Fault prediction method and apparatus, device, storage medium, and program product

Abstract

Disclosed are a fault prediction method and apparatus, a device, a storage medium, and a program product, relating to the technical field of cloud computing and fault prediction. First prediction information is obtained from a prediction model by means of prediction, and second prediction information is obtained by verifying a first fault prediction result on the basis of expert experience. Thus, a cloud management platform determines a target fault prediction result of a prediction object by combining prediction information obtained in two different manners. When the first fault prediction result is different from a second fault prediction result, it can be determined that there is an incorrect fault prediction result; and when the first fault prediction result is the same as the second fault prediction result, it can be determined that the first fault prediction result is correct; therefore, compared with the solution in which the first fault prediction result outputted from the model is determined as the target fault prediction result and whether the target fault prediction result is correct cannot be verified, the accuracy of the determined target fault prediction result of the prediction object is improved.

Description

A method, apparatus, device, storage medium, and program product for fault prediction.

[0001] This application claims priority to Chinese Patent Application No. 202510121502.8, filed on January 23, 2025, entitled "A Fault Prediction Method, Apparatus, Device, Storage Medium and Program Product", the entire contents of which are incorporated herein by reference. Technical Field

[0002] This application relates to the fields of cloud computing and fault prediction technology, and in particular to a fault prediction method, apparatus, device, storage medium and program product. Background Technology

[0003] Machine learning models predict faults in objects based on real-time data from infrastructure managed by a cloud management platform, and the model's output is directly used as the predicted fault outcome. However, due to the potential for false positives and false negatives, the accuracy of fault prediction is relatively poor. Summary of the Invention

[0004] This application provides a fault prediction method, apparatus, device, storage medium, and program product that can improve the accuracy of fault prediction.

[0005] Firstly, a fault prediction method is provided, applied to a cloud management platform for managing infrastructure. A prediction model runs on the infrastructure, which includes at least one prediction object. The method includes: the cloud management platform acquiring first prediction information from the prediction model; wherein the prediction model is used to predict faults in the prediction object based on data to be predicted, the data to be predicted indicating the operational data of the prediction object, and the first prediction information including a first predicted fault result. The cloud management platform verifies the first predicted fault result based on expert experience to obtain second prediction information; the second prediction information includes a second predicted fault result. The cloud management platform determines the target predicted fault result for the prediction object based on the first and second prediction information.

[0006] Based on the aforementioned fault prediction method, since the first prediction information is obtained from the prediction model, and the second prediction information is obtained by verifying the first predicted fault result based on expert experience, the cloud management platform integrates the prediction information obtained from these two different methods to determine the target predicted fault result. When the first and second predicted fault results are different, it can be determined that one of the predicted fault results is incorrect; when the first and second predicted fault results are the same, it can be determined that the first predicted fault result is correct. Compared to the scheme that directly determines the first predicted fault result output by the model as the target predicted fault result without being able to determine whether the target predicted fault result is correct, this method improves the accuracy of the determined target predicted fault result.

[0007] As one possible implementation, the cloud management platform determines the performance metrics of the prediction model based on the target predicted fault results and the actual fault results of the predicted object. If the performance metrics of the prediction model fall below a first preset threshold, the cloud management platform optimizes the prediction model. In this way, by acquiring the target predicted fault results and the actual fault results of the predicted object, the cloud management platform monitors the performance metrics of the prediction model and optimizes the prediction model when the performance metrics fall below the first preset threshold, thereby improving the accuracy of subsequent fault predictions.

[0008] As one possible implementation, the cloud management platform determines the accuracy of the second predicted fault result based on the target predicted fault result and the actual fault result of the predicted object. If the accuracy of the second predicted fault result is lower than a second preset threshold, the cloud management platform optimizes expert experience. By determining the accuracy of the second predicted fault result, the cloud management platform verifies the effectiveness of expert experience validation. Optimizing expert experience when the accuracy of the second predicted fault result is lower than the second preset threshold improves the accuracy of subsequent fault predictions.

[0009] Optionally, the cloud management platform determines sample data for the prediction object based on the first predicted failure result, the second predicted failure result, and the actual failure result. The sample data includes one or more combinations of high-confidence predicted samples, corrected error samples, false positive samples, and missed positive samples. Based on the number of sample data, the performance index of the prediction model is determined. Based on the corrected error samples, the accuracy of the second predicted failure result is determined.

[0010] The high-confidence sample is the data to be predicted that is consistent with the first predicted fault result, the second predicted fault result, and the actual fault result.

[0011] The erroneous sample is the data to be predicted where the first predicted fault result is consistent with the actual fault result, but the second predicted fault result is different from the first predicted fault result;

[0012] The false alarm sample is the first predicted fault result indicating a fault, but the second predicted fault result and the actual fault result both indicate that there is no fault in the predicted data.

[0013] The missed sample is the data to be predicted, where the first predicted fault result indicates no fault, but both the second predicted fault result and the actual fault result indicate a fault; or, if the second predicted fault result is not predicted, the missed sample is the data to be predicted, where the first predicted fault result indicates no fault, but the actual fault result indicates a fault.

[0014] Based on the above method, the cloud management platform classifies the data to be predicted. Using the classified sample data, it can promptly determine the performance metrics of the prediction model and the accuracy of the second predicted fault result, thereby improving the real-time performance of model optimization and expert experience optimization. Furthermore, the classification criteria for the data to be predicted include both actual fault results and second predicted fault results, avoiding sample contamination caused by directly applying the first predicted fault result to fault handling, ensuring the training quality of the model, and further improving the accuracy of model predictions.

[0015] Optionally, when the number of false positives exceeds a corresponding first preset threshold, the cloud management platform determines that the performance index of the prediction model is lower than the first preset threshold; and / or, when the number of false negatives exceeds a corresponding second preset threshold, the cloud management platform determines that the performance index of the prediction model is lower than the first preset threshold. In this way, the cloud management platform can determine, through the classified samples, that the model performance degradation is in accuracy and / or false positive rate, thereby enabling targeted retraining of the prediction model and improving the efficiency of model optimization.

[0016] Optionally, when the number of corrected error samples exceeds a corresponding third preset threshold, the cloud management platform determines that the accuracy of the second predicted fault result is lower than the second preset threshold. In this way, the cloud management platform can promptly determine the accuracy of expert experience based on the number of corrected error samples after classification, thereby improving the effectiveness of verification based on expert experience.

[0017] One possible implementation is through pre-defined expert rules based on expert experience. Predictions are made on the data to be predicted according to these pre-defined expert rules to determine the second prediction information. These pre-defined expert rules specify the predicted fault results and the probability of occurrence of the predicted fault results corresponding to the operational data of at least one predicted object. Thus, by determining the second prediction information through manually set pre-defined expert rules, the accuracy of the second predicted fault result is generally higher than the first predicted fault result output by the model, since these rules are based on expert experience. Furthermore, because pre-defined expert rule prediction and model prediction are two completely different prediction methods, the accuracy of the target predicted fault result determined based on the first and second prediction information is improved.

[0018] As one possible implementation method, expert experience suggests conducting online load testing. Based on the data to be predicted, online load testing is performed on the object to be predicted to determine secondary predictive information. Online load testing can more accurately determine whether a failure will occur, thereby improving the accuracy of predicting target failures.

[0019] As one possible implementation, the first prediction information also includes the probability of a first predicted failure result occurring, and the second prediction information also includes the probability of a second predicted failure result occurring. When the first and second predicted failure results differ, the cloud management platform determines the target predicted failure result based on the probabilities of occurrence of the first and second predicted failure results. In this way, the cloud management platform combines two different methods to determine the target predicted failure result of the predicted object. When the first and second predicted failure results differ, it can determine that one of the predicted failure results is incorrect, and by determining the target predicted failure result based on its probability of occurrence, the accuracy of the determined target predicted failure result is improved.

[0020] Optionally, if the first predicted fault result and the second predicted fault result differ, the target predicted fault result is the one with the higher probability of occurrence between the first and second predicted fault results. Thus, when the first and second predicted fault results differ, to avoid false alarms and missed alarms, the cloud management platform selects the predicted fault result with the higher probability of occurrence, thereby improving the accuracy of fault prediction.

[0021] Specifically, when the first predicted fault result and the second predicted fault result are different, and the first predicted fault result indicates that the predicted object is about to fail, the target predicted fault result is the second predicted fault result. Thus, for the data to be predicted where the first predicted fault result indicates that the predicted object is about to fail, to avoid false alarms, the second predicted fault result, which has higher accuracy and indicates that the predicted object will not fail, is determined as the target predicted fault result, thereby improving the accuracy of fault prediction.

[0022] Specifically, when the first predicted fault result and the second predicted fault result differ, and the second predicted fault result is obtained through online stress testing, the target predicted fault result is the second predicted fault result. When the first predicted fault result and the second predicted fault result differ, the second predicted fault result obtained through online stress testing, which has higher accuracy, is determined as the target predicted fault result, thereby improving the accuracy of fault prediction.

[0023] Where the first predicted fault result and the second predicted fault result are different, and the first predicted fault result is that the predicted object will not fail, and the second predicted fault result is obtained by re-prediction based on preset expert rules, the target predicted fault result is the predicted fault result with a higher probability of occurrence between the first predicted fault result and the second predicted fault result.

[0024] As one possible implementation, if the first predicted fault result of the data to be predicted is determined to be a predicted fault, the first predicted fault result is verified based on expert experience to obtain second predicted information. Thus, the cloud management platform only re-predicts the data to be predicted where the first predicted fault result indicates a predicted fault, thereby obtaining the second predicted information. Compared to a scheme that re-predicts all data to be predicted without distinguishing between first and second predicted fault results, this approach improves the accuracy of fault prediction while minimizing computational resource consumption.

[0025] As one possible implementation, a fault alarm is issued when the predicted fault result of the target object is a fault. Since the accuracy of the target predicted fault result determined by combining two different prediction methods is improved, the accuracy of the fault alarm based on the target predicted fault result is improved.

[0026] In a second aspect, a fault prediction apparatus is provided, the apparatus comprising modules for performing the fault prediction method in the first aspect or any possible implementation thereof.

[0027] The fault prediction device described in the second aspect can be a terminal device or a network device, or it can be a chip (system), network card or other component or assembly that can be set in a terminal device or a network device, or it can be a device that includes a terminal device or a network device. This application does not limit it in this regard.

[0028] Furthermore, the technical effects of the fault prediction device described in the second aspect can be referred to the technical effects of the fault prediction method described in the first aspect, and will not be repeated here.

[0029] Thirdly, a computing device cluster is provided, including at least one computing device, each computing device including a processor and memory;

[0030] The processor of the at least one computing device is configured to execute instructions stored in the memory of the at least one computing device to cause the cluster of computing devices to perform the operational steps of the method as described in any possible implementation of the first aspect.

[0031] Fourthly, a computer-readable storage medium is provided, including computer program instructions that, when executed by a cluster of computing devices, cause the cluster of computing devices to perform operational steps of the method as described in the first aspect or any possible implementation thereof.

[0032] Fifthly, a chip system is provided. The chip system includes a memory and at least one processor. The memory stores a set of computer instructions, which, when executed by the processor, perform the operational steps of the method described in any possible implementation of the first aspect.

[0033] In a sixth aspect, a computer program product containing instructions is provided, wherein when the instructions are run by a cluster of computing devices, the cluster of computing devices performs the operation steps of the method as described in any possible implementation of the first aspect.

[0034] Based on the implementation methods provided in the above aspects, this application can be further combined to provide more implementation methods. Attached Figure Description

[0035] Figure 1 is a schematic diagram of the architecture of a cloud management platform provided in an embodiment of this application;

[0036] Figure 2 is a flowchart illustrating a fault prediction method provided in an embodiment of this application;

[0037] Figure 3 is a schematic diagram of a fault prediction method provided in an embodiment of this application;

[0038] Figure 4 is a schematic diagram of a fault alarm process provided in an embodiment of this application;

[0039] Figure 5 is a schematic diagram of a sample classification process provided in an embodiment of this application;

[0040] Figure 6 is a schematic diagram of a model performance optimization process provided in an embodiment of this application;

[0041] Figure 7 is a schematic diagram of a fault prediction device 700 provided in an embodiment of this application;

[0042] Figure 8 is a schematic diagram of the structure of a computing device provided in an embodiment of this application;

[0043] Figure 9 is a schematic diagram of the structure of a computing device cluster provided in an embodiment of this application;

[0044] Figure 10 is a schematic diagram of a network connection between computing devices provided in an embodiment of this application. Detailed Implementation

[0045] This application can be applied not only to existing cloud service technologies and fault prediction scenarios, but also to future cloud service technologies and fault prediction scenarios.

[0046] To facilitate understanding, the relevant terms involved in the embodiments of this application will be introduced below.

[0047] (1) Online stress testing

[0048] Online load testing refers to simulating real-world usage scenarios to test the performance of the device hardware under those scenarios.

[0049] (2) Expert experience

[0050] Expert experience refers to the unique insights, skills, knowledge, and judgment acquired through long-term practice, learning, research, and exploration in a specific field or profession.

[0051] In this embodiment, expert experience can refer to preset expert rules or online stress testing of the predicted object. Preset expert rules are a rule system built based on the knowledge and experience of domain experts. For example, in this embodiment, preset expert rules refer to a set of judgment rules for the data to be predicted when experts determine whether the predicted object in the infrastructure managed by the cloud management platform will fail.

[0052] (3) Prediction Model

[0053] A predictive model is a machine learning model that is trained on historical data to predict whether a failure will occur.

[0054] For example, the prediction model can be a decision tree model, a convolutional neural network model, or a recurrent neural network model from machine learning models.

[0055] (4) Cloud Management Platform

[0056] A cloud management platform is a unified management platform for data center resources. It can manage multiple open-source or heterogeneous cloud computing technologies or products.

[0057] The cloud management platform provides unified cloud services, enabling organizations to quickly transform their existing virtual machine infrastructure into a highly scalable private cloud while making full use of public cloud resources.

[0058] (5) Infrastructure

[0059] Infrastructure refers to the hardware, software, and network resources required to support the normal operation of a cloud management platform.

[0060] (6) Prediction Object

[0061] The prediction target can be the entire infrastructure, or servers, switches, storage devices, etc. within the infrastructure, or specific hardware within servers, switches, and storage devices.

[0062] (7) Data to be predicted

[0063] The data to be predicted is the operational data of the objects to be predicted in the infrastructure.

[0064] The data to be predicted includes at least one of the following: health indicators, performance indicators, and environmental indicators of the object to be predicted.

[0065] Health metrics data include the number of times the predicted object is launched, process runtime, temperature, etc.; performance metrics data include the memory utilization, latency, remaining memory, etc.; and environmental metrics data include the temperature, humidity, etc., of the location of the predicted object.

[0066] To make the technical problem of poor accuracy in fault prediction in the background technology easier to understand, the hardware fault alarm method will be explained in conjunction with Figure 1 below.

[0067] In today's digital age, cloud services have become central to enterprise operations, offering flexible resources and efficient services. The reliability of cloud services directly impacts user experience and business operations, and hardware failures in the infrastructure managed by the cloud management platform are a major factor affecting cloud service reliability. Infrastructure hardware failures include switch failures, server failures, and storage device failures. Switch and server failures can lead to cloud service interruptions, while storage device failures can result in data loss. Therefore, hardware failures in the infrastructure can diminish user trust in the cloud management platform.

[0068] To avoid cloud service interruptions, the cloud management platform's hardware fault alarm system needs to predict potential infrastructure hardware failures and issue fault alarms, triggering subsequent fault handling operations such as hardware replacement, thereby preventing infrastructure hardware failures from occurring.

[0069] The hardware fault alarm method is applied in a cloud management platform. Figure 1 is a schematic diagram of the architecture of a cloud management platform provided in an embodiment of this application. As shown in Figure 1, the cloud management platform 100 is used to manage the infrastructure 200. The cloud management platform 100 and the infrastructure 200 run in a cloud environment. The infrastructure 200 includes at least one server, at least one switch, and at least one storage device, etc.

[0070] A predictive model runs on the infrastructure 200. This model is used to predict faults in the target object based on the data to be predicted. The infrastructure 200 includes the target object, which can be the entire infrastructure 200, servers, switches, storage devices, or specific hardware within those devices. The data to be predicted includes the operational data of the target object, specifically the operational data of the infrastructure.

[0071] The cloud management platform 100 acquires the data to be predicted uploaded by the infrastructure, inputs this data into a prediction model for fault prediction, and directly generates a maintenance order based on the predicted fault results output by the model. The maintenance order is then sent to the user for subsequent fault handling. Fault handling includes hardware replacement and hardware repair.

[0072] In the above hardware fault alarm method, the cloud management platform 100 uses a predictive model to make predictions and directly generates a maintenance order based on the predicted fault results output by the predictive model. However, since the predicted fault results may produce false alarms and missed alarms, the accuracy of fault prediction is poor.

[0073] To address the issue of poor accuracy in fault prediction, this application provides a fault prediction method, specifically a method for determining a target predicted fault result based on first predicted information output by a prediction model and second predicted information obtained by verifying the first predicted fault result using expert experience. Since the first predicted information is obtained from the prediction model, and the second predicted information is obtained by verifying the first predicted fault result using expert experience, the cloud management platform integrates the predicted information obtained from these two different methods to determine the target predicted fault result. When the first and second predicted fault results differ, it can be determined that one of the predicted fault results is incorrect; when the first and second predicted fault results are the same, it can be determined that the first predicted fault result is correct. Compared to a scheme that determines the first predicted fault result output by the model as the target predicted fault result and cannot detect whether the target predicted fault result is correct, this method improves the accuracy of the determined target predicted fault result.

[0074] The following explanation uses the cloud management platform 100 in Figure 1 as an example for fault alarm. Figure 2 is a flowchart illustrating a fault prediction method provided in this embodiment. The cloud management platform 100 in Figure 2 can be referred to the relevant description in Figure 1 above; this embodiment will not repeat the details here. As shown in Figure 2, the fault prediction method may include steps 201 to 209.

[0075] Step 201: The cloud management platform 100 obtains the first prediction information from the prediction model.

[0076] The prediction model is used to predict the failure of the object based on the data to be predicted, and the data to be predicted is used to indicate the operating data of the object to be predicted.

[0077] For example, the data to be predicted could be the operational data of the infrastructure at a certain point in time when the cloud management platform 100 performs a fault prediction. The cloud management platform 100 inputs the data to be predicted into the prediction model and obtains the first prediction information output by the prediction model. The data to be predicted could also include multiple batches of data. For each batch of data, the cloud management platform 100 inputs it into the prediction model and obtains the first prediction information output by the prediction model.

[0078] Among them, the health indicator data, performance indicator data and environmental indicator data of the object to be predicted come from different data sources. For example, the data source can be the operating system of the object to be predicted, the temperature and humidity sensor of the location of the object to be predicted, etc.

[0079] The predicted object refers to a device, device cluster, or hardware within the infrastructure managed by the cloud management platform 100. For example, if the predicted object is a device cluster, then the predicted object is the infrastructure managed by the management platform 100. Another example is an electronic device; this predicted object can be any electronic device within the infrastructure, and any electronic device is a device within the infrastructure for which the cloud management platform 100 needs to predict whether a hardware failure will occur.

[0080] The first prediction information includes a first predicted failure result. This first predicted failure result includes predictions that the object will fail and predictions that the object will not fail. When the first predicted failure result indicates that the object will fail, it may also include the failure type, such as storage unit failure or communication failure. A first predicted failure result indicating that the object will fail can also be called a predicted failure, and a first predicted failure result indicating that the object will not fail can also be called a predicted no failure.

[0081] As one possible implementation, a device containing a prediction model makes a prediction on the data to be predicted to obtain first prediction information, and the cloud management platform 100 receives the first prediction information sent by the device containing the prediction model.

[0082] The prediction model is a machine learning model capable of predicting faults. The prediction model can be an existing model or a model retrained based on sample data after sample classification; this application does not impose any limitations on this.

[0083] Step 202: The cloud management platform 100 verifies the first predicted fault result based on expert experience and obtains the second predicted information.

[0084] The second prediction information includes a second predicted failure result. The second predicted failure result indicates whether the predicted object is likely to fail or will not fail.

[0085] As one possible implementation, the cloud management platform 100 can predict the data to be predicted according to preset expert rules and determine the second prediction information of the data to be predicted.

[0086] Among them, the preset expert rule refers to the judgment rule determined by expert experience and pre-set manually. The preset expert rule is used to specify the predicted failure result and the probability of occurrence of the predicted failure result corresponding to the operating data of at least one predicted object; at least one predicted object includes the predicted object, the predicted failure result corresponding to the operating data within the preset data range is failure, and the predicted failure result corresponding to the operating data outside the preset data range is no failure. The type of preset expert rule can be confidence rule, prior rule, etc., and this application embodiment does not limit this. The probability of occurrence in the preset expert rule is pre-set by the user.

[0087] The cloud management platform 100 can be a preset expert rule that receives user input, or it can be a preset expert rule that is pre-saved in a database or the Internet. This application embodiment does not limit this.

[0088] For example, data to be predicted falling within a preset data range can mean that health indicator data is greater than a corresponding preset health indicator threshold, environmental indicator data is greater than a corresponding preset environmental indicator threshold, and performance indicator data is less than a corresponding preset performance indicator threshold. Data to be predicted falling outside a preset data range can mean that health indicator data is less than a corresponding preset health indicator threshold, environmental indicator data is less than a corresponding preset environmental indicator threshold, and performance indicator data is greater than a corresponding preset performance indicator threshold.

[0089] For example, when the health indicator data in the data to be predicted is greater than the corresponding preset health indicator threshold, and / or the environmental indicator data is greater than the corresponding preset environmental indicator threshold, and / or the performance indicator data is less than the corresponding preset performance indicator threshold, the cloud management platform 100 determines that the second predicted fault result of the data to be predicted is a predicted fault. When the health indicator data in the data to be predicted is less than the corresponding preset health indicator threshold, and the environmental indicator data is less than the corresponding preset environmental indicator threshold, and the performance indicator data is greater than the corresponding preset performance indicator threshold, the cloud management platform 100 determines that the second predicted fault result of the data to be predicted is that the predicted fault will not occur.

[0090] The above content is merely an example of determining the second prediction information of the data to be predicted based on preset expert rules, and does not constitute a limitation on the solution. The embodiments of this application may also determine the second prediction information of the data to be predicted based on other preset expert rules.

[0091] As one possible implementation, the cloud management platform 100 can also perform online stress testing on the object to be predicted based on the data to be predicted, and determine the second prediction information of the data to be predicted.

[0092] Online load testing, also known as online stress testing, refers to simulating real-world usage scenarios to test the hardware's performance under those scenarios.

[0093] For example, based on performance, health, and environmental metrics from the data to be predicted, when the cloud platform's operational data reaches the predicted data, it is determined that the cloud platform is in a simulated real-world usage scenario. The cloud management platform 100 obtains the simulation test results of the cloud platform and determines these results as the second predicted failure outcome. The model test result may indicate that a failure has occurred or not, with the probability of occurrence in the second predicted failure outcome approaching 100%.

[0094] As one possible implementation, the cloud management platform 100 could also, after obtaining the first prediction information of the data to be predicted, determine the first predicted fault result of the data to be predicted as the predicted object's expected fault, verify the first predicted fault result based on expert experience, and obtain the second prediction information. Alternatively, if the cloud management platform 100 determines that the first predicted fault result indicates the predicted object will not experience a fault, it may not verify the first predicted fault result, but instead directly determine the target predicted fault result as the first predicted fault result.

[0095] Specifically, the first predicted fault result is verified to obtain the second predicted information only if the first predicted fault result indicates a predicted fault. If the first predicted fault result indicates a predicted non-fault, the first predicted fault result is not verified. The fault alarm scheme can be referred to the relevant descriptions in Figures 3 and 4 below, and will not be elaborated here in the embodiments of this application.

[0096] Step 203: The cloud management platform 100 determines the target predicted fault result of the predicted object based on the first prediction information and the second prediction information.

[0097] As one possible implementation, when the first predicted failure result and the second predicted failure result are consistent, the cloud management platform 100 determines the target predicted failure result of the predicted object as the first predicted failure result.

[0098] When both the first and second predicted failure results indicate a predicted failure, the cloud management platform 100 determines that the target predicted failure result indicates a predicted failure. Conversely, when both the first and second predicted failure results indicate no predicted failure, the cloud management platform 100 determines that the target predicted failure result indicates no predicted failure.

[0099] As one possible implementation, the first prediction information further includes the probability of a first predicted failure result occurring, and the second prediction information further includes the probability of a second predicted failure result occurring. When the first predicted failure result and the second predicted failure result differ, the target predicted failure result is determined based on the probability of the first predicted failure result occurring and the probability of the second predicted failure result occurring.

[0100] Optionally, when the first predicted fault result and the second predicted fault result are different, the target predicted fault result is the predicted fault result with the higher probability of occurrence between the first and second predicted fault results. Alternatively, when the first and second predicted fault results are different, but the probability of occurrence of the first and second predicted fault results is the same, the target predicted fault result can be the second predicted fault result.

[0101] Specifically, when the first predicted failure result and the second predicted failure result are different, and the first predicted failure result indicates that the predicted object will not fail, and the second predicted failure result is obtained by re-prediction based on preset expert rules, the cloud management platform 100 determines the target predicted failure result for the predicted object to be the predicted failure result with the higher probability of occurrence between the first and second predicted failure results. In other words, when the first predicted failure result indicates that the predicted object will not fail, and the second predicted failure result indicates that the predicted object is expected to fail, the cloud management platform 100 determines the target predicted failure result for the predicted object to be the predicted failure result with the higher probability of occurrence between the first and second predicted failure results.

[0102] Optionally, when the first predicted fault result and the second predicted fault result are different, if the first predicted fault result is the predicted fault of the object, or the second predicted fault result is determined based on the online stress test method, the target predicted fault result is the second predicted fault result.

[0103] In cases where the first predicted failure result differs from the second predicted failure result, and the first predicted failure result indicates a predicted failure of the object, the second predicted failure result, which indicates the object will not fail, is obtained through online stress testing or preset expert rules. These preset expert rules include various scenarios where failure will not occur based on expert experience. Therefore, the probability of the second predicted failure result for the data to be predicted satisfying the preset expert rules is relatively high. For example, the probability of the second predicted failure result for the data to be predicted satisfying the preset expert rules could be 100%, 99%, 98%, or similarly high. The probability of the second predicted failure result obtained through online stress testing is also close to 100%. Since the probability of the second predicted failure result obtained through both methods is higher than the probability of the first predicted failure result from the prediction model, the cloud management platform 100 determines the target predicted failure result for the object to be the second predicted failure result. In other words, when the first predicted failure result indicates a predicted failure of the object, and the second predicted failure result indicates the object will not fail, the cloud management platform 100 determines the target predicted failure result for the object to be the object that will not fail.

[0104] In cases where the first predicted fault result differs from the second predicted fault result, and the second predicted fault result is determined based on online stress testing, the probability of the second predicted fault result obtained by online stress testing can be considered to be close to 100%. The probability of the second predicted fault result is higher than the probability of the first predicted fault result output by the prediction model. Therefore, the cloud management platform 100 determines the target predicted fault result of the predicted object as the second predicted fault result.

[0105] Step 204: When the predicted failure result of the predicted object is a failure, the cloud management platform 100 issues a fault alarm.

[0106] When the predicted failure result of the target object indicates a predicted failure, the cloud management platform 100 will output a fault alarm.

[0107] Step 205: The cloud management platform 100 determines the performance indicators of the prediction model and the accuracy of the second predicted failure result based on the target predicted failure result and the actual failure result of the predicted object. Then, it executes steps 206 and / or 208.

[0108] When the operational data of the infrastructure 200 managed by the cloud management platform 100 is the data to be predicted, the cloud management platform 100 obtains the actual failure result of the predicted object. The actual failure result is either the predicted object actually fails or the predicted object does not actually fail.

[0109] As one possible implementation, the cloud management platform 100 can test the prediction model based on a test set and determine the performance metrics of the prediction model based on the test results. Alternatively, the cloud management platform 100 can determine the loss function of the prediction model and determine the performance metrics of the prediction model based on the loss function. Alternatively, the cloud management platform 100 can perform sample classification on the test data and determine the performance metrics of the prediction model based on the classification results.

[0110] Accuracy refers to the proportion of predicted data where both the initial predicted fault result and the actual fault result indicate either a fault or no fault, out of the total number of predicted data. Recall refers to the proportion of predicted data where both the initial predicted fault result and the actual fault result indicate a fault, out of the total number of predicted data where the actual fault result indicates a fault. False alarm rate refers to the proportion of predicted data where the initial predicted fault result is a predicted fault but the actual fault result indicates no fault, out of the total number of predicted data where the actual fault result indicates no fault.

[0111] As one possible implementation, the cloud management platform 100 may determine the accuracy of the second predicted failure result by the ratio of a first number of second predicted failure results that are identical to the actual failure result and the total number of data to be predicted. Alternatively, the cloud management platform may also determine a first number of second predicted failure results that are identical to the actual failure result, and determine the reciprocal of the first number as the accuracy of the second predicted failure result.

[0112] Optionally, when the cloud management platform 100 performs sample classification on the data to be predicted, the cloud management platform 100 may determine the sample data of the prediction object based on the first predicted failure result, the second predicted failure result, and the actual failure result. The sample data includes one of the following categories: high-confidence predicted samples, error-corrected samples, false positive samples, and missed positive samples. The cloud management platform 100 determines the performance metrics of the prediction model based on the quantity of sample data. The cloud management platform 100 determines the accuracy of the second predicted failure result based on the corrected error samples.

[0113] Specifically, the cloud management platform 100 may determine at least one of the following parameters for the prediction model: accuracy, recall, and false positive rate, based on the quantity of sample data. The cloud management platform 100 may use accuracy, recall, or false positive rate as the performance metric for the prediction model; alternatively, it may obtain the performance metric by weighted averaging of at least two of the accuracy, recall, and false positive rate. Furthermore, the cloud management platform 100 may determine whether the performance metric of the prediction model is lower than a first preset threshold based on the relationship between the quantity of sample data and a corresponding quantity threshold.

[0114] Specifically, the cloud management platform 100 can determine the ratio of the second number of corrected error samples to the total number of data to be predicted based on the second number of erroneous samples and the total number of data to be predicted, and determine the difference between 1 and the ratio as the accuracy of the second prediction failure result. Alternatively, the cloud management platform 100 can also determine the difference between the total number of data to be predicted and the second number of corrected error samples, and determine the reciprocal of the difference as the accuracy of the second prediction failure result.

[0115] In this embodiment, the high-confidence prediction sample is the data to be predicted where the first predicted failure result, the second predicted failure result, and the actual failure result are all the same. In other words, if the first predicted failure result, the second predicted failure result, and the actual failure result of the data to be predicted are all the same, all indicating a failure or no failure, then the data to be predicted is a high-confidence prediction sample.

[0116] Among them, high-confidence samples can be used as sample data for the next training of the prediction model.

[0117] The corrected error sample is the data to be predicted where the first predicted fault result is consistent with the actual fault result, but the second predicted fault result is different from the first predicted fault result. That is, in this embodiment, when the first predicted fault result output by the prediction model is the same as the actual fault result (both indicating fault or no fault), but the second predicted fault result is different from the actual fault result, then the data to be predicted is the corrected error sample. The corrected error sample includes data to be predicted where both the first predicted fault result and the actual fault result indicate fault, but the second predicted fault result indicates no fault, and data to be predicted where both the first predicted fault result and the actual fault result indicate no fault, but the second predicted fault result indicates fault.

[0118] A false alarm sample is a data to be predicted where the first predicted fault result indicates a fault, but both the second predicted fault result and the actual fault result indicate no fault. That is, in this embodiment of the application, when the first predicted fault result of the data to be predicted indicates a fault, but both the second predicted fault result and the actual fault result indicate no fault, the cloud management platform 100 determines that the data to be predicted is a false alarm sample.

[0119] A missed sample is a data point to be predicted where the first predicted fault result indicates no fault, but both the second predicted fault result and the actual fault result indicate a fault. In other words, in this embodiment, when the first predicted fault result indicates no fault, but both the second predicted fault result and the actual fault result indicate a fault, the cloud management platform 100 determines that the data to be predicted is a missed sample.

[0120] Alternatively, without predicting the second predicted fault result for the data to be predicted, the first predicted fault result for the missed sample indicates data that is not faulty but the actual fault result indicates faulty. That is, in this embodiment of the application, without predicting the second predicted fault result for the data to be predicted, when the first predicted fault result for the data to be predicted indicates that it is not faulty but the actual fault result indicates faulty, the cloud management platform 100 determines that the data to be predicted is a missed sample.

[0121] As one possible implementation, when the cloud management platform 100 re-predicts the data to be predicted, which indicates a predicted fault in the first predicted fault result, to obtain a second predicted fault result, the process of classifying the data to be predicted by the cloud management platform 100 can be referred to the relevant descriptions in Figures 3 and 5 below, and will not be repeated here in the embodiments of this application.

[0122] Step 206: The cloud management platform 100 determines whether the performance index of the prediction model is lower than the first preset threshold. If yes, proceed to step 207; otherwise, return to step 205.

[0123] As one possible implementation, the cloud management platform 100 can also determine the weighted average of accuracy, recall, and false positive rate, and when the weighted average is lower than the corresponding preset threshold, determine that the performance index of the prediction model is lower than the first preset threshold.

[0124] As one possible implementation, the cloud management platform 100 may determine that the performance index of the prediction model is lower than a first preset threshold when the accuracy is lower than a preset accuracy threshold, and / or the recall is lower than a preset accuracy threshold, and / or the false positive rate is higher than a preset false positive rate threshold.

[0125] As one possible implementation, the cloud management platform 100 may determine that the performance index of the prediction model is lower than a first preset threshold if the number of false positive samples and / or the number of false negative samples in the sample data is greater than the corresponding quantity threshold.

[0126] Optionally, when the cloud management platform 100 classifies the data to be predicted into one of the following categories: high-confidence prediction samples, error correction samples, false positive samples, and missed positive samples, the cloud management platform 100 may also determine that the performance index of the prediction model is lower than the first preset threshold when the number of false positive samples exceeds the corresponding first preset threshold; and / or when the number of missed positive samples exceeds the corresponding second preset threshold.

[0127] Specifically, the cloud management platform 100 can determine the first preset number threshold of false positive samples corresponding to the first preset threshold set by the user, based on the correspondence between the preset number threshold corresponding to the false positive samples and the preset threshold of the performance index; or, when the performance index and the false positive rate are negatively correlated, the cloud management platform 100 can obtain the false positive rate threshold set by the user. The cloud management platform 100 can determine the first product value of the total number of data to be predicted and the false positive rate threshold based on the total number of data to be predicted and the false positive rate threshold. The cloud management platform 100 determines the first preset number threshold corresponding to the false positive samples as the first product value. When the number of false positive samples is greater than the corresponding first preset number threshold, the cloud management platform 100 determines that the performance index of the prediction model is lower than the first preset threshold.

[0128] The cloud management platform can also determine a second preset threshold for the number of missed samples corresponding to a first preset threshold set by the user, based on the correspondence between a preset threshold for the number of missed samples and a preset threshold for performance indicators; or, when performance indicators and the missed rate are negatively correlated, the cloud management platform 100 obtains a user-preset missed rate threshold. The cloud management platform 100 can determine a second product value between the total number of data to be predicted and the missed rate threshold, based on the total number of data to be predicted and the missed rate threshold. The cloud management platform 100 determines the second preset threshold for the number of false positive samples as the second product value. When the number of missed samples is greater than the corresponding second preset threshold, the cloud management platform 100 determines that the performance indicator of the prediction model is lower than the first preset threshold.

[0129] Step 207: The cloud management platform 100 retrains the prediction model.

[0130] As one possible implementation, when the performance index of the prediction model is lower than a first preset threshold, the cloud management platform 100 retrains the prediction model based on false positive samples and / or false negative samples.

[0131] Optionally, when the number of false positive samples exceeds the corresponding first preset threshold, the cloud management platform 100 retrains the prediction model based on the false positive samples; and / or, when the number of false negative samples exceeds the corresponding second preset threshold, the cloud management platform 100 retrains the prediction model based on the false negative samples.

[0132] Specifically, the cloud management platform 100 can be retrained using reinforcement learning or transfer learning methods.

[0133] Reinforcement learning is a form of experiential learning that involves a model continuously exploring and developing its environment and learning from feedback. In reinforcement learning, the model perceives state information in the environment, searches for a strategy to select the optimal action, thereby causing a change in state and receiving a delayed reward. This updates the evaluation function, and after completing one learning cycle, it enters the next round of training. This iterative cycle continues until the conditions for the entire learning process are met, at which point the learning terminates.

[0134] Transfer learning is the process of improving new tasks by transferring knowledge from related tasks that have already been learned.

[0135] Step 208: The cloud management platform 100 determines whether the accuracy of the second predicted fault result is lower than the second preset threshold. If yes, proceed to step 209; otherwise, return to step 205.

[0136] As one possible implementation, the cloud management platform 100 optimizes expert experience when the accuracy of the second predicted failure result is lower than a second preset threshold.

[0137] Optionally, when the cloud management platform 100 classifies the data to be predicted into one of the following categories: high-confidence prediction samples, error correction samples, false positive samples, and missed positive samples, the cloud management platform 100 may also determine that the accuracy of the second prediction failure result is lower than a second preset threshold when the number of error correction samples is greater than the corresponding third quantity threshold.

[0138] Specifically, the cloud management platform 100 can determine a third preset threshold for the number of false alarm samples corresponding to a second preset threshold set by the user, based on the correspondence between a preset threshold for the number of corrected error samples and a preset threshold for performance indicators. Alternatively, since the second preset threshold refers to the minimum accuracy of the second predicted fault result, the cloud management platform 100 can also determine the minimum number of corrected samples that meet the second preset threshold based on the total number of data to be predicted and the second preset threshold. The cloud management platform 100 determines the maximum number of corrected error samples as the third preset threshold, based on the difference between the total number of data to be predicted and the minimum number of corrected error samples. When the number of corrected error samples exceeds the corresponding third preset threshold, the cloud management platform 100 determines that the accuracy of the second predicted fault result is lower than the second preset threshold.

[0139] Step 209: Optimize the cloud management platform with expert experience.

[0140] As one possible implementation, the cloud management platform 100 obtains the optimized preset expert rules that have been manually modified by experts and input by the user, and updates the preset expert rules based on the optimized preset expert rules.

[0141] As one possible implementation, when the cloud management platform 100 determines the first predicted failure result of the predicted object, verifies the first predicted failure result based on expert experience, and obtains the second prediction information, the process of the cloud management platform 100 performing prediction model optimization and expert experience optimization can be referred to the relevant descriptions in Figures 3 and 6 below. The embodiments of this application will not be elaborated here.

[0142] Based on the description of steps 201 to 209 in Figure 2 above, the cloud management platform 100 obtains the first prediction information output by the prediction model and the second prediction information obtained by re-predicting the data to be predicted. Since the first prediction information is obtained by the prediction model, and the second prediction information is obtained by verifying the first prediction failure result, the target prediction failure result of the prediction object is determined by combining the prediction information obtained by the two different methods. When the first prediction failure result and the second prediction failure result are different, it can be determined that one of the prediction failure results is wrong. When the first prediction failure result and the second prediction failure result are the same, it can be determined that the first prediction failure result is correct. Compared with the scheme of determining the first prediction failure result output by the model as the target prediction failure result and being unable to detect whether the target prediction failure result is correct, the accuracy of the determined target prediction failure result is improved.

[0143] In addition, the cloud management platform 100 combines the first predicted fault result, the second predicted fault result verified by expert experience, and the actual fault result to classify the data to be predicted. Based on false positive samples and / or false negative samples, the platform evaluates and optimizes the prediction model, avoiding sample contamination caused by directly handling faults based on the first predicted fault result. This ensures the quality of the data used for model training and guarantees the sustainability of the model optimization and evaluation process.

[0144] Finally, the cloud management platform 100 monitors the performance of the prediction model in real time through model evaluation and retraining. When the model's performance deteriorates, it automatically retrains the model to ensure the accuracy of fault prediction. The cloud management platform 100 also monitors the effectiveness of expert experience in real time by determining the accuracy of the second predicted fault result. When the effectiveness of expert presentations deteriorates, it optimizes the system to ensure the accuracy of fault prediction.

[0145] The complete process of the fault prediction method has been described above. Next, taking the cloud management platform 100 in Figure 1 as an example, a specific embodiment will be used to describe the fault prediction method when re-predicting the data to be predicted, representing the first predicted fault result, to obtain the second predicted fault result. Figure 3 is a schematic diagram of a fault prediction method provided in an embodiment of this application.

[0146] As shown in Figure 3, the cloud management platform 100 obtains the first predicted fault result output by the prediction model, verifies the first predicted fault result based on expert experience to obtain the second predicted fault result, and determines the target predicted fault result based on the first and second predicted fault results. The target predicted fault result indicates whether the predicted object will experience a fault or will not experience a fault.

[0147] When both the first and second predicted failure results are predicted failures, the cloud management platform 100 determines the target predicted failure result as a predicted failure.

[0148] When the first predicted failure result is predicted as no failure, and when the first predicted failure result is predicted as a failure but the second predicted failure result is predicted as no failure, the cloud management platform 100 determines that the target predicted failure result is predicted as no failure.

[0149] The cloud management platform 100 issues fault alarms based on the predicted fault results of the target fault.

[0150] The specific solution for fault alarm in the cloud management platform 100 can be referred to the embodiment shown in Figure 4 below. The embodiment of this application will not be described in detail here.

[0151] Based on the target predicted fault results and the actual fault results, the cloud management platform 100 determines the data to be predicted to be one of the following categories: high-confidence predicted samples, error-corrected samples, false alarm samples, or missed alarm samples. The target predicted fault results include two cases: predicted fault and predicted no fault. The actual fault results include two cases: actual fault and actual no fault.

[0152] When the target predicted fault result is a predicted fault and the actual fault result is an actual fault, the cloud management platform 100 determines the data to be predicted as a high-confidence sample for prediction.

[0153] When the target predicted fault result is predicted as no fault and the actual fault result is actual fault, the cloud management platform 100 determines the data to be predicted as a sample for error correction.

[0154] When the target predicted fault result is a predicted non-fault obtained based on the first predicted fault result being a predicted fault but the second predicted fault result being a predicted non-fault, and the actual fault result being an actual non-fault, the cloud management platform 100 determines the data to be predicted to be a false alarm sample.

[0155] When the target predicted fault result is a predicted non-fault obtained based on the first predicted fault result being a predicted non-fault, and the actual fault result is an actual fault, the cloud management platform 100 determines the data to be predicted as a missed sample.

[0156] The specific scheme for sample classification by the cloud management platform 100 can be referred to in the embodiment shown in Figure 5 below, and will not be described in detail in this application embodiment.

[0157] The cloud management platform 100 determines the accuracy, recall, and false positive rate of the prediction model to evaluate its performance.

[0158] When the performance metrics of the prediction model fall below the preset performance metric threshold, the cloud management platform 100 retrains the prediction model and optimizes the preset expert rules.

[0159] The specific scheme for optimizing model performance in the cloud management platform 100 can be found in the embodiment shown in Figure 6 below, which will not be elaborated upon here.

[0160] Based on the fault prediction method shown in Figure 3 above, when the first predicted fault result of the prediction model represents the predicted fault, the cloud management platform 100 verifies the first predicted fault result based on expert experience to obtain a second predicted fault result. Cross-validation between the second and first predicted fault results ensures the accuracy of the target predicted fault result. Furthermore, the cloud management platform 100 combines the first, second, and actual fault results to classify the data to be predicted, avoiding sample contamination caused by directly applying the first predicted fault result to fault handling. This ensures the quality of data used for model training and guarantees the sustainability of the model optimization and evaluation process. Finally, the cloud management platform 100 monitors the performance of the prediction model in real time and automatically retrains the model when its performance deteriorates, ensuring the accuracy of fault prediction.

[0161] The above describes the overall process of the fault prediction method. Next, a specific solution for fault alarm of cloud management platform 100 will be described through a specific embodiment.

[0162] Figure 4 is a schematic diagram of a fault alarm process provided in an embodiment of this application.

[0163] As shown in Figure 4, the input data includes health indicator data, performance indicator data, and environmental indicator data.

[0164] The cloud management platform 100 determines the first predicted fault result obtained after inputting the input data into the prediction model. The first predicted fault result is either predicted fault or predicted no fault.

[0165] The cloud management platform 100 re-predicts the data to be predicted based on the first predicted fault result, after being judged by preset expert rules or through online stress testing, to obtain the second predicted fault result. The second predicted fault result is either a predicted fault or a predicted no fault.

[0166] The cloud management platform 100 determines the target predicted fault result based on the first and second predicted fault results. The target predicted fault result is either predicted fault or predicted no fault.

[0167] The target predicted fault result can be obtained from the first predicted fault result and the second predicted fault result. When the first predicted fault result is a predicted fault and the second predicted fault result is a predicted fault, the predicted fault of the target predicted fault result can be called a high-confidence fault.

[0168] The predicted failure result for a target that is not predicted to fail can be obtained from either a first predicted failure result and a second predicted failure result, or from the first predicted failure result. Where the first predicted failure result is a predicted failure and the second predicted failure result is a predicted failure, the predicted failure result for the target predicted failure result can also be called a corrected failure result.

[0169] The cloud management platform 100 issues a fault alarm when the predicted fault result of the target is a high-confidence fault.

[0170] Based on the above embodiments, when the first predicted fault result of the prediction model is a predicted fault, the cloud management platform 100 cross-validates the second predicted fault result with the first predicted fault result to ensure the accuracy of the fault alarm prediction. A fault alarm is only issued when the target predicted fault result is a high-confidence fault, where both the first and second predicted fault results are predicted faults, thus ensuring the accuracy of the fault alarm.

[0171] The above describes the specific process of fault alarm. Next, we will describe the specific scheme for sample classification of the cloud management platform 100 through a specific embodiment.

[0172] Figure 5 is a schematic diagram of a sample classification process provided in an embodiment of this application.

[0173] As shown in Figure 5, the input data includes the first predicted fault result, the second predicted fault result, and the actual fault result.

[0174] The cloud management platform 100 determines that the first predicted fault result, the second predicted fault result, and the actual fault result all represent high-confidence samples of the fault prediction. These high-confidence samples are used as sample data for the next training iteration of the prediction model.

[0175] The cloud management platform 100 identifies corrected error samples where both the first predicted fault result and the actual fault result indicate a fault, but the second predicted fault result indicates no fault. These corrected error samples serve as a reference for experts when modifying preset expert rules.

[0176] The cloud management platform 100 identifies false alarm samples where the first predicted fault result indicates a fault, but the second predicted fault result and the actual fault result both indicate no fault. These false alarm samples are used to evaluate the false alarm rate of the prediction model and optimize the model algorithm.

[0177] The cloud management platform 100 identifies missed samples where the first predicted fault result is not a fault, but the actual fault result represents a fault. These missed samples are crucial for improving the prediction model, and analyzing them helps identify model weaknesses.

[0178] The cloud management platform 100 performs sample analysis and sample feedback. Sample analysis refers to using data visualization tools to display the sample distribution and classification results for each type of sample data. Sample feedback refers to using the classification results to evaluate and optimize the performance of the prediction model.

[0179] Based on the above embodiments, the cloud management platform 100 classifies the data to be predicted by combining the first predicted fault result, the second predicted fault result after re-prediction, and the actual fault result. This avoids sample contamination caused by directly handling faults based on the first predicted fault result, ensures the quality of data used for model training, and ensures the sustainability of the model optimization and evaluation process.

[0180] The above describes the specific process of sample classification. Next, a specific implementation plan for optimizing the model performance of the cloud management platform 100 will be described.

[0181] Figure 6 is a schematic diagram of a model performance optimization process provided in an embodiment of this application.

[0182] As shown in Figure 6, the cloud management platform 100 obtains the classification results.

[0183] The cloud management platform 100 sets performance metrics. These metrics include accuracy, recall, and false positive rate.

[0184] The cloud management platform acquires performance metrics data in real time. When the performance metrics of the prediction model fall below the preset performance metric threshold, model optimization is automatically triggered.

[0185] The cloud management platform 100 performs model optimization. Model optimization includes model retraining and optimization of preset expert rules. Model retraining by the cloud management platform 100 can improve the predictive model's adaptability to new data environments. Optimization of preset expert rules by the cloud management platform 100 can improve the accuracy of re-predictions based on preset expert rules.

[0186] The cloud management platform 100 provides feedback and iterative updates. After optimization, the cloud management platform 100 re-evaluates the performance of the predictive model and compares the new results with the previous performance to verify the optimization effect. This process is a continuous loop to ensure that the predictive model always remains in optimal condition.

[0187] Based on the above embodiments, the cloud management platform 100 monitors the performance of the prediction model in real time. When the model's performance deteriorates, the cloud management platform 100 automatically retrains the model to ensure the accuracy of fault prediction. Furthermore, the cloud management platform 100 updates the preset expert rules to better adapt them to the new data environment, thereby further improving the accuracy of fault prediction.

[0188] The fault prediction device provided in this embodiment will be described below with reference to Figure 7. Figure 7 is a schematic diagram of a fault prediction device 700 provided in an embodiment of this application. The fault prediction device 700 is applied to a cloud management platform, which is used to manage infrastructure. A prediction model runs on the infrastructure, and the infrastructure includes at least one prediction object. The application of the fault prediction device 700 to the cloud management platform means that the fault prediction device 700 can be a functional module in the cloud management platform, or it can refer to the hardware configuration where the cloud management platform is located.

[0189] As shown in Figure 7, the fault prediction device 700 includes:

[0190] The acquisition module 701 is used to acquire first prediction information from the prediction model, wherein the prediction model is used to predict the fault of the prediction object based on the data to be predicted, the data to be predicted is used to indicate the operating data of the prediction object, and the first prediction information includes a first predicted fault result. For example, the acquisition module 701 is used to execute step 201 shown in Figure 2 above.

[0191] The determination module 702 is used to verify the first predicted fault result based on expert experience, obtain second predicted information, the second predicted information including the second predicted fault result; and determine the target predicted fault result of the predicted object based on the first predicted information and the second predicted information. For example, the determination module 702 is used to execute steps 202 and 203 shown in Figure 2 above.

[0192] As one possible implementation, the determination module 702 is also used to determine the performance index of the prediction model and the accuracy of the second predicted fault result based on the target predicted fault result and the actual fault result of the predicted object.

[0193] The fault prediction device 700 also includes:

[0194] The processing module 703 is used to optimize the prediction model when the performance index of the prediction model is lower than a first preset threshold, and / or to optimize expert experience when the accuracy of the second predicted fault result is lower than a second preset threshold.

[0195] As one possible implementation, module 702 is specifically used to determine sample data of the prediction object based on the first predicted fault result, the second predicted fault result, and the actual fault result. The sample data includes one or more combinations of high-confidence predicted samples, error-corrected samples, false alarm samples, and missed alarm samples. Based on the number of sample data, the performance index of the prediction model is determined; and based on the number of error-corrected samples, the accuracy of the second predicted fault result is determined. The high-confidence predicted samples are the data to be predicted that are consistent with the first predicted fault result, the second predicted fault result, and the actual fault result.

[0196] The erroneous sample is the data to be predicted where the first predicted fault result is consistent with the actual fault result, but the second predicted fault result is different from the first predicted fault result.

[0197] The false alarm sample is the data to be predicted where the first predicted fault result indicates a fault, but the second predicted fault result and the actual fault result both indicate that there is no fault.

[0198] The missed sample is the data to be predicted, which is not faulty according to the first predicted fault result but is faulty according to both the second predicted fault result and the actual fault result; or, in the case where the second predicted fault result of the data to be predicted is not predicted, the missed sample is the data to be predicted, which is not faulty according to the first predicted fault result but is faulty according to the actual fault result.

[0199] As one possible implementation, the processing module 703 is specifically used to determine that the performance index of the prediction model is lower than the first preset threshold when the number of false positive samples is greater than the corresponding first preset threshold; and / or, when the number of false negative samples is greater than the corresponding second preset threshold, determine that the performance index of the prediction model is lower than the first preset threshold; and / or, when the number of corrected erroneous samples is greater than the corresponding third preset threshold, determine that the accuracy of the second prediction fault result is lower than the second preset threshold.

[0200] As one possible implementation, expert experience is based on preset expert rules. The determination module 702 is specifically used to predict the data to be predicted according to the preset expert rules, and to determine the second prediction information; the preset expert rules are used to specify the predicted fault results and the probability of occurrence of the predicted fault results corresponding to the operating data of at least one predicted object.

[0201] As one possible implementation method, expert experience suggests conducting online load testing. Module 702 is specifically used to perform online load testing on the object to be predicted based on the data to be predicted, thereby determining the second prediction information.

[0202] As one possible implementation, the first prediction information further includes the probability of occurrence of a first predicted fault result, and the second prediction information further includes the probability of occurrence of a second predicted fault result. The determining module 702 is specifically used to determine the target predicted fault result based on the probability of occurrence of the first predicted fault result and the probability of occurrence of the second predicted fault result when the first predicted fault result and the second predicted fault result are different.

[0203] As one possible implementation, the first predicted fault result and the second predicted fault result are different, and the target predicted fault result is the predicted fault result with a higher probability of occurrence between the first predicted fault result and the second predicted fault result.

[0204] As one possible implementation, the determining module 702 is specifically used to verify the first predicted fault result and obtain the second predicted information based on expert experience when the first predicted fault result is determined to be a predicted fault of the object.

[0205] As one possible implementation, the fault prediction device 700 also includes:

[0206] The alarm module 704 is used to issue a fault alarm when the predicted fault result of the predicted object is a fault.

[0207] The acquisition module 701, determination module 702, alarm module 704, and processing module 703 can all be implemented in software or in hardware. For example, the implementation of the acquisition module 701 will be described below. Similarly, the implementation of the determination module 702, alarm module 704, and processing module 703 can refer to the implementation of the acquisition module 701.

[0208] As an example of a software functional unit, module 701 may include code running on a computing instance. The computing instance may include at least one of a physical host (computing device), a virtual machine, or a container. Further, the aforementioned computing instance may be one or more. For example, module 701 may include code running on multiple hosts / virtual machines / containers. It should be noted that the multiple hosts / virtual machines / containers used to run the code may be distributed within the same region or in different regions. Further, the multiple hosts / virtual machines / containers used to run the code may be distributed within the same availability zone (AZ) or in different AZs, each AZ including one or more geographically proximate data centers. Typically, a region may include multiple AZs.

[0209] Similarly, multiple hosts / virtual machines / containers used to run this code can be distributed within the same Virtual Private Cloud (VPC) or across multiple VPCs. Typically, a VPC is set up within a region. Communication between two VPCs within the same region, as well as between VPCs in different regions, requires a communication gateway to be set up within each VPC to enable interconnection between VPCs.

[0210] As an example of a hardware functional unit, the acquisition module 701 may include at least one computing device, such as a server. Alternatively, the acquisition module 701 may also be a device implemented using an application-specific integrated circuit (ASIC) or a programmable logic device (PLD). The PLD may be implemented using a complex programmable logical device (CPLD), a field-programmable gate array (FPGA), generic array logic (GAL), or any combination thereof.

[0211] The multiple computing devices included in the acquisition module 701 can be distributed in the same region or in different regions. Similarly, the multiple computing devices included in the acquisition module 701 can be distributed in the same Availability Zone (AZ) or in different AZs. Likewise, the multiple computing devices included in the acquisition module 701 can be distributed in the same Virtual Private Cloud (VPC) or in multiple VPCs. These multiple computing devices can be any combination of computing devices such as servers, ASICs, PLDs, CPLDs, FPGAs, and GALs.

[0212] It should be noted that, in other embodiments, the acquisition module 701 can be used to execute any step in the fault prediction method, the determination module 702 can be used to execute any step in the fault prediction method, the processing module 703 can be used to execute any step in the fault prediction method, and the alarm module 704 can be used to execute any step in the fault prediction method. The steps implemented by the acquisition module 701, the determination module 702, the processing module 703, and the alarm module 704 can be specified as needed. By implementing different steps in the fault prediction method through the acquisition module 701, the determination module 702, the processing module 703, and the alarm module 704, all functions of the fault prediction device 700 can be realized.

[0213] It should be noted that the grouping of modules in this embodiment is illustrative and represents only one logical functional grouping. In actual implementation, there may be other grouping methods. For example, in other embodiments, the functional modules in the fault prediction device 700 include a dual verification module, an alarm module, a sample feedback module, a model management module, and a model optimization module.

[0214] The dual verification module combines model prediction with expert experience. Specifically, it determines the target predicted fault result by combining the first predicted information output by the prediction model with the second predicted information obtained based on expert experience and the data to be predicted. This dual verification module reduces the risk of false alarms and improves the accuracy of the target predicted fault result. For example, the dual verification module is used to implement the function of determining the target predicted fault result in the determination module 702 of the above embodiment.

[0215] The alarm module is used to send the target predicted fault result output by the dual verification module to the alarm system. When the target predicted fault result indicates a predicted fault in the object, it generates an alarm repair order to realize fault alarm and subsequent fault handling. For example, the alarm module is used to implement the fault alarm function of alarm module 704 in the above embodiment.

[0216] The sample feedback module combines the first predicted fault result, the re-predicted second predicted fault result, and the actual fault result to classify the data to be predicted. This classification result ensures the sustainability and accuracy of model evaluation and iterative optimization. For example, the sample feedback module is used to implement the sample classification function of the determination module 702 in the above embodiment.

[0217] The model management module is used to acquire performance metrics data of the prediction model in real time based on the classified sample data. When performance degrades, model optimization is automatically triggered, utilizing false positive and false negative samples to optimize the prediction model and ensure prediction accuracy. For example, the module management module implements the function of acquiring performance metrics data of the prediction model as described in the above embodiment.

[0218] The model optimization module is used to retrain the prediction model when the performance index data of the prediction model obtained by the model management module is lower than a preset performance index threshold. For example, the model optimization module is used to implement the model retraining function of the processing module 703 in the above embodiment.

[0219] This application also provides a computing device 800. Figure 8 is a schematic diagram of the structure of a computing device provided in an embodiment of this application. As shown in Figure 8, the computing device 800 includes: a bus 802, a processor 804, a memory 806, and a communication interface 808. The processor 804, the memory 806, and the communication interface 808 communicate with each other via the bus 802. The computing device 800 can be a server or a terminal device. It should be understood that this application does not limit the number of processors and memories in the computing device 800.

[0220] Bus 802 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of illustration, only one line is used in Figure 8, but this does not imply that there is only one bus or one type of bus. Bus 802 can include pathways for transmitting information between various components of computing device 800 (e.g., memory 806, processor 804, communication interface 808).

[0221] Processor 804 may include any one or more processors such as a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor (MP), or a digital signal processor (DSP).

[0222] The memory 806 may include volatile memory, such as random access memory (RAM). The processor 804 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0223] The memory 806 stores executable program code, and the processor 804 executes this executable program code to implement the functions of the aforementioned acquisition module 701, determination module 702, processing module 703, and alarm module 704, thereby realizing the fault prediction method. That is, the memory 806 stores instructions for executing the fault prediction method.

[0224] The communication interface 808 uses transceiver modules such as, but not limited to, network interface cards and transceivers to enable communication between the computing device 800 and other devices or communication networks.

[0225] This application also provides a computing device cluster. The computing device cluster includes at least one computing device. The computing device can be a server, such as a central server, an edge server, or a local server in a local data center. In some embodiments, the computing device can also be a terminal device such as a desktop computer, a laptop computer, or a smartphone.

[0226] Figure 9 is a schematic diagram of a computing device cluster provided in an embodiment of this application. As shown in Figure 9, the computing device cluster includes at least one computing device 800. The memory 806 of one or more computing devices 800 in the computing device cluster may store the same instructions for executing the fault prediction method.

[0227] In some possible implementations, the memory 806 of one or more computing devices 800 in the computing device cluster may also store partial instructions for executing the fault prediction method. In other words, a combination of one or more computing devices 800 can jointly execute the instructions for executing the fault prediction method.

[0228] It should be noted that the memory 806 in different computing devices 800 within the computing device cluster can store different instructions, each used to execute a portion of the functions of the fault prediction device. That is, the instructions stored in the memory 806 of different computing devices 800 can implement the functions of one or more modules among the acquisition module 701, determination module 702, processing module 703, and alarm module 704.

[0229] In some possible implementations, one or more computing devices in a computing device cluster can be connected via a network. This network can be a wide area network (WAN) or a local area network (LAN), etc. Figure 10 illustrates one possible implementation. Figure 10 is a schematic diagram of a network connection between computing devices according to an embodiment of this application. As shown in Figure 10, two computing devices 800A and 800B are connected via a network. Specifically, they are connected to the network through communication interfaces in each computing device. In this type of possible implementation, the memory 806 in computing device 800A stores instructions for executing the functions of the acquisition module 701 and the alarm module 704. Simultaneously, the memory 806 in computing device 800B stores instructions for executing the functions of the determination module 702 and the processing module 703.

[0230] It should be understood that the functions of computing device 800A shown in Figure 10 can also be performed by multiple computing devices 800. Similarly, the functions of computing device 800B can also be performed by multiple computing devices 800.

[0231] This application also provides a computer program product containing instructions. The computer program product may be a software or program product containing instructions, capable of running on a computing device or stored on any usable medium. When the computer program product is run on at least one computing device, it causes the at least one computing device to perform a fault prediction method.

[0232] This application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that a computing device can store, or a data storage device such as a data center that includes one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive). The computer-readable storage medium includes instructions that instruct the computing device to perform a fault prediction method.

[0233] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A fault prediction method, characterized in that, The method is applied to a cloud management platform for managing infrastructure, on which a predictive model runs, and the infrastructure includes at least one predictive object. The method includes: Obtain first prediction information from the prediction model, wherein the prediction model is used to predict the failure of the prediction object based on the data to be predicted, the data to be predicted is used to indicate the operating data of the prediction object, and the first prediction information includes a first predicted failure result; Based on expert experience, the first predicted fault result is verified to obtain second predicted information, which includes the second predicted fault result. Based on the first prediction information and the second prediction information, the target predicted fault result of the predicted object is determined.

2. The method according to claim 1, characterized in that, The method further includes: Based on the predicted fault results of the target and the actual fault results of the predicted object, the performance index of the prediction model and the accuracy of the second predicted fault result are determined. If the performance index of the prediction model is lower than a first preset threshold, the prediction model is optimized, and / or, if the accuracy of the second predicted fault result is lower than a second preset threshold, the expert experience is optimized.

3. The method according to claim 2, characterized in that, The step of determining the performance index of the prediction model and the accuracy of the second predicted fault result based on the target predicted fault result and the actual fault result of the predicted object includes: Based on the first predicted fault result, the second predicted fault result, and the actual fault result, the sample data of the predicted object is determined. The sample data includes one or more combinations of high-confidence predicted samples, error-corrected samples, false alarm samples, and missed alarm samples. The performance metrics of the prediction model are determined based on the quantity of the sample data. Based on the corrected error samples, the accuracy of the second predicted fault result is determined; wherein, The high-confidence prediction sample is the data to be predicted that is consistent with the first predicted fault result, the second predicted fault result, and the actual fault result. The corrected error sample is the data to be predicted where the first predicted fault result is consistent with the actual fault result, but the second predicted fault result is different from the first predicted fault result; The false alarm sample is the data to be predicted, where the first predicted fault result indicates a fault, but the second predicted fault result and the actual fault result both indicate that there is no fault. The missed sample is the data to be predicted where the first predicted fault result indicates no fault, but both the second predicted fault result and the actual fault result indicate a fault; or, in the case where the second predicted fault result of the data to be predicted is not predicted, the missed sample is the data to be predicted where the first predicted fault result indicates no fault, but the actual fault result indicates a fault.

4. The method according to claim 3, characterized in that, Determining the performance metrics of the prediction model based on the quantity of sample data includes: When the number of false positive samples exceeds a corresponding first preset threshold, the performance index of the prediction model is determined to be lower than the first preset threshold; and / or, When the number of missed samples is greater than the corresponding second preset threshold, it is determined that the performance index of the prediction model is lower than the first preset threshold. Determining the accuracy of the second predicted fault result based on the corrected error samples includes: When the number of corrected error samples is greater than the corresponding third preset threshold, the accuracy of the second predicted fault result is determined to be lower than the second preset threshold.

5. The method according to any one of claims 1-4, characterized in that, The expert experience is a preset expert rule. The step of verifying the first predicted fault result based on the expert experience to obtain the second predicted information includes: The data to be predicted is predicted according to the preset expert rules to determine the second prediction information; the preset expert rules are used to specify the predicted fault results corresponding to the operating data of the predicted object and the probability of occurrence of the predicted fault results.

6. The method according to any one of claims 1-4, characterized in that, The expert experience is used for online stress testing. The step of verifying the first predicted fault result based on the expert experience to obtain the second predicted information includes: Based on the data to be predicted, online stress testing is performed on the object to be predicted to determine the second prediction information.

7. The method according to any one of claims 1-6, characterized in that, The first prediction information further includes the probability of occurrence of the first predicted failure result, and the second prediction information further includes the probability of occurrence of the second predicted failure result; determining the target predicted failure result of the prediction object based on the first prediction information and the second prediction information includes: When the first predicted fault result and the second predicted fault result are different, the target predicted fault result is determined based on the probability of occurrence of the first predicted fault result and the probability of occurrence of the second predicted fault result.

8. The method according to claim 7, characterized in that, If the first predicted fault result and the second predicted fault result are different, the target predicted fault result is the predicted fault result with a higher probability of occurrence between the first predicted fault result and the second predicted fault result.

9. The method according to any one of claims 1-8, characterized in that, The step of verifying the first predicted fault result based on expert experience to obtain the second predicted information includes: If the first predicted fault result is determined to be a predicted fault for the predicted object, the first predicted fault result is verified based on expert experience to obtain second prediction information.

10. The method according to any one of claims 1-9, characterized in that, The method further includes: If the predicted fault result of the predicted object is a fault, a fault alarm will be issued.

11. A fault prediction device, characterized in that, Applied to a cloud management platform for managing infrastructure on which a predictive model runs, the infrastructure including at least one predictive object; the device includes: An acquisition module is used to acquire first prediction information from the prediction model, wherein the prediction model is used to predict the failure of the prediction object based on the data to be predicted, the data to be predicted is used to indicate the operating data of the prediction object, and the first prediction information includes a first predicted failure result. The determination module is used to verify the first predicted fault result based on expert experience, obtain second prediction information, the second prediction information including the second predicted fault result; and determine the target predicted fault result of the predicted object based on the first prediction information and the second prediction information.

12. The apparatus according to claim 11, characterized in that, The determining module is further configured to determine the performance index of the prediction model and the accuracy of the second predicted fault result based on the target predicted fault result and the actual fault result of the predicted object; The device further includes: The processing module is configured to optimize the prediction model when the performance index of the prediction model is lower than a first preset threshold, and / or optimize the expert experience when the accuracy of the second predicted fault result is lower than a second preset threshold.

13. The apparatus according to claim 12, characterized in that, The determining module is specifically used to determine sample data of the prediction object based on the first predicted fault result, the second predicted fault result, and the actual fault result. The sample data includes one or more combinations of high-confidence prediction samples, error correction samples, false positive samples, and false negative samples. Based on the number of sample data, the module determines the performance index of the prediction model. Based on the number of error correction samples, the module determines the accuracy of the second predicted fault result. The high-confidence prediction samples are the prediction data where the first predicted fault result, the second predicted fault result, and the actual fault result are consistent. The corrected error sample is the data to be predicted where the first predicted fault result is consistent with the actual fault result, but the second predicted fault result is different from the first predicted fault result; The false alarm sample is the data to be predicted, where the first predicted fault result indicates a fault, but the second predicted fault result and the actual fault result both indicate that there is no fault. The missed sample is the data to be predicted where the first predicted fault result indicates no fault, but both the second predicted fault result and the actual fault result indicate a fault; or, in the case where the second predicted fault result of the data to be predicted is not predicted, the missed sample is the data to be predicted where the first predicted fault result indicates no fault, but the actual fault result indicates a fault.

14. The apparatus according to claim 13, characterized in that, The processing module is specifically configured to determine that the performance index of the prediction model is lower than the first preset threshold when the number of false positive samples is greater than the corresponding first preset threshold; and / or, determine that the performance index of the prediction model is lower than the first preset threshold when the number of false negative samples is greater than the corresponding second preset threshold. And / or, when the number of corrected error samples is greater than the corresponding third preset number threshold, the accuracy of the second predicted fault result is determined to be lower than the second preset threshold.

15. The apparatus according to any one of claims 11-14, characterized in that, The expert experience is a preset expert rule. The determining module is specifically used to predict the data to be predicted according to the preset expert rule and determine the second prediction information. The preset expert rule is used to specify the predicted fault result corresponding to the running data of at least one predicted object and the probability of occurrence of the predicted fault result.

16. The apparatus according to any one of claims 11-14, characterized in that, The expert experience is used for online stress testing. The determination module is specifically used to perform online stress testing on the prediction object based on the data to be predicted, and to determine the second prediction information.

17. The apparatus according to any one of claims 11-16, characterized in that, The first prediction information further includes the probability of occurrence of the first predicted fault result, and the second prediction information further includes the probability of occurrence of the second predicted fault result; the determining module is specifically used to determine the target predicted fault result based on the probability of occurrence of the first predicted fault result and the probability of occurrence of the second predicted fault result when the first predicted fault result and the second predicted fault result are different.

18. The apparatus according to claim 17, characterized in that, If the first predicted fault result and the second predicted fault result are different, the target predicted fault result is the predicted fault result with a higher probability of occurrence between the first predicted fault result and the second predicted fault result.

19. The apparatus according to any one of claims 11-18, characterized in that, The determining module is specifically used to verify the first predicted fault result and obtain second prediction information based on expert experience when the first predicted fault result is determined to be a predicted fault of the predicted object.

20. The apparatus according to any one of claims 11-19, characterized in that, The device further includes: The alarm module is used to issue a fault alarm when the target predicted fault result of the predicted object is a fault.

21. A computing device cluster, characterized in that, It includes at least one computing device, each computing device including a processor and memory; The processor of the at least one computing device is configured to execute instructions stored in the memory of the at least one computing device to cause the cluster of computing devices to perform the operational steps of the method as described in any one of claims 1-10.

22. A computer program product containing instructions, characterized in that, When the instruction is executed by the computing device cluster, the computing device cluster causes the computing device cluster to perform the operation steps of the method as described in any one of claims 1-10.

23. A computer-readable storage medium, characterized in that, It includes computer program instructions, which, when executed by a cluster of computing devices, perform the operational steps of the method as described in any one of claims 1-10.

Images