Fault prediction method, device, computer equipment and storage medium

Abstract

The present invention relates to the field of fault prediction technology, and discloses a fault prediction method, device, computer equipment and storage medium, the method comprising: determining the time series data of the target data transmission unit; inputting the time series data into a pre-built fault prediction model to obtain a fault prediction result; the fault prediction model is determined by training the pre-trained model in the first stage and the second stage in sequence; the first stage is used to adjust the hyperparameter values ​​of the pre-trained model and the weight values ​​of the rule items of the loss function to obtain a first prediction model; the second stage is used to train the first prediction model according to the introduced new labeled data set, and adjust the parameter values ​​of the parameters of the first prediction model. In the first stage of the present invention, the parameters of the model are adjusted through prior knowledge and constraint rules, and further optimized in the second stage, which can improve the generalization ability of the model and the accuracy of fault prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of fault prediction, and in particular to a fault prediction method, device, computer equipment and storage medium. Background Art

[0002] With the continuous advancement of technology, especially the rapid development of the Internet of Things and artificial intelligence technology, the amount of data transmission is increasing. Therefore, it is very important to predict whether the data transmission unit (DTU) will fail, which is directly related to the stability, efficiency, safety and cost control of the system.

[0003] In the related art, a series of rules and thresholds are usually set to determine whether the working state of the data transmission unit is abnormal, such as timeout, packet loss rate exceeding the threshold, etc. This method requires manual setting of rules, and it is difficult to accurately predict complex faults, and the accuracy is low.

[0004] Therefore, how to improve the accuracy of data transmission unit failure prediction becomes a technical problem that technical personnel in this field need to solve urgently. Summary of the invention

[0005] In view of this, the present invention provides a fault prediction method, apparatus, computer equipment and storage medium to solve the problem of low accuracy of data transmission unit fault prediction in the related art.

[0006] In a first aspect, the present invention provides a fault prediction method, comprising:

[0007] Determine the timing data of the target data transmission unit;

[0008] Inputting the time series data of the target data transmission unit into a pre-built fault prediction model to obtain a fault prediction result of the target data transmission unit;

[0009] Among them, the fault prediction model is determined by training the pre-trained model in the first stage and the second stage in sequence; the first stage is used to adjust the parameter values ​​of the parameters of the pre-trained model, and the parameter values ​​of the parameters of the pre-trained model include: the hyperparameter values ​​of the pre-trained model and the weight values ​​of the rule items of the loss function; the pre-trained model is trained in the first stage to obtain a first prediction model; the second stage is used to train the first prediction model according to the data set and adjust the parameter values ​​of the parameters of the first prediction model; the data set is an introduced new labeled data set.

[0010] In an optional implementation, before inputting the time series data of the target data transmission unit into a pre-built fault prediction model to obtain a fault prediction result of the target data transmission unit, the method further includes:

[0011] Determine an initial training model, and train the initial training model to determine the pre-training model;

[0012] Performing the first stage of training on the pre-trained model to obtain a first prediction model;

[0013] The second stage of training is performed on the first prediction model to obtain the fault prediction model.

[0014] In an optional implementation, the performing the first stage of training on the pre-trained model to obtain the first prediction model includes:

[0015] Determine prior knowledge and constraints;

[0016] Determine a rule item according to the prior knowledge, and add the rule item to the loss function;

[0017] When performing the first stage of training on the pre-trained model, determining a performance indicator of the pre-trained model;

[0018] When the performance index of the pre-trained model does not meet the first preset condition, adjusting the hyperparameter value of the pre-trained model and the weight value of the rule item of the loss function according to the constraint rule until the performance index of the pre-trained model meets the first preset condition;

[0019] When the performance indicator of the pre-trained model meets the first preset condition, the pre-trained model is determined as the first prediction model.

[0020] In an optional implementation, the first stage is also used to adjust the structure of the pre-trained model, and the training of the pre-trained model in the first stage to obtain the first prediction model includes:

[0021] Adjusting the structure of the pre-trained model by replacing the last several layers of the pre-trained model with new layers and freezing the remaining layers of the pre-trained model except the last several layers;

[0022] The adjusted pre-trained model is trained to obtain a first prediction model.

[0023] In an optional implementation, the performing the second stage of training on the first prediction model to obtain the fault prediction model includes:

[0024] Evaluating the performance indicators of the first prediction model to determine parameters to be optimized;

[0025] The first prediction model is trained according to the data set, and the parameter values ​​of the parameters to be optimized of the first prediction model are adjusted to obtain the fault prediction model.

[0026] In an optional implementation, the training of the first prediction model according to the data set and adjusting the parameter value of the parameter to be optimized of the first prediction model to obtain the fault prediction model includes:

[0027] Determine target prior knowledge and target constraint rules; the target prior knowledge is the re-determined prior knowledge; the target constraint rules are the re-determined constraint rules;

[0028] Determine a target rule item according to the target prior knowledge, and add the target rule item to the loss function; the target rule item is a rule item re-determined according to the target prior knowledge;

[0029] Replacing the rule term of the loss function with the target rule term;

[0030] When training the first prediction model according to the data set, determining a performance indicator of the first prediction model;

[0031] When the performance index of the first prediction model does not meet the second preset condition, adjusting the parameter value of the parameter to be optimized of the first prediction model according to the target constraint rule until the performance index of the first prediction model meets the second preset condition; the parameter value of the parameter to be optimized of the first prediction model includes: the hyperparameter value of the first prediction model or the weight value of the target rule item of the loss function;

[0032] When the performance indicator of the first prediction model meets the second preset condition, the first prediction model is determined as the fault prediction model.

[0033] In an optional implementation, the determining the initial training model and training the initial training model to determine the pre-training model include:

[0034] Get historical time series data;

[0035] Determine the working status corresponding to the historical time series data;

[0036] The historical time series data is labeled according to the working state corresponding to the historical time series data to obtain historical labeled data; wherein the working state corresponding to the historical time series data includes: a fault state or a normal state;

[0037] The historical annotated data is used as a training set to train the initial training model to determine the pre-training model.

[0038] In a second aspect, the present invention provides a fault prediction device, comprising:

[0039] A first processing module, used to determine the timing data of the target data transmission unit;

[0040] A second processing module, used for inputting the time series data of the target data transmission unit into a pre-built fault prediction model to obtain a fault prediction result of the target data transmission unit;

[0041] Among them, the fault prediction model is determined by training the pre-trained model in the first stage and the second stage in sequence; the first stage is used to adjust the parameter values ​​of the parameters of the pre-trained model, and the parameter values ​​of the parameters of the pre-trained model include: the hyperparameter values ​​of the pre-trained model and the weight values ​​of the rule items of the loss function; the pre-trained model is trained in the first stage to obtain a first prediction model; the second stage is used to train the first prediction model according to the data set and adjust the parameter values ​​of the parameters of the first prediction model; the data set is an introduced new labeled data set.

[0042] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the fault prediction method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0043] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the fault prediction method of the first aspect or any corresponding embodiment thereof.

[0044] The technical solution provided by the present invention has the following technical effects:

[0045] The present invention adjusts the parameters of the model in combination with prior knowledge, constraint rules and performance indicators of the model, so that the model can better learn the fault characteristics in the data and improve the accuracy of fault prediction. In the present invention, by performing the first stage of training on the pre-trained model, a first prediction model is obtained, and the first prediction model is adjusted on the basis of the pre-trained model to further improve the performance of the model and the accuracy of fault prediction. Finally, by performing the second stage of training on the first prediction model, a fault prediction model is obtained, and the fault prediction model is further optimized on the basis of the first prediction model to improve the generalization ability of the model and the accuracy of fault prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0047] Figure 1 is a flow chart of a fault prediction method according to an embodiment of the present invention;

[0048] Figure 2 is a flow chart of another fault prediction method according to an embodiment of the present invention;

[0049] Figure 3 is a structural block diagram of a fault prediction device according to an embodiment of the present invention;

[0050] Figure 4 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0051] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0052] In the related art, whether a data transmission unit fails is usually predicted by the following method.

[0053] Rule-based method: By setting a series of rules and thresholds, it is determined whether the working status of the data transmission unit is abnormal, such as timeout, packet loss rate exceeding the threshold, etc. This method is simple and intuitive, but it requires manual setting of rules and is difficult to accurately predict complex faults.

[0054] Methods based on statistical analysis: predict future failures by analyzing the statistical characteristics of historical data, such as time series analysis, regression analysis, etc. This method has high data requirements and cannot capture complex nonlinear relationships.

[0055] Machine learning-based methods: Use traditional machine learning algorithms such as decision trees and support vector machines to model and predict the failure of data transmission units. This method requires manual feature extraction and is not flexible enough for complex data feature extraction.

[0056] Therefore, the embodiments of the present invention provide a fault prediction method, apparatus, computer device and storage medium to solve the above problems.

[0057] According to an embodiment of the present invention, a fault prediction method embodiment is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0058] Figure 1 4 is a flow chart of a fault prediction method according to an embodiment of the present invention.

[0059] like Figure 1 As shown, a fault prediction method provided in an embodiment of the present invention is applied to a data transmission unit to predict whether a fault occurs in the data transmission unit.

[0060] The fault prediction method includes:

[0061] S101: Determine the timing data of the target data transmission unit.

[0062] In this embodiment, the target data transmission unit is a data transmission unit for which it is to be determined whether a failure occurs.

[0063] In this embodiment, it is necessary to collect data information of the target data transmission unit, and store the data information of the target data transmission unit in the form of time series to obtain the time series data of the target data transmission unit. The data information of the target data transmission unit includes: acquisition time, real-time data, control parameters, operating parameters and communication data, etc.

[0064] The acquisition time indicates the time for acquiring time series data, such as the time for acquiring control parameters.

[0065] Real-time data may include data collected in real time by sensors such as temperature, humidity, pressure, current, and voltage.

[0066] The control parameters may include data that change over time, such as set temperature, set pressure, set speed, etc. The control parameters are used to monitor the operating state of the target data transmission unit and control the operation of the target data transmission unit.

[0067] The operating parameters may include the on / off time, fault alarm records, maintenance records, etc. of the target data transmission unit. By recording the operating parameters, various events and operations during the operation of the target data transmission unit can be recorded for fault diagnosis and prediction.

[0068] Communication data refers to the communication data between the target data transmission unit and other data transmission units or systems. Communication data may include data transmission rate, communication delay, packet loss rate, etc., which helps to monitor the stability and efficiency of data communication.

[0069] S102: Inputting the time series data of the target data transmission unit into a pre-built fault prediction model to obtain a fault prediction result of the target data transmission unit.

[0070] In this embodiment, the fault prediction model is determined by training the pre-trained model in the first stage and the second stage in sequence. The first stage is used to adjust the parameter values ​​of the parameters of the pre-trained model, and the parameter values ​​of the parameters of the pre-trained model include: the hyperparameter values ​​of the pre-trained model and the weight values ​​of the rule items of the loss function. The pre-trained model is trained in the first stage to obtain a first prediction model. The second stage is used to train the first prediction model according to the data set and adjust the parameter values ​​of the parameters of the first prediction model.

[0071] In this embodiment, the rule item is used to limit the range learned by the fault prediction model during the training process, which can be the range of time series data, for example, it can be the range of temperature when the target data transmission unit is in a normal state.

[0072] In this embodiment, the loss function is:

[0073]

[0074] Among them, L represents the loss function, N represents the number of training sets, and y i Indicates the real working status label corresponding to the i-th historical time series data, Represents the predicted value corresponding to the i-th historical time series data.

[0075] In this embodiment, the fault prediction result of the target data transmission unit is equivalent to the working state of the target data transmission unit. If the fault prediction result of the target data transmission unit is fault, it means that the target data transmission unit is in a fault state. If the fault prediction result of the target data transmission unit is normal, it means that the target data transmission unit is in a normal state. In this embodiment, when the fault prediction result of the target data transmission unit is fault, the target data transmission unit can be repaired in time to prevent data loss.

[0076] Before inputting the timing data of the target data transmission unit into a pre-built fault prediction model to obtain the fault prediction result of the target data transmission unit, it is necessary to train the initial training model to obtain the final fault prediction model. Therefore, before inputting the timing data of the target data transmission unit into a pre-built fault prediction model to obtain the fault prediction result of the target data transmission unit, the fault prediction method also includes: a training process of the fault prediction model.

[0077] In this embodiment, the fault prediction model is a model built based on a convolutional neural network.

[0078] Figure 2 FIG. 1 is a flow chart of another fault prediction method according to an embodiment of the present invention. Figure 2 As shown, the fault prediction method also includes a fault prediction model training method of S201-S203.

[0079] The fault prediction method includes:

[0080] S201: Determine an initial training model, and train the initial training model to determine a pre-training model.

[0081] In this embodiment, the initial training model can be a model trained on a large data set. For example, a model trained on an ImageNet data set can be determined as the initial training model. The initial training model has learned rich feature representations and can be applied to different tasks. The advantage of using the initial training model is that the features it has learned can be used to reduce the time and computational cost of training from scratch on fault prediction. The pre-trained model determined by training the initial training model is obtained by training based on historical time series data on the basis of the initial training model. The initial training model is trained on a large number of general data sets and can obtain the ability to extract multiple features. In this way, the pre-trained model determined by training based on the initial training model also has the ability to extract multiple features, which not only reduces the cost of model training, improves training efficiency, but also improves the accuracy of fault prediction.

[0082] S202: Perform a first stage of training on the pre-trained model to obtain a first prediction model.

[0083] S203: Performing second-stage training on the first prediction model to obtain a fault prediction model.

[0084] S204: Determine the timing data of the target data transmission unit. For details, refer to the relevant description of S101, which will not be repeated here.

[0085] S205: Input the time series data of the target data transmission unit into the pre-built fault prediction model to obtain the fault prediction result of the target data transmission unit. For details, refer to the relevant description of S102, which will not be repeated here.

[0086] In an optional implementation, S202 performs a first phase of training on the pre-trained model to obtain a first prediction model, including:

[0087] S2021: Determine prior knowledge and constraints.

[0088] S2022: Determine the rule item based on prior knowledge, and add the rule item to the loss function.

[0089] In this embodiment, the prior knowledge can be the knowledge and experience of the field, or the fault characteristics in the process of training the initial training model to determine the pre-training model can be used as the prior knowledge. The rule items can be defined according to the prior knowledge, such as the identification of the working state of the data transmission unit, the threshold of the time series data, etc. The rule items can be regularization items or constraints, and the learning space of the pre-training model can be limited through the rule items. The defined rule items are added to the loss function and combined with the original loss function.

[0090] S2023: When performing the first phase of training on the pre-trained model, determine the performance indicators of the pre-trained model.

[0091] In this embodiment, the performance indicators of the pre-trained model may include fault prediction accuracy, average prediction error, fit, etc.

[0092] S2024: When the performance indicator of the pre-trained model does not meet the first preset condition, adjust the hyperparameter value of the pre-trained model and the weight value of the rule item of the loss function according to the constraint rule until the performance indicator of the pre-trained model meets the first preset condition.

[0093] In this embodiment, the hyperparameters that need to be adjusted among all the hyperparameters of the pre-trained model can be comprehensively determined based on the performance indicators, prior knowledge and constraint rules of the pre-trained model. The hyperparameters that need to be adjusted among all the hyperparameters of the pre-trained model can be determined as the hyperparameters to be adjusted. The parameter values ​​of the hyperparameters to be adjusted of the pre-trained model and the weight values ​​of the rule items of the loss function can be adjusted according to the constraint rules.

[0094] In this embodiment, the hyperparameters may include parameters such as learning rate, optimizer selection, loss function selection, batch size, etc. The training process is optimized by selecting and adjusting the hyperparameters of the pre-trained model, such as learning rate, batch size, etc. The complexity of the pre-trained model can be controlled and the generalization ability can be improved by adding a rule term to the loss function and adjusting its weight.

[0095] S2025: When the performance indicator of the pre-trained model meets the first preset condition, determine the pre-trained model as the first prediction model.

[0096] In this embodiment, when the performance indicator of the pre-trained model meets the first preset condition, the first stage of training the pre-trained model is terminated, and the final pre-trained model is determined as the first prediction model.

[0097] In this embodiment, a corresponding first preset condition may be set for the performance indicator of the pre-trained model, and the first preset condition may be set and modified according to actual needs.

[0098] In the present invention, in the first stage, in addition to the scheme of adjusting the parameter values ​​of the parameters of the pre-trained model through prior knowledge and constraint rules to obtain the first prediction model as shown in the above S2021-S2025, the first stage can also be used to adjust the structure of the pre-trained model. In the first stage, the structure of the pre-trained model can also be adjusted to perform the first stage of training on the pre-trained model to obtain the first prediction model.

[0099] In an optional implementation, the above-mentioned solution of adjusting the structure of the pre-trained model and performing the first stage of training on the pre-trained model to obtain the first prediction model includes:

[0100] The structure of the pre-trained model is adjusted by replacing the last several layers of the pre-trained model with new layers and freezing the remaining layers of the pre-trained model except the last several layers. The adjusted pre-trained model is trained to obtain a first prediction model.

[0101] In this embodiment, the structure of the pre-trained model can be adjusted by data migration, and specifically, the structure of the pre-trained model can be adjusted by replacement, freezing, unfreezing, and fine-tuning.

[0102] In this embodiment, the pre-trained model (such as a convolutional neural network model trained on ImageNet) has learned some common features, such as edges, textures, shapes, etc. These features are very useful for many visual tasks (such as image classification, object detection). However, the output layer of the pre-trained model is designed for the original task (such as ImageNet's 1000-class classification task). The goal of the fault prediction task of the data transmission unit is to predict the working status of the data transmission unit through time series data, which is different from the original task of the pre-trained model. Therefore, the output layer of the model needs to be modified to match the output requirements of the new task. Therefore, when the pre-trained model is trained in the first phase, the structure of the pre-trained model needs to be adjusted.

[0103] In this embodiment, the last several layers of the pre-trained model are output layers or classification layers. The structure of the pre-trained model can be adjusted by replacing the last several layers of the pre-trained model with new layers and freezing some or all of the remaining layers of the pre-trained model except the last several layers.

[0104] As an example, the last layer of a pre-trained model is usually a fully connected layer, which is used to map the previous features to a specific number of classification categories. For example, when ResNet is trained on ImageNet, the last layer is a fully connected layer that outputs a probability distribution of 1000 categories.

[0105] As an example, the structure of the pre-trained model can be adjusted by replacing the fully connected layer of the pre-trained model with a new fully connected layer, while freezing some or all of the remaining layers of the pre-trained model except the fully connected layer. In this embodiment, if the fault prediction task is a binary classification problem, a fully connected layer with an output of 2 can be added. If it is a multi-classification problem, the corresponding fully connected layer is added according to the number of categories. Freezing some or all of the remaining layers of the pre-trained model except the fully connected layer can retain the features learned in the pre-trained model, and some or all of the convolutional layers of the pre-trained model can be frozen, and only the newly added fully connected layer can be trained.

[0106] In this embodiment, on the basis of freezing some or all of the remaining layers of the pre-trained model except the last several layers, some of the top layers or all of the layers can be unfrozen, and the final first prediction model can be better adapted to the new task through fine-tuning.

[0107] In this embodiment, the adjusted pre-trained model can be trained using the data set, and the parameters of the pre-trained model can be updated through back propagation and an optimizer to obtain a first prediction model.

[0108] In the present invention, the first stage of training the pre-trained model is carried out on the basis of the above embodiment, mainly including the above two methods: the first method is to adjust the parameter values ​​of the parameters of the pre-trained model through prior knowledge and constraint rules. The second method is to adjust the parameter values ​​of the parameters of the pre-trained model by re-training by adjusting the structure of the pre-trained model.

[0109] In the present invention, the above two methods can be applied in combination. For example, the first method can be applied on the basis of the second method. By applying the above two methods in combination, the training speed of the model and the accuracy of fault prediction can be improved.

[0110] In the present invention, the first stage of training the pre-trained model may also include other methods, such as adjusting the optimization strategy, such as using a specific optimizer, to train the pre-trained model in the first stage to obtain a first prediction model. In the present invention, the method of adjusting the optimization strategy can be applied in combination with the first method and the second method mentioned above. The purpose of the present invention is to improve the accuracy of model fault prediction by training the pre-trained model in the first stage by combining multiple methods.

[0111] In an optional implementation, the above S203 performs a second phase of training on the first prediction model to obtain a fault prediction model, specifically including:

[0112] S2031: Evaluate the performance indicators of the first prediction model to determine the parameters to be optimized.

[0113] In this embodiment, the performance indicators of the first prediction model, such as fault prediction accuracy, loss, etc., can be evaluated through the validation set to determine the parameters to be optimized. Specifically, the parameters to be optimized can be determined when the performance indicators of the first prediction model do not meet the second preset condition.

[0114] S2032: Training a first prediction model according to the data set, and adjusting parameter values ​​of parameters to be optimized of the first prediction model to obtain a fault prediction model.

[0115] In an optional implementation, the above S2032 trains the first prediction model according to the data set, and adjusts the parameter values ​​of the parameters to be optimized of the first prediction model to obtain the fault prediction model, specifically including:

[0116] When new prior knowledge and new constraint rules are introduced, steps Sa1-Sa2 are executed to redefine the rule items and constraint rules, and then steps Sa3-Sa6 are executed to adjust the parameter values ​​of the parameters to be optimized of the first prediction model according to the newly determined target constraint rules.

[0117] Sa1: Determine target prior knowledge and target constraint rules.

[0118] In this embodiment, the target prior knowledge is the re-determined prior knowledge, and the target constraint rule is the re-determined constraint rule.

[0119] Sa2: Determine the target rule term based on the target prior knowledge and add the target rule term to the loss function.

[0120] In this embodiment, the target rule item is a rule item that is re-determined based on target prior knowledge.

[0121] Sa3: Replace the rule term of the loss function with the target rule term.

[0122] Sa4: When training the first prediction model according to the data set, determine a performance indicator of the first prediction model.

[0123] In this embodiment, the data set is a newly introduced labeled data set.

[0124] In this embodiment, if a new labeled data set is introduced in the second stage, it is necessary to adjust the parameter value of the parameter to be optimized of the first prediction model according to the performance indicator of the first prediction model corresponding to the new labeled data set.

[0125] Sa5: When the performance indicator of the first prediction model does not meet the second preset condition, adjust the parameter value of the parameter to be optimized of the first prediction model according to the target constraint rule until the performance indicator of the first prediction model meets the second preset condition.

[0126] In this embodiment, the parameter value of the parameter to be optimized of the first prediction model includes: a hyperparameter value of the first prediction model or a weight value of a target rule item of a loss function.

[0127] Sa6: When the performance indicator of the first prediction model meets the second preset condition, the first prediction model is determined as the fault prediction model.

[0128] In this embodiment, when the performance indicator of the first prediction model meets the second preset condition, the second stage of training the first prediction model is terminated, and the finally obtained first prediction model is determined as the fault prediction model.

[0129] In this embodiment, the parameters of the optimizer, such as the learning rate, may also be dynamically adjusted based on the performance indicator of the first prediction model.

[0130] In this embodiment, before the second stage of training the first prediction model, it is necessary to evaluate the performance indicators of the first prediction model obtained after the first stage adjustment through a verification set. When the performance indicators of the first prediction model do not meet the second preset conditions, the second stage of training is performed on the first prediction model. When the performance indicators of the first prediction model meet the second preset conditions, the first prediction model obtained after the first stage adjustment is determined as a fault prediction model.

[0131] In this embodiment, after the first stage of training, a new labeled data set can be introduced to train the first prediction model according to the new labeled data set, and the parameter values ​​of the parameters to be optimized of the first prediction model can be adjusted to obtain a fault prediction model. Specifically, the parameter values ​​of the parameters to be optimized of the first prediction model can be updated by forward propagation and back propagation using the new labeled data set. Combined with new prior knowledge and new constraint rules, the rule items and the weights of the rule items in the loss function are adjusted to make the model better adapt to new task requirements.

[0132] In the first and second stages of the present invention, the parameter values ​​of the parameters of the pre-trained model are adjusted or the parameter values ​​of the parameters to be optimized of the first prediction model are adjusted. Specifically, the loss can be calculated through a loss function, and the parameter values ​​of the parameters of the model are adjusted according to the loss. Loss calculation: The loss of the model on the current batch of data is calculated through forward propagation. Parameter adjustment: Using the back propagation algorithm, the model parameters are gradient updated according to the loss. This can be done through an optimizer. The process of adjusting the parameter values ​​of the parameters of the model according to the loss is repeated in each training step, by gradually adjusting the parameters of the model to minimize the loss function.

[0133] In the present invention, the method for adjusting the weight value of a rule item or a target rule item includes:

[0134] Sb1: Initialize the sampling points and construct the objective function according to the optimization goal. For example, the objective function is constructed with the minimum loss as the optimization goal. Initialize the sampling points: Randomly select several initial weight values ​​in the parameter space and calculate the corresponding objective function value.

[0135] Sb2: Build a priori model: Use Gaussian process to fit the distribution of the target function.

[0136] Sb3: Select the next sampling point: Based on the prediction of the prior model, select the next weight value that is most likely to improve the objective function value.

[0137] Sb4: Update the prior model: Update the prior model according to the results of the new sampling points.

[0138] Repeat Sb3 and Sb4: continue sampling and updating until a preset stopping condition is reached (such as the maximum number of iterations or the objective function value converges).

[0139] Sb5: Select the optimal weight: Finally, the weight value with the optimal objective function value is selected as the weight value of the rule item.

[0140] In an optional implementation, the above S201 determines the initial training model, and trains the initial training model to determine the pre-training model, including:

[0141] Get historical time series data.

[0142] Determine the working status corresponding to the historical time series data.

[0143] The historical time series data are labeled according to the working status corresponding to the historical time series data to obtain historical labeled data.

[0144] The historical annotated data is used as a training set to train the initial training model to determine the pre-training model.

[0145] In this embodiment, the working state corresponding to the historical time series data includes: a fault state or a normal state.

[0146] In this embodiment, the initial training model can be a model trained on a large data set, or a newly constructed convolutional neural network model. In the present invention, the model trained on a large data set is preferably used.

[0147] In this embodiment, the historical time series data is annotated according to the working status corresponding to the historical time series data, and the corresponding working status label is added. As an example, the historical time series data can be annotated according to the working status corresponding to the historical time series data by manual annotation, and the working status label is added. When the working status corresponding to the historical time series data is a normal state, the historical time series data is first annotated, for example, a working status label of "0" can be added. When the working status corresponding to the historical time series data is a fault state, the historical time series data is secondly annotated, for example, a working status label of "1" is added.

[0148] In this embodiment, a convolutional neural network model is constructed to design a neural network structure including an input layer, a convolution layer, a pooling layer, and a fully connected layer.

[0149] Input layer: The historical time series data of the data transmission unit is used as the input layer data. The data information of each time step can be represented as a feature vector. The historical time series data is input into the input layer of the convolutional neural network model in chronological order.

[0150] Convolutional layer 1: First, the first convolutional layer is set to capture the spatial features in the historical time series data. As an example, it can capture the spatial features in the sensor data. Set 4 3×3 convolution kernels, the step size is 1, and the ReLU activation function is used.

[0151] Pooling layer 1: To reduce the size of the feature map and retain the main features, a pooling layer is added. A maximum pooling operation is performed using a 2×2 pooling kernel with a stride of 2.

[0152] Convolutional layer 2: The second convolutional layer is used to further extract abstract features from historical time series data. Six 3×3 convolution kernels are set with a step size of 1 and a ReLU activation function.

[0153] Pooling layer 2: Also use a 2×2 pooling kernel for maximum pooling with a stride of 2.

[0154] Fully connected layer: Flatten the feature map output by the pooling layer and use it as the input of the fully connected layer. You can add a fully connected layer with 128 neurons and apply the ReLU activation function.

[0155] Output layer: The last layer is the output layer, which is a fully connected layer with 2 neurons for binary classification problems (normal / faulty). Sigmoid activation function is used for binary classification prediction.

[0156] Feature extraction: The local features in the historical time series data are extracted through the convolution layer, and then the pooling layer is used to reduce the dimension and retain the most significant features. The first convolution layer contains 4 3×3 convolution kernels with a step size of 1 and a ReLU activation function. These convolution kernels will slide through the input historical time series data to extract features from local areas. Next, a 2×2 maximum pooling operation with a step size of 2 is applied to the feature map output by the convolution layer. This can reduce the size of the feature map by half while retaining the most significant features. Through such a feature extraction process, the convolutional neural network model can effectively learn local features from historical time series data, while reducing the data dimension, retaining important information, and providing more accurate and effective feature representation for subsequent fault prediction tasks. This feature extraction method that combines convolution layers and pooling layers helps to improve the model's ability to predict faults.

[0157] Activation function: Introduce nonlinear activation functions such as the ReLU activation function to enhance the expressiveness of the model, so that it can better learn complex features and patterns. The structure of the convolution layer and the pooling layer is set, and the ReLU activation function is introduced after each convolution layer. The role of the ReLU activation function is to introduce nonlinearity in the convolutional neural network, allowing the model to learn more complex feature representations. When the feature map output by the convolution layer passes through the ReLU activation function, the negative value will be converted to zero, while the positive value remains unchanged. This nonlinear conversion can help the model better adapt to the nonlinear relationship of the data and improve the fitting and expression capabilities of the model. By introducing the ReLU activation function, the convolutional neural network model can better learn the complex features in the historical time series data and improve the accuracy of fault prediction. The use of this nonlinear activation function helps to accelerate the convergence process of the model and improve the generalization ability of the model, so that it can be more effectively applied to the fault prediction task of the data transmission unit.

[0158] In this embodiment, the combination of the above-mentioned multi-layer convolutional layers and activation functions can help the model learn more complex patterns and rules in historical time series data and improve the accuracy of prediction. Therefore, it can better capture the complex nonlinear relationships in historical time series data, which has advantages over traditional statistical analysis methods. This solves the problem that statistical analysis-based methods have high data requirements and cannot capture complex nonlinear relationships.

[0159] In this embodiment, the time correlation and sequence pattern of historical time series data can be captured through convolution operations.

[0160]

[0161] Among them, Z[i] represents the convolution result of the i-th layer output, W[j] represents the j-th convolution kernel, X[i+j] represents the local window of the input data, b represents the bias term, and k represents the k×k convolution kernel.

[0162] In this embodiment, the parameter values ​​of the parameters of the convolutional neural network model can be updated by the back propagation algorithm until the specified number of iterations is reached or the loss converges, so that the prediction result is close to the true label. An optimizer can be used to adjust the parameter values ​​of the parameters of the convolutional neural network model to reduce the loss function.

[0163] The loss function is:

[0164]

[0165] Among them, L represents the loss function, N represents the number of training sets, and y i Indicates the real working status label corresponding to the i-th historical time series data, Represents the predicted value corresponding to the i-th historical time series data.

[0166] In this embodiment, the convolutional neural network model is trained using existing labeled historical time series data to learn the patterns and rules in the historical time series data, thereby improving the accuracy of fault prediction.

[0167] In the present invention, the loss function of the introduced rule term is:

[0168]

[0169] Among them, R(W) represents the rule item, and λ represents the weight value of the rule item.

[0170] In the present invention, the optimization of the model in the first stage and the second stage mainly includes the following aspects:

[0171] Optimize the loss function: It includes the combination of the original loss function and the rule term. By optimizing this combined loss function, the model parameters are updated.

[0172] Adjust the weight value of the rule item: Dynamically adjust the weight value of the rule item according to the training results of the model to balance the fitting ability and generalization ability of the model.

[0173] Application of transfer learning: Use pre-trained models and adapt them to new task requirements by replacing, freezing, unfreezing, and fine-tuning.

[0174] Optimizer and hyperparameter tuning: Choose the appropriate optimizer and hyperparameter settings to ensure that the model training process is stable and efficient.

[0175] In the present invention, a pre-trained convolutional neural network model (pre-trained model) is used for transfer learning, and fault prediction can be performed based on a small amount of labeled time series data, without the need to use a large amount of labeled time series data. By fine-tuning the pre-trained model, the performance of the statistical analysis method can be improved.

[0176] In the present invention, transfer learning is applied, and the pre-trained model is used to train the pre-trained model according to historical time series data. The pre-trained model has good feature extraction capabilities. Freezing the parameter values ​​of the parameters of the pre-trained model can keep the learned features unchanged. In order to adapt to the fault prediction task of the data transmission unit, the fully connected layer (output layer or classification layer) of the pre-trained model is replaced to adapt to the output requirements of the fault prediction task. Unfreeze some of the top or all layer parameters of the pre-trained model, and fine-tune the model through the introduction of a new labeled data set to adapt to the fault prediction task. The adjusted pre-trained model can be trained with the data set, and the parameters of the pre-trained model can be updated through back propagation and an optimizer to obtain a first prediction model. Select an optimizer and set hyperparameters such as the learning rate.

[0177] The present invention integrates prior knowledge and constraint rules into the model, and by optimizing the loss function (original loss function and rule terms), applying the back propagation algorithm, the parameter values ​​of the model parameters are updated on the basis of considering both the prediction accuracy and the constraint rules, and the fitting ability and generalization ability of the model are balanced. Adding rule terms to the loss function can enhance the robustness and interpretability of the model. The rule terms require that the features learned by the model are within a certain range. For example, the temperature features learned by the model fluctuate within a certain range to enhance the robustness of the model. In order to introduce rule terms, rule terms can be added to the loss function so that the model considers the constraint conditions of the features during the training process. As an example, taking the mean square error as the loss function, the rule terms require that the temperature features fluctuate within a certain range and do not exceed a threshold.

[0178] The technical effects that can be achieved by the present invention are:

[0179] Can improve prediction accuracy: The technical solution of the present invention can more effectively learn the features and patterns in the timing data of the data transmission unit by introducing advanced technical means such as convolutional neural networks, thereby improving the accuracy and precision of fault prediction.

[0180] Reducing human intervention: The technical solution of the present invention can utilize automated technical means to reduce reliance on manually set rules and manual feature extraction, reduce human errors, and improve the degree of automation of the system.

[0181] Processing complex relationships: The technical solution of the present invention can better process complex data features and nonlinear relationships, improve the model's ability to identify complex fault modes, and thus more comprehensively predict fault conditions.

[0182] Improve prediction efficiency: The technical solution of the present invention can speed up the prediction speed and the efficiency of model training by optimizing the algorithm and model architecture, thereby achieving more timely fault prediction and response.

[0183] Improve generalization ability: The technical solution of the present invention can enhance the generalization ability of the model, so that it can maintain good prediction performance in different data sets and environments, and improve the applicability and robustness of the model.

[0184] Improve early warning effect: The technical solution of the present invention can detect potential failure risks in advance through more accurate fault prediction, reduce the downtime of the data transmission unit, and improve production efficiency and equipment reliability.

[0185] It should be noted that the contents not described in detail in the specification of the present invention belong to the common knowledge of those skilled in the art.

[0186] In this embodiment, a fault prediction device is also provided, which is used to implement the above-mentioned embodiment and optional implementation methods, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0187] Figure 3 is a structural block diagram of a fault prediction device according to an embodiment of the present invention.

[0188] This embodiment provides a fault prediction device, such as Figure 3 As shown, the fault prediction device comprises:

[0189] The first processing module 11 is used to determine the timing data of the target data transmission unit.

[0190] The second processing module 12 is used to input the time series data of the target data transmission unit into a pre-built fault prediction model to obtain a fault prediction result of the target data transmission unit.

[0191] The fault prediction model is determined by training the pre-trained model in the first and second stages in sequence. The first stage is used to adjust the parameter values ​​of the parameters of the pre-trained model, and the parameter values ​​of the parameters of the pre-trained model include: the hyperparameter values ​​of the pre-trained model and the weight values ​​of the rule items of the loss function. The pre-trained model is trained in the first stage to obtain the first prediction model. The second stage is used to train the first prediction model according to the data set and adjust the parameter values ​​of the parameters of the first prediction model. The data set is a newly introduced labeled data set.

[0192] In an optional implementation, the fault prediction device further includes: a training module.

[0193] The training module is used to input the timing data of the target data transmission unit into the pre-built fault prediction model in the second processing module 12, and before obtaining the fault prediction result of the target data transmission unit, determine the initial training model, train the initial training model to determine the pre-training model. Perform the first stage of training on the pre-training model to obtain the first prediction model. Perform the second stage of training on the first prediction model to obtain the fault prediction model.

[0194] In an optional implementation, the training module includes a first processing unit.

[0195] The first processing unit is used to determine prior knowledge and constraint rules.

[0196] The rule items are determined based on prior knowledge and added to the loss function.

[0197] When performing the first phase of training on the pre-trained model, the performance indicators of the pre-trained model are determined.

[0198] When the performance indicator of the pre-trained model does not meet the first preset condition, the hyperparameter value of the pre-trained model and the weight value of the rule item of the loss function are adjusted according to the constraint rule until the performance indicator of the pre-trained model meets the first preset condition.

[0199] When the performance indicator of the pre-trained model meets the first preset condition, the pre-trained model is determined as the first prediction model.

[0200] In an optional implementation, the training module further includes a second processing unit.

[0201] The second processing unit is used to adjust the structure of the pre-trained model by replacing the last several layers of the pre-trained model with new layers and freezing the remaining layers of the pre-trained model except the last several layers.

[0202] The adjusted pre-trained model is trained to obtain a first prediction model.

[0203] In an optional implementation, the training module further includes a third processing unit.

[0204] The third processing unit includes a first processing sub-unit and a second processing sub-unit.

[0205] The first processing subunit is used to evaluate the performance index of the first prediction model to determine the parameters to be optimized.

[0206] The second processing subunit is used to train the first prediction model according to the data set, and adjust the parameter values ​​of the parameters to be optimized of the first prediction model to obtain a fault prediction model.

[0207] In an optional implementation, the third processing unit includes a second processing subunit, which is specifically used to determine target prior knowledge and target constraint rules. The target prior knowledge is the re-determined prior knowledge. The target constraint rules are the re-determined constraint rules.

[0208] The target rule item is determined based on the target prior knowledge and added to the loss function. The target rule item is the rule item re-determined based on the target prior knowledge.

[0209] Replace the rule term of the loss function with the target rule term.

[0210] When training the first prediction model according to the data set, a performance indicator of the first prediction model is determined.

[0211] When the performance index of the first prediction model does not meet the second preset condition, the parameter value of the parameter to be optimized of the first prediction model is adjusted according to the target constraint rule until the performance index of the first prediction model meets the second preset condition. The parameter value of the parameter to be optimized of the first prediction model includes: the hyperparameter value of the first prediction model or the weight value of the target rule item of the loss function.

[0212] When the performance indicator of the first prediction model meets the second preset condition, the first prediction model is determined as the fault prediction model.

[0213] In an optional implementation, the training module further includes a fourth processing unit.

[0214] The fourth processing unit is used to obtain historical time series data.

[0215] Determine the working status corresponding to the historical time series data.

[0216] The historical time series data is labeled according to the working state corresponding to the historical time series data to obtain historical labeled data, wherein the working state corresponding to the historical time series data includes: a fault state or a normal state.

[0217] The historical annotated data is used as a training set to train the initial training model to determine the pre-training model.

[0218] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0219] The fault prediction device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0220] The embodiment of the present invention also provides a computer device having the above Figure 3 The fault prediction device shown.

[0221] See also Figure 4 , Figure 4 Schematic diagram of the hardware structure of the computer device according to the embodiment of the present invention. Figure 4 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In an optional embodiment, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 4 A processor 10 is taken as an example.

[0222] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0223] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0224] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function. The data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage devices. In an optional embodiment, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0225] The memory 20 may include a volatile memory, such as a random access memory. The memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive. The memory 20 may also include a combination of the above-mentioned types of memory.

[0226] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0227] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc. Further, the storage medium can also include a combination of the above-mentioned types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0228] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0229] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A fault prediction method, characterized in that: include: Determine the timing data of the target data transmission unit; Inputting the time series data of the target data transmission unit into a pre-built fault prediction model to obtain a fault prediction result of the target data transmission unit; Before inputting the time series data of the target data transmission unit into a pre-built fault prediction model to obtain the fault prediction result of the target data transmission unit, determining an initial training model, and training the initial training model to determine a pre-training model; Performing the first stage of training on the pre-trained model to obtain a first prediction model; The first stage of training the pre-trained model to obtain a first prediction model includes: Determine a rule item based on prior knowledge, and add the rule item to the loss function; the rule item is used to limit the learning space of the pre-trained model; When the performance index of the pre-trained model does not meet the first preset condition, adjusting the hyperparameter value of the pre-trained model and the weight value of the rule item of the loss function according to the constraint rule until the performance index of the pre-trained model meets the first preset condition; When the performance indicator of the pre-trained model meets the first preset condition, determining the pre-trained model as the first prediction model; Performing a first phase of training on the pre-trained model to obtain a first prediction model also includes: adjusting the structure of the pre-trained model by replacing the last several layers of the pre-trained model with new layers and freezing the remaining layers of the pre-trained model except the last several layers; training the adjusted pre-trained model to obtain the first prediction model; Among them, the fault prediction model is determined by training the pre-trained model in the first and second stages in sequence; the second stage is used to train the first prediction model according to the data set and adjust the parameter values ​​of the parameters of the first prediction model; the data set is an introduced new labeled data set.

2. The method according to claim 1, characterized in that Before inputting the time series data of the target data transmission unit into a pre-built fault prediction model to obtain a fault prediction result of the target data transmission unit, the method further includes: The second stage of training is performed on the first prediction model to obtain the fault prediction model.

3. The method according to claim 2, characterized in that The performing the second stage of training on the first prediction model to obtain the fault prediction model includes: Evaluating the performance indicators of the first prediction model to determine parameters to be optimized; The first prediction model is trained according to the data set, and the parameter values ​​of the parameters to be optimized of the first prediction model are adjusted to obtain the fault prediction model.

4. The method according to claim 3, characterized in that The step of training the first prediction model according to the data set and adjusting the parameter value of the parameter to be optimized of the first prediction model to obtain the fault prediction model includes: Determine target prior knowledge and target constraint rules; the target prior knowledge is the re-determined prior knowledge; the target constraint rules are the re-determined constraint rules; Determine a target rule item according to the target prior knowledge, and add the target rule item to the loss function; the target rule item is a rule item re-determined according to the target prior knowledge; Replacing the rule term of the loss function with the target rule term; When training the first prediction model according to the data set, determining a performance indicator of the first prediction model; When the performance index of the first prediction model does not meet the second preset condition, adjusting the parameter value of the parameter to be optimized of the first prediction model according to the target constraint rule until the performance index of the first prediction model meets the second preset condition; the parameter value of the parameter to be optimized of the first prediction model includes: the hyperparameter value of the first prediction model or the weight value of the target rule item of the loss function; When the performance indicator of the first prediction model meets the second preset condition, the first prediction model is determined as the fault prediction model.

5. The method according to claim 1, characterized in that The determining of the initial training model and training the initial training model to determine the pre-training model include: Get historical time series data; Determine the working status corresponding to the historical time series data; The historical time series data is labeled according to the working state corresponding to the historical time series data to obtain historical labeled data; wherein the working state corresponding to the historical time series data includes: a fault state or a normal state; The historical annotated data is used as a training set to train the initial training model to determine the pre-training model.

6. A fault prediction device, characterized in that: include: A first processing module, used to determine the timing data of the target data transmission unit; A second processing module, used for inputting the time series data of the target data transmission unit into a pre-built fault prediction model to obtain a fault prediction result of the target data transmission unit; A training module, used for inputting the timing data of the target data transmission unit into a pre-built fault prediction model before the second processing module obtains the fault prediction result of the target data transmission unit, determining an initial training model, and training the initial training model to determine a pre-training model; Performing the first stage of training on the pre-trained model to obtain a first prediction model; The training module comprises: a first processing unit and a second processing unit; A first processing unit is used to determine a rule item according to prior knowledge and add the rule item to the loss function; the rule item is used to limit the learning space of the pre-trained model; When the performance index of the pre-trained model does not meet the first preset condition, adjusting the hyperparameter value of the pre-trained model and the weight value of the rule item of the loss function according to the constraint rule until the performance index of the pre-trained model meets the first preset condition; When the performance indicator of the pre-trained model meets the first preset condition, determining the pre-trained model as the first prediction model; The second processing unit is used to adjust the structure of the pre-trained model by replacing the last several layers of the pre-trained model with new layers and freezing the remaining layers of the pre-trained model except the last several layers; and train the adjusted pre-trained model to obtain a first prediction model; Among them, the fault prediction model is determined by training the pre-trained model in the first and second stages in sequence; the second stage is used to train the first prediction model according to the data set and adjust the parameter values ​​of the parameters of the first prediction model; the data set is an introduced new labeled data set.

7. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the steps of the method according to any one of claims 1 to 5 by executing the computer instructions.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the steps of the method according to any one of claims 1 to 5.

Images