model.

[0038] In a fourth aspect, the present application provides a computer-readable storage medium, configured to store a computer program; wherein, when the computer program is executed by a processor, the foregoing operation and maintenance fault prediction method based on a large model is implemented.

[0039] In this application, first, initial alarm data is obtained based on preset data level conditions and preset time conditions, and the target instruction template corresponding to the initial alarm data is determined; then, data preprocessing is performed on the initial alarm data according to the target instruction template and the target large model to obtain corresponding target alarm data; finally, a target operation and maintenance failure prediction model is constructed based on a preset collaborative neural network and historical alarm data, so as to use the target operation and maintenance failure prediction model to output the target operation and maintenance failure prediction result corresponding to the target alarm data, and send the target operation and maintenance failure prediction result based on a preset notification method. As can be seen from the above, in this application, initial alarm data is first obtained, and the target instruction template corresponding to the initial alarm data is determined. Then, data preprocessing is performed on the initial alarm data using the target instruction template and the target large model to obtain target alarm data. Finally, the target operation and maintenance failure prediction model constructed based on the preset collaborative neural network is used to perform operation and maintenance failure prediction on the target alarm data to obtain the target operation and maintenance failure prediction result, and the target operation and maintenance failure prediction result is sent. In this way, in this application, by combining the data processing ability of the large model and the adaptive learning characteristics of the collaborative neural network, effective and accurate prediction of operation and maintenance failures is achieved. It effectively solves the problem of processing operation and maintenance alarm data with high dimensions, non-linearity, and strong time series, and solves the problem of unbalanced data samples. In this way, this application uses the large model and the collaborative neural network, which can significantly improve the efficiency and accuracy of operation and maintenance failure prediction. Description of the Drawings

[0040] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained according to the provided drawings.

[0041] Figure 1 Flowchart of a method for predicting operation and maintenance failures based on a large model provided by this application;

[0042] Figure 2 Flowchart of a specific method for predicting operation and maintenance failures based on a large model provided by this application;

[0043] Figure 3 Architecture diagram of a specific operation and maintenance failure prediction model provided by this application;

[0044] Figure 4 Flowchart of a specific method for predicting operation and maintenance failures based on a large model provided by this application;

[0045] Figure 5 Structure diagram of a device for predicting operation and maintenance failures based on a large model provided by this application;

[0046] Figure 6 A structural diagram of an electronic device provided for this application. Specific implementation manners

[0047] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0048] With the continuous expansion of the scale and the increasing complexity of information systems, operation and maintenance management work is facing unprecedented severe challenges. Timely prediction of operation and maintenance failures has become a key means to improve operation and maintenance efficiency and ensure service continuity, and can provide sufficient time windows for the operation and maintenance team to take preventive measures, thereby effectively reducing the occurrence rate of failures. In addition, accurate failure prediction can ensure that users continuously obtain stable and high-quality services, thereby significantly improving user satisfaction. However, the current operation and maintenance failure prediction field still faces many problems. Especially when dealing with operation and maintenance failure data with high dimensionality, non-linearity, and strong time series characteristics, traditional methods often fail to achieve ideal results. And due to the problem of unbalanced data samples, it is difficult for the prediction model to accurately capture the precursor characteristics of failures, so it cannot issue early warnings in time, and further exacerbates the risks and costs of operation and maintenance management. For this reason, this application provides an operation and maintenance failure prediction scheme based on a large model, which can achieve efficient and accurate prediction of operation and maintenance failures.

[0049] See Figure 1 As shown, the embodiments of the present invention disclose an operation and maintenance failure prediction method based on a large model, which may include:

[0050] Step S11, obtain initial alarm data based on a preset data level condition and a preset time condition, and determine a target instruction template corresponding to the initial alarm data.

[0051] In this embodiment, the obtaining of the initial alarm data based on the preset data level condition and the preset time condition may include: determining the data level corresponding to each alarm data in a preset data repository, and determining the to-be-extracted alarm data that meets the preset data level condition from the preset data repository based on the data level, so as to determine the initial alarm data based on the to-be-extracted alarm data in the preset data repository and the preset time condition. It can be understood that the alarm data can be stored in a data lake or a database. Specifically, the alarm data can be classified first, and corresponding data levels can be set, so as to extract the alarm data with a data level of S from the data lake or the database.

[0052] It should be noted that, in order to ensure the timeliness and accuracy of data, reduce the burden on operation and maintenance personnel for manually collecting data, and improve the overall work efficiency. The determination of the initial alarm data based on the to-be-extracted alarm data and the preset time condition in the preset data repository may include: determining a preset data extraction period, and obtaining the to-be-extracted alarm data in the preset data repository within the current time period every the preset data extraction period to obtain corresponding data to be processed; or, obtaining the to-be-extracted alarm data in the preset data repository in real time to obtain corresponding data to be processed; filtering the data to be processed based on a preset filtering condition to obtain the initial alarm data, and determining the target performance indicator corresponding to the initial alarm data. Specifically, according to business requirements, the data to be processed can be determined automatically from the alarm data with a data level of S in the data lake or database according to the set data extraction period or real-time trigger condition. After that, in order to improve the quality of the alarm data, the data to be processed needs to be further filtered, frequent alarm information is eliminated, and relevant performance indicators are added, where the time interval of the alarm data is in hours.

[0053] In this embodiment, the determination of the target instruction template corresponding to the initial alarm data may include: dividing the initial alarm data based on a preset area condition, and determining the target instruction template corresponding to the divided initial alarm data according to a preset data format condition. Specifically, the initial alarm data can be divided by alarm area, and an appropriate Prompt instruction can be defined according to the preset data format condition.

[0054] Step S12, perform data preprocessing on the initial alarm data according to the target instruction template and the target large model to obtain corresponding target alarm data.

[0055] It can be understood that, in order to ensure that the target operation and maintenance fault prediction model can obtain high-quality data information and effectively avoid prediction errors caused by incompatible data formats or improper preprocessing. The above data preprocessing of the initial alarm data according to the target instruction template and the target large model to obtain the corresponding target alarm data may include: inputting the divided initial alarm data into a large model based on a preset neural network architecture, and using the large model based on the preset neural network architecture to output the corresponding target alarm data according to the target instruction template. Specifically, the initial alarm data can be processed by calling a large model based on the Transformer framework. In a specific implementation, it can be defined that "mapping the input alarm data 'BmcTrap,cmp045 / KYTenantVServerHightCpuUsageCritical' to a set of 0-1 encodings representing the alarm name, such as {1,0,1,0,...,0} representing the input alarm data 'Alarm 0 BmcTrap and Alarm 2 cmp045 / KYTenantVServerHightCpuUsageCritical, and no other alarms occurred'."

[0056] Step S13, construct a target operation and maintenance fault prediction model based on a preset collaborative neural network and historical alarm data, so as to use the target operation and maintenance fault prediction model to output the target operation and maintenance fault prediction result corresponding to the target alarm data, and send the target operation and maintenance fault prediction result based on a preset notification method.

[0057] In this embodiment, the adaptive learning characteristics of the collaborative neural network can be utilized to construct a target operation and maintenance fault prediction model consisting of an embedding layer, a Memory layer (memory layer), an Unflatten layer (unflattening layer), a Syn layer (neural network layer), a Flatten layer (flattening layer), and an output layer based on historical alarm data, and use the target operation and maintenance fault prediction model to determine the target operation and maintenance fault prediction result corresponding to the target alarm data.

[0058] It should be noted that sending the target operation and maintenance failure prediction result based on the preset notification method may include: receiving in real time the target operation and maintenance failure prediction result corresponding to the target alarm data output by the target operation and maintenance failure prediction model based on the preset data interface; using the preset data interface to send the target operation and maintenance failure prediction result according to the preset hierarchical notification strategy and the preset notification method, and receiving the target sending result corresponding to the target operation and maintenance failure prediction result, so as to process the target sending result based on the preset error handling mechanism. Specifically, the target operation and maintenance failure prediction result can be received in real time based on the API interface. After that, a hierarchical notification strategy can be formulated according to the severity or impact scope of the failure. For example, a serious failure can be immediately notified to senior operation and maintenance management personnel, and a general failure can be notified later or only recorded in the log. Then, a corresponding notification email can be generated based on the target operation and maintenance failure prediction result, and the content of the email can include specific information of the target operation and maintenance failure prediction result, such as the failure type, occurrence time, impact scope, etc. The notification email can support batch sending to notify multiple operation and maintenance management personnel at the same time. After that, the sending result of the notification email can be monitored, and the target sending result can be processed based on the preset error handling mechanism. In a specific implementation manner, in order to avoid excessive retries causing too high a system load, the number of retries and the retry interval can be set. For the notifications that fail to be sent, they are sent again based on the number of retries and the retry interval. If the retry fails, an alarm can be triggered and relevant personnel can be notified for handling. For serious errors or errors that cannot be automatically processed, the error log can be recorded and manual intervention can be requested.

[0059] As can be seen from the above, in this embodiment, initial alarm data is first obtained based on a preset data level condition and a preset time condition, and a target instruction template corresponding to the initial alarm data is determined; then, data preprocessing is performed on the initial alarm data according to the target instruction template and a target large model to obtain corresponding target alarm data; finally, a target operation and maintenance failure prediction model is constructed based on a preset collaborative neural network and historical alarm data, so as to output a target operation and maintenance failure prediction result corresponding to the target alarm data by using the target operation and maintenance failure prediction model, and send the target operation and maintenance failure prediction result based on a preset notification method. As can be seen from the above, in this embodiment, initial alarm data is first obtained, and a target instruction template corresponding to the initial alarm data is determined. Then, data preprocessing is performed on the initial alarm data by using the target instruction template and the target large model to obtain target alarm data. Finally, the target operation and maintenance failure prediction model constructed based on the preset collaborative neural network is used to perform operation and maintenance failure prediction on the target alarm data to obtain a target operation and maintenance failure prediction result, and the target operation and maintenance failure prediction result is sent. In this way, in this embodiment, by combining the data processing ability of the large model and the adaptive learning characteristics of the collaborative neural network, effective and accurate prediction of operation and maintenance failures is realized. The problem of processing operation and maintenance alarm data with high dimensionality, non-linearity, and strong time series is effectively solved, and the problem of unbalanced data samples is also solved. In this way, in this embodiment, by using the large model and the collaborative neural network, the efficiency and accuracy of operation and maintenance failure prediction can be significantly improved.

[0060] Based on the previous embodiment, it can be known that this application can combine the data processing ability of the large model and the adaptive learning characteristics of the collaborative neural network to realize the prediction of operation and maintenance failures. Next, in this embodiment, how to construct a target operation and maintenance failure prediction model based on the collaborative neural network will be elaborated in detail. Refer to Figure 2 As shown, an embodiment of the present invention further discloses an operation and maintenance failure prediction method based on a large model, which may include:

[0061] Step S21: Obtain initial alarm data based on a preset data level condition and a preset time condition, and determine a target instruction template corresponding to the initial alarm data.

[0062] Step S22: Perform data preprocessing on the initial alarm data according to the target instruction template and a target large model to obtain corresponding target alarm data.

[0063] Step S23: Construct a target operation and maintenance failure prediction model based on a preset collaborative neural network and historical alarm data, so as to output a target operation and maintenance failure prediction result corresponding to the target alarm data by using the target operation and maintenance failure prediction model, and send the target operation and maintenance failure prediction result based on a preset notification method.

[0064] Refer to Figure 3As shown, in this embodiment, to solve the problem of unbalanced historical alarm sequence data samples and improve the accuracy of operation and maintenance fault prediction results, the above-mentioned target operation and maintenance fault prediction model constructed based on a preset collaborative neural network and historical alarm data may include: constructing a preset embedding layer of the target operation and maintenance fault prediction model based on a preset multi-layer perceptron and a preset rectified linear unit, so as to encode the input binary data into a target floating-point number by using the preset embedding layer; constructing a preset memory layer of the target operation and maintenance fault prediction model according to a preset bias-free multi-layer perceptron, and constructing a preset unflattening layer of the target operation and maintenance fault prediction model according to a first target dimension; constructing a preset neural network layer of the target operation and maintenance fault prediction model based on a preset activation function, a preset backpropagation algorithm, a preset normalization technique, and a preset number of iterations, and constructing a preset flattening layer of the target operation and maintenance fault prediction model according to a second target dimension; constructing a preset output layer of the target operation and maintenance fault prediction model based on the preset bias-free multi-layer perceptron, so as to construct the target operation and maintenance fault prediction model according to the preset embedding layer, the preset memory layer, the preset unflattening layer, the preset neural network layer, the preset flattening layer, the preset output layer, and the historical alarm data. Specifically, the target operation and maintenance fault prediction model consists of six parts: an embedding layer, a Memory layer, an Unflatten layer, a Syn layer, a Flatten layer, and an output layer.

[0065] It can be understood that the embedding layer of the target operation and maintenance fault prediction model uses an MLP layer (Multilayer Perceptron), and the activation function uses ReLU (Rectified Linear Unit). The embedding layer changes the input dimension to emb_dim, and the dimensions of other layers remain emb_dim. Since binary data is not conducive to the training process of the neural network, the binary data can be encoded to improve the training effect of the neural network. The embedding layer is used to format the network input and can encode the input 0-1 binary data into a floating-point number. The embedding layer can be understood as a form of dynamic encoding. Compared with static encoding, the encoding of the embedding layer can be dynamically adjusted during the training of the neural network, and the encoding effect is automatically optimized with the working effect of the neural network, which is beneficial to further improving the training effect of the neural network. The multi-layer embedding layer can be understood as the non-linear encoding of the input. Compared with the single embedding layer, the multi-layer embedding layer can optimize the data distribution and reduce the fitting difficulty of the neural network, which is beneficial to further improving the training effect of the neural network. The number of embedding layers emb_layer_num and the dimension emb_dim are hyperparameters and can be determined by experience or grid search.

[0066] It should be noted that the Memory layer of the target operation and maintenance fault prediction model, that is, the memory layer, uses an unbiased MLP layer without an activation function. The Memory layer takes the inner product of the input of the embedding layer and the stored memory represented by the parameters within the layer to obtain the cooperative order parameter, and the data dimension is scaled from [batch_num, emb_layer_num] to [batch_num, pattern_multiplier * class_num]. Among them, the purpose of introducing pattern_multiplier is to improve the network performance. The cooperative neural network realizes pattern classification by constructing an in-built memory storage, where class_num refers to the number of categories of the original pattern classification problem. To improve the classification performance, the number of stored memories can be multiplied by pattern_multiplier, so that pattern_multiplier memories correspond to the same category to improve the network performance.

[0067] In this embodiment, the Unflatten layer of the target operation and maintenance fault prediction model, that is, the unflattening layer, has no activation function. The Unflatten layer changes the dimension of the output of the Memory layer from [batch_num, pattern_multiplier * class_num] to [batch_num, pattern_multiplier, class_num].

[0068] It should be noted that the Syn layer of the target operation and maintenance fault prediction model, that is, the neural network layer, is a cyclic syn layer that uses the syn activation function. The Syn layer first receives the input x from the Unflatten layer with a dimension of [batch_num, pattern_multiplier, class_num], and then receives the output output of the previous Syn layer with a dimension of [batch_num, pattern_multiplier, class_num]. After cycling {syn_iter} times, the final output output has a dimension of [batch_num, pattern_multiplier, class_num]. The specific details are as follows:

[0069] (1) Define the cooperative attention hyperparameter lambda = torch.ones(size=(1, pattern_multiplier, class_num), requires_grad=True). Among them, requires_grad=True enables lambda to perform self-regulation using the BP algorithm (Backpropagation Algorithm) to form a negatively correlated distribution corresponding to the sample labels.

[0070] (2) Normalize x. The Syn layer requires the input to be normalized data.

[0071] (3) First loop: output = syn(x) = (0.5 * x ** 3).detach() + (0.5 * lambda + 1) * x. Among them, the detach() high-order term can avoid gradient explosion and disappearance.

[0072] (4) Second loop: output = syn(output).

[0073] (4) Loop {syn_iter} times to obtain the final output output.

[0074] The Syn layer can record the label distribution of the samples, making the collaborative attention hyperparameter lambda self-learn to be negatively correlated with the label distribution, and using lambda to assign higher attention to small-class samples to achieve the adaptability of the network to the data distribution.

[0075] In this embodiment, the Flatten layer of the target operation and maintenance fault prediction model, that is, the flattening layer. The Flatten layer changes the dimension of the output of the Memory layer from [batch_num, pattern_multiplier, class_num] to [batch_num, pattern_multiplier * class_num]. The output layer of the target operation and maintenance fault prediction model uses an unbiased MLP layer, which can change the dimension from [batch_num, pattern_multiplier * class_num] to [batch_num, class_num].

[0076] It should be noted that constructing the target operation and maintenance fault prediction model based on the preset embedding layer, the preset memory layer, the preset non-flattening layer, the preset neural network layer, the preset flattening layer, the preset output layer, and the historical alarm data may include: constructing an initial operation and maintenance fault prediction model based on the preset embedding layer, the preset memory layer, the preset non-flattening layer, the preset neural network layer, the preset flattening layer, and the preset output layer; determining the target gradient based on a preset cross-entropy loss function, a preset error backpropagation algorithm, the historical alarm data, and the initial operation and maintenance fault prediction model; using a preset optimizer to update the model parameters corresponding to the initial operation and maintenance fault prediction model according to a preset corrected weight decay technique and the target gradient to obtain the target operation and maintenance fault prediction model. Specifically, the EBP algorithm (Error Back Propagation) can be used to complete model training, and Softmax cross-entropy can be selected as the loss function of the model. The Softmax cross-entropy loss function is a loss function commonly used in classification problems, and the Softmax cross-entropy loss function can directly reflect the distribution of the difference between the model prediction result and the true label. The optimizer can be selected as AdamW (Adaptive Moment Estimation with Weight Decay), and the parameters are the default configuration. Compared with the SGD method (Stochastic Gradient Descent), AdamW can calculate the adaptive learning rate, thereby improving the training speed of the network; compared with the Adam method (Adaptive Moment Estimation), AdamW can further improve the training speed of the network and enhance the generalization effect of the network by introducing the corrected weight decay technique, so as to obtain the target operation and maintenance fault prediction model.

[0077] Among them, for the more specific processing procedures of the above steps S21 and S22, reference can be made to the corresponding content disclosed in the foregoing embodiments, and details are not described herein again.

[0078] As can be seen from the above, in this embodiment, an operation and maintenance fault prediction model consisting of six parts, namely, a preset embedding layer, a preset memory layer, a preset non-flattening layer, a preset neural network layer, a preset flattening layer and a preset output layer, is constructed, and based on the preset error back propagation algorithm, the preset cross entropy loss function and the preset optimizer, the target operation and maintenance fault prediction model is determined according to the preset corrected weight decay technology. In this way, this embodiment designs and trains a collaborative neural network, which specifically solves the sample imbalance problem existing in the historical alarm data, thereby achieving a significant improvement in the accuracy of the operation and maintenance fault prediction results. It helps operation and maintenance personnel to promptly discover potential fault risks, effectively shorten fault handling time, improve operation and maintenance efficiency, and ensure the continuity and stability of system services.

[0079] See also Figure 4 As shown, in a specific implementation, the operation and maintenance fault prediction method process based on the big model can be specifically as follows: first, capture and extract alarm sequence data from the data lake or database in a scheduled or real-time manner; then define a suitable Prompt instruction and call the big model to complete data preprocessing; then design and train a collaborative neural network model consisting of six modules; then call the collaborative neural network model to realize operation and maintenance fault prediction; finally, output the operation and maintenance fault prediction results and send a notification email in a timely manner.

[0080] As can be seen from the above, this application proposes an automated and intelligent operation and maintenance fault prediction method based on a large model and a collaborative neural network. In this way, this application can effectively overcome the problems of data imbalance and high-dimensional nonlinear processing, and significantly improve the accuracy of operation and maintenance fault prediction. In addition, it can provide strong intelligent decision-making support for operation and maintenance management, and accelerate the pace of transformation of operation and maintenance process automation and intelligence. Through the application of operation and maintenance fault prediction methods based on large models, not only can the continuous stability of system services be ensured, but also the operation and maintenance costs can be greatly reduced, which effectively promotes the comprehensive improvement and iterative upgrade of the operation and maintenance management level.

[0081] Accordingly, see Figure 5 As shown, the embodiment of the present application also provides an operation and maintenance fault prediction device based on a large model, which may include:

[0082] The target instruction template determination module 11 is used to obtain the initial alarm data based on the preset data level condition and the preset time condition, and determine the target instruction template corresponding to the initial alarm data;

[0083] A target warning data determination module 12 is used to perform data preprocessing on the initial warning data according to the target instruction template and the target large model to obtain corresponding target warning data;

[0084] The target operation and maintenance fault prediction result output module 13 is used to construct a target operation and maintenance fault prediction model based on a preset collaborative neural network and historical alarm data, so as to output a target operation and maintenance fault prediction result corresponding to the target alarm data by using the target operation and maintenance fault prediction model, and send the target operation and maintenance fault prediction result based on a preset notification method.

[0085] As can be seen from the above, in this application, initial alarm data is first obtained based on preset data level conditions and preset time conditions, and a target instruction template corresponding to the initial alarm data is determined; then, data preprocessing is performed on the initial alarm data according to the target instruction template and the target large model to obtain corresponding target alarm data; finally, a target operation and maintenance fault prediction model is constructed based on a preset collaborative neural network and historical alarm data, so as to output a target operation and maintenance fault prediction result corresponding to the target alarm data by using the target operation and maintenance fault prediction model, and send the target operation and maintenance fault prediction result based on a preset notification method. As can be seen from the above, in this application, initial alarm data is first obtained, and a target instruction template corresponding to the initial alarm data is determined. Then, data preprocessing is performed on the initial alarm data by using the target instruction template and the target large model to obtain target alarm data. Finally, the target operation and maintenance fault prediction model constructed based on the preset collaborative neural network is used to perform operation and maintenance fault prediction on the target alarm data to obtain a target operation and maintenance fault prediction result, and the target operation and maintenance fault prediction result is sent. In this way, in this application, by combining the data processing ability of the large model and the adaptive learning characteristics of the collaborative neural network, effective and accurate prediction of operation and maintenance faults is achieved. The problem of processing operation and maintenance alarm data with high dimensions, non-linearity, and strong time series is effectively solved, and the problem of unbalanced data samples is also solved. In this way, this application uses the large model and the collaborative neural network to significantly improve the efficiency and accuracy of operation and maintenance fault prediction.

[0086] In some specific embodiments, the target instruction template determination module 11 may include:

[0087] The to-be-extracted alarm data determination sub-module is used to determine the data level corresponding to each alarm data in the preset data repository, and determine the to-be-extracted alarm data that meets the preset data level conditions from the preset data repository based on the data level, so as to determine the initial alarm data based on the to-be-extracted alarm data and the preset time conditions in the preset data repository.

[0088] In some specific embodiments, the to-be-extracted alarm data determination sub-module may include:

[0089] A to-be-processed data determination unit, configured to determine a preset data extraction period, and obtain the to-be-extracted alarm data in the preset data repository within the current time period every said preset data extraction period, so as to obtain corresponding to-be-processed data; or, obtain the to-be-extracted alarm data in the preset data repository in real time, so as to obtain corresponding to-be-processed data;

[0090] An initial alarm data determination unit, configured to filter the to-be-processed data based on a preset filtering condition to obtain the initial alarm data, and determine a target performance indicator corresponding to the initial alarm data.

[0091] In some specific embodiments, the target instruction template determination module 11 may include:

[0092] A target instruction template determination unit, configured to divide the initial alarm data based on a preset region condition, and determine the target instruction template corresponding to the divided initial alarm data according to a preset data format condition;

[0093] Correspondingly, the target alarm data determination module 12 may include:

[0094] A target alarm data determination unit, configured to input the divided initial alarm data into a large model based on a preset neural network architecture, and use the large model based on the preset neural network architecture to output corresponding target alarm data according to the target instruction template.

[0095] In some specific embodiments, the target operation and maintenance fault prediction result output module 13 may include:

[0096] A preset embedding layer construction sub-module, configured to construct a preset embedding layer of the target operation and maintenance fault prediction model based on a preset multi-layer perceptron and a preset rectified linear unit, so as to use the preset embedding layer to encode input binary data into a target floating-point number;

[0097] A preset memory layer construction sub-module, configured to construct a preset memory layer of the target operation and maintenance fault prediction model according to a preset bias-free multi-layer perceptron, and construct a preset non-flattening layer of the target operation and maintenance fault prediction model according to a first target dimension;

[0098] A preset flattening layer construction sub-module, configured to construct a preset neural network layer of the target operation and maintenance fault prediction model based on a preset activation function, a preset backpropagation algorithm, a preset normalization technique, and a preset number of iterations, and construct a preset flattening layer of the target operation and maintenance fault prediction model according to a second target dimension;

[0099] The target operation and maintenance fault prediction model construction sub-module is used to construct the preset output layer of the target operation and maintenance fault prediction model based on the preset unbiased multi-layer perceptron, so as to construct the target operation and maintenance fault prediction model according to the preset embedding layer, the preset memory layer, the preset non-flattening layer, the preset neural network layer, the preset flattening layer, the preset output layer, and the historical alarm data.

[0100] In some specific embodiments, the target operation and maintenance fault prediction model construction sub-module may include:

[0101] The initial operation and maintenance fault prediction model construction unit is used to construct an initial operation and maintenance fault prediction model according to the preset embedding layer, the preset memory layer, the preset non-flattening layer, the preset neural network layer, the preset flattening layer, and the preset output layer;

[0102] The target gradient determination unit is used to determine the target gradient based on a preset cross-entropy loss function, a preset error backpropagation algorithm, the historical alarm data, and the initial operation and maintenance fault prediction model;

[0103] The target operation and maintenance fault prediction model construction unit is used to update the model parameters corresponding to the initial operation and maintenance fault prediction model by using a preset optimizer according to the preset correction weight decay technique and the target gradient to obtain the target operation and maintenance fault prediction model.

[0104] In some specific embodiments, the target operation and maintenance fault prediction result output module 13 may include:

[0105] The target operation and maintenance fault prediction result determination unit is used to receive in real time the target operation and maintenance fault prediction result corresponding to the target alarm data output by the target operation and maintenance fault prediction model based on a preset data interface;

[0106] The target operation and maintenance fault prediction result sending unit is used to send the target operation and maintenance fault prediction result according to a preset hierarchical notification policy and the preset notification method by using the preset data interface, and receive the target sending result corresponding to the target operation and maintenance fault prediction result, so as to process the target sending result based on a preset error handling mechanism.

[0107] Furthermore, the embodiments of the present application also disclose an electronic device, Figure 6It is a structural diagram of an electronic device 20 shown according to an exemplary embodiment. The content in the figure should not be regarded as any limitation on the scope of use of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. Among them, the memory 22 is used to store a computer program, and the computer program is loaded and executed by the processor 21 to implement the relevant steps in the operation and maintenance fault prediction method based on a large model disclosed in any of the foregoing embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0108] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows is any communication protocol applicable to the technical solution of this application, and no specific limitation is imposed on it here; the input / output interface 25 is used to obtain external input data or output data to the outside, and its specific interface type can be selected according to specific application needs, and no specific limitation is made here.

[0109] In addition, as a carrier for resource storage, the memory 22 may be a read-only memory, a random access memory, a magnetic disk, or an optical disc, etc. The resources stored thereon may include an operating system 221, a computer program 222, etc., and the storage method may be temporary storage or permanent storage.

[0110] Among them, the operating system 221 is used to manage and control each hardware device and the computer program 222 on the electronic device 20, and it may be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program that can be used to complete the operation and maintenance fault prediction method based on a large model executed by the electronic device 20 disclosed in any of the foregoing embodiments, the computer program 222 may further include a computer program that can be used to complete other specific tasks.

[0111] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the operation and maintenance fault prediction method based on a large model disclosed above. For the specific steps of this method, reference may be made to the corresponding content disclosed in the foregoing embodiments, and details will not be repeated here.

[0112] In this specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. The same or similar parts among the various embodiments can be referred to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0113] Those skilled in the art can further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described according to functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of this application.

[0114] The steps of the methods or algorithms described in combination with the embodiments disclosed herein can be directly implemented by hardware, software modules executed by a processor, or a combination of both. The software modules can be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium well-known in the technical field.

[0115] Finally, it should also be noted that in this document, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or also includes elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "comprising a..." does not exclude the presence of additional identical elements in the process, method, article or device comprising the element.

[0116] The technical solutions provided in this application have been introduced in detail above. Specific examples have been used in this article to elaborate on the principles and implementation manners of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A method for predicting operation and maintenance faults based on a large model, characterized in that, Including: Obtain initial alarm data based on preset data level conditions and preset time conditions, and determine a target instruction template corresponding to the initial alarm data; Perform data preprocessing on the initial alarm data according to the target instruction template and a target large model to obtain corresponding target alarm data; Construct a target operation and maintenance fault prediction model based on a preset collaborative neural network and historical alarm data, so as to use the target operation and maintenance fault prediction model to output a target operation and maintenance fault prediction result corresponding to the target alarm data, and send the target operation and maintenance fault prediction result based on a preset notification method.

2. The operation and maintenance fault prediction method based on a large model according to claim 1, wherein The obtaining of the initial alarm data based on preset data level conditions and preset time conditions includes: Determine the data level corresponding to each alarm data in a preset data repository, and determine, based on the data level, the alarm data to be extracted that meets the preset data level conditions from the preset data repository, so as to determine the initial alarm data based on the alarm data to be extracted and the preset time conditions in the preset data repository.

3. The operation and maintenance fault prediction method based on a large model according to claim 2, wherein The determining of the initial alarm data based on the alarm data to be extracted and the preset time conditions in the preset data repository includes: Determine a preset data extraction period, and obtain the alarm data to be extracted in the preset data repository within the current time period every the preset data extraction period to obtain corresponding data to be processed; Or, obtain the alarm data to be extracted in the preset data repository in real time to obtain corresponding data to be processed; Perform filtering processing on the data to be processed based on preset filtering conditions to obtain the initial alarm data, and determine a target performance index corresponding to the initial alarm data.

4. The operation and maintenance fault prediction method based on a large model according to claim 1, wherein The determining of the target instruction template corresponding to the initial alarm data includes: Divide the initial alarm data based on preset area conditions, and determine the target instruction template corresponding to the divided initial alarm data according to preset data format conditions; Correspondingly, the performing of data preprocessing on the initial alarm data according to the target instruction template and a target large model to obtain corresponding target alarm data includes: Input the divided initial alarm data into a large model based on a preset neural network architecture, and use the large model based on the preset neural network architecture to output the corresponding target alarm data according to the target instruction template.

5. The operation and maintenance fault prediction method based on a large model according to claim 1, wherein The constructing of a target operation and maintenance fault prediction model based on a preset collaborative neural network and historical alarm data includes: Construct a preset embedding layer of the target operation and maintenance fault prediction model based on a preset multi-layer perceptron and a preset rectified linear unit, so as to use the preset embedding layer to encode input binary data into a target floating point number; Construct a preset memory layer of the target operation and maintenance fault prediction model according to a preset bias-free multi-layer perceptron, and construct a preset unflattened layer of the target operation and maintenance fault prediction model according to a first target dimension; Construct the preset neural network layer of the target operation and maintenance fault prediction model based on a preset activation function, a preset backpropagation algorithm, a preset normalization technique, and a preset number of iterations, and construct the preset flattening layer of the target operation and maintenance fault prediction model according to the second target dimension; Construct the preset output layer of the target operation and maintenance fault prediction model based on the preset unbiased multi-layer perceptron, so as to construct the target operation and maintenance fault prediction model according to the preset embedding layer, the preset memory layer, the preset non-flattening layer, the preset neural network layer, the preset flattening layer, the preset output layer, and the historical alarm data.

6. The operation and maintenance fault prediction method based on a large model according to claim 5, wherein, The constructing the target operation and maintenance fault prediction model according to the preset embedding layer, the preset memory layer, the preset non-flattening layer, the preset neural network layer, the preset flattening layer, the preset output layer, and the historical alarm data includes: Construct an initial operation and maintenance fault prediction model according to the preset embedding layer, the preset memory layer, the preset non-flattening layer, the preset neural network layer, the preset flattening layer, and the preset output layer; Determine the target gradient based on a preset cross-entropy loss function, a preset error backpropagation algorithm, the historical alarm data, and the initial operation and maintenance fault prediction model; Use a preset optimizer to update the model parameters corresponding to the initial operation and maintenance fault prediction model according to a preset modified weight decay technique and the target gradient to obtain the target operation and maintenance fault prediction model.

7. The operation and maintenance fault prediction method based on a large model according to any one of claims 1 to 6, characterized in that, The sending the target operation and maintenance fault prediction result based on a preset notification method includes: Receive in real time the target operation and maintenance fault prediction result corresponding to the target alarm data output by the target operation and maintenance fault prediction model based on a preset data interface; Use the preset data interface to send the target operation and maintenance fault prediction result according to a preset hierarchical notification policy and the preset notification method, and receive the target sending result corresponding to the target operation and maintenance fault prediction result, so as to process the target sending result based on a preset error handling mechanism.

8. An operation and maintenance fault prediction device based on a large model, characterized in that, including: A target instruction template determination module, configured to obtain initial alarm data based on a preset data level condition and a preset time condition, and determine a target instruction template corresponding to the initial alarm data; A target alarm data determination module, configured to perform data preprocessing on the initial alarm data according to the target instruction template and a target large model to obtain corresponding target alarm data; A target operation and maintenance fault prediction result output module, configured to construct a target operation and maintenance fault prediction model based on a preset collaborative neural network and historical alarm data, so as to output a target operation and maintenance fault prediction result corresponding to the target alarm data by using the target operation and maintenance fault prediction model, and send the target operation and maintenance fault prediction result based on a preset notification method.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory; wherein, the memory is used to store a computer program, and the computer program is loaded and executed by the processor to implement the operation and maintenance fault prediction method based on a large model according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that For storing a computer program, which, when executed by a processor, implements the operation and maintenance fault prediction method based on a large model according to any one of claims 1 to 7.

Images