Hard disk failure prediction method and apparatus

Abstract

The present disclosure provides a hard disk failure prediction method and device, comprising: obtaining real-time log data associated with the state of the hard disk; inputting the real-time log data into a failure prediction model to obtain a hard disk failure prediction result; wherein the failure prediction model is obtained by training based on negative samples of failed hard disks and positive samples of normal hard disks; the sampling time interval of the negative samples of the failed hard disks is a time period from a time point one predetermined sampling period earlier than a failure reporting time point to the failure reporting time point. The log data of a period of time before the failure reporting point is used to train the failure prediction model, so that the failure prediction model can learn the features before the hard disk fails during training, which is beneficial to the failure prediction model to make more accurate and timely prediction of the hard disk failure, leaving time for repair or replacement of the hard disk, and ensuring the normal operation of the service based on the hard disk.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a method and apparatus for predicting hard disk failures. Background Technology

[0002] Hard drive failure prediction can determine whether a hard drive has failed through a failure prediction model. However, the accuracy of hard drive failure prediction is low in related technologies. It can usually only predict hard drive failure when a serious failure has occurred. The probability of predicting hard drive failure when the hard drive is about to fail or has a minor failure is low. This makes it difficult to repair or replace the hard drive in a timely manner based on the prediction results, and it is difficult to ensure the normal operation of hard drive-based services. Summary of the Invention

[0003] This disclosure provides a hard disk failure prediction method and apparatus to address the shortcomings of existing technologies where hard disk failure prediction accuracy is low when using failure prediction models. The method enables the failure prediction model to predict more accurate results based on the characteristics of the hard disk before failure, thereby facilitating timely repair or replacement of the hard disk based on the prediction results and ensuring the normal operation of hard disk-based services.

[0004] This disclosure provides a hard disk failure prediction method, comprising: acquiring real-time log data associated with the hard disk status; inputting the real-time log data into a failure prediction model to obtain a hard disk failure prediction result; wherein the failure prediction model is trained based on negative samples of failed hard disks and positive samples of normal hard disks; and the sampling time interval of the negative samples of failed hard disks is the time period between a time point one predetermined sampling period earlier than the failure reporting time point and the failure reporting time point.

[0005] According to the hard disk failure prediction method provided in this disclosure, the real-time log data includes:

[0006] The log data includes first log data based on the fault information corresponding to the hard drive failure work order, second log data based on the fault information corresponding to the alarm and error information of the faulty hard drive, third log data corresponding to the normal hard drive, and fourth log data corresponding to the normal hard drive.

[0007] Furthermore, during the training process of the fault prediction model, the negative samples of the faulty hard drive include the first log data and the second log data; the positive samples of the normal hard drive include the third log data and the fourth log data.

[0008] According to a hard disk failure prediction method provided in this disclosure, the first log data and the third log data belong to SAMRT log data;

[0009] The second log data and the fourth log data belong to the Message log data.

[0010] According to the hard disk failure prediction method provided in this disclosure, the failure prediction model is dynamically updated according to a preset update strategy.

[0011] According to the hard disk failure prediction method provided in this disclosure, the preset update strategy includes: scheduled task-triggered update, random-triggered update, and automatic update triggered when the failure prediction result deviates from the evaluation index.

[0012] According to the hard disk failure prediction method provided in this disclosure, the step of updating according to a preset update strategy further includes:

[0013] In the feature pool, feature analysis and selection are performed on the data obtained by data fusion based on the negative samples of the faulty hard drive and the positive samples of the normal hard drive, and data with correlation exceeding a set threshold are extracted as training dataset.

[0014] Based on the training dataset, different algorithm models in the algorithm pool are trained, and the optimal algorithm model is extracted as the updated fault prediction model; wherein, the different algorithm models belong to the classification algorithms in machine learning.

[0015] This disclosure also provides a hard disk failure prediction device, comprising:

[0016] The first acquisition module is used to acquire real-time log data associated with the hard disk status;

[0017] The first acquisition module is used to input the real-time log data into the fault prediction model to obtain hard disk fault prediction results;

[0018] The fault prediction model is trained based on negative samples of faulty hard drives and positive samples of normal hard drives. The sampling time interval of the negative samples of faulty hard drives is the time period between the time point one predetermined sampling period earlier than the fault reporting time point and the fault reporting time point.

[0019] This disclosure also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the hard disk failure prediction methods described above.

[0020] This disclosure also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described hard disk failure prediction methods.

[0021] This disclosure also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above-described hard disk failure prediction methods.

[0022] The hard disk failure prediction method and apparatus disclosed herein use features reflecting the hard disk status in log data from a period prior to the failure reporting point as negative samples of the failed hard disk to train the failure prediction model. This allows the model to learn the features of the hard disk before failure during training, enabling it to provide early warnings when predicting real-time log data associated with the hard disk status. This early warnings facilitate timely repair or replacement of the hard disk, ensuring the normal operation of hard disk-based services. Attached Figure Description

[0023] To more clearly illustrate the technical solutions in this disclosure or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0024] Figure 1 This is one of the flowcharts of the hard disk failure prediction method provided in this disclosure;

[0025] Figure 2 This is the second flowchart of the hard disk failure prediction method provided in this publication;

[0026] Figure 3 This is the third flowchart of the hard drive failure prediction method provided in this publication;

[0027] Figure 4 This is a schematic diagram of the hard disk failure prediction device provided in this disclosure;

[0028] Figure 5 This is a schematic diagram of the structure of the electronic device provided in this disclosure. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this disclosure clearer, the technical solutions of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0030] For ease of understanding, the terms used in this disclosure will first be explained.

[0031] (1) Self-monitoring analysis and reporting technology (SMART)

[0032] SMART is an automatic hard drive status monitoring and alerting system and standard. Most modern hard drives integrate this technology, and the SMART log allows users to view the hard drive's health status.

[0033] The following is combined Figures 1-5 This disclosure describes a hard disk failure prediction method and apparatus.

[0034] Figure 1 This is one of the flowcharts illustrating the hard disk failure prediction method provided in this disclosure. It can be understood that... Figure 1 The methods described can be executed by a hard disk failure prediction device. For example... Figure 1 As shown, the hard disk failure prediction method provided in this disclosure includes the following steps:

[0035] Step 110: Obtain real-time log data associated with the hard disk status.

[0036] In one embodiment, the hard disk is a hard disk drive (HDD).

[0037] Among them, the real-time log data associated with the hard drive status reflects the status of the hard drive, such as whether the hard drive has failed and the type of hard drive failure.

[0038] Step 120: Input the real-time log data into the fault prediction model to obtain the hard disk fault prediction result.

[0039] The fault prediction model is trained based on negative samples of faulty hard drives and positive samples of normal hard drives. The sampling time interval of the negative samples of faulty hard drives is the time period between the time point one predetermined sampling period earlier than the fault reporting time point and the fault reporting time point.

[0040] The fault reporting time point can be understood as the time when the fault occurs; that is, the fault is reported as soon as it occurs, and the reporting time is negligible. Positive and negative samples can be log data associated with the hard drive status. The fault prediction model is trained on the algorithm model based on the negative and positive samples. The algorithm model can be one of the classification algorithms in machine learning, such as support vector machines, decision trees, random forests, reinforcement learning, XGBoost, or deep learning algorithms.

[0041] Among them, the hard drive failure prediction result is used to indicate whether the hard drive has failed.

[0042] It's understandable that features reflecting the hard drive's status in the log data corresponding to the fault reporting time point are typically used as samples to train the fault prediction model. However, since the fault prediction model trained using these features might not be usable for hard drive fault prediction even after a failure has occurred, a solution is to use the time interval between a predetermined sampling period earlier than the fault reporting time point and the fault reporting time point as the sampling time interval for negative samples. This allows the model to obtain log data prior to the hard drive failing or becoming unrepairable. This enables the fault prediction model to learn the features prior to the hard drive failure during training, leading to more accurate predictions. This, in turn, facilitates timely repair or replacement of the hard drive based on the prediction results, ensuring the normal operation of hard drive-based services.

[0043] The hard disk failure prediction method disclosed herein uses features reflecting the hard disk status in log data from a period of time prior to the failure reporting point as negative samples of the failed hard disk to train the failure prediction model. This allows the failure prediction model to learn the features of the hard disk before failure during training, thereby enabling the model to provide a fault warning when predicting real-time log data associated with the hard disk status, indicating that the corresponding hard disk is about to fail or has a minor failure. This facilitates timely repair or replacement of the hard disk, ensuring the normal operation of hard disk-based services.

[0044] Based on any of the above embodiments, in this embodiment...

[0045] The real-time log data includes:

[0046] The log data includes first log data based on the fault information corresponding to the hard drive failure work order, second log data based on the fault information corresponding to the alarm and error information of the faulty hard drive, third log data corresponding to the normal hard drive, and fourth log data corresponding to the normal hard drive.

[0047] Furthermore, during the training process of the fault prediction model, the negative samples of the faulty hard drive include the first log data and the second log data; the positive samples of the normal hard drive include the third log data and the fourth log data.

[0048] A hard drive failure work order can be understood as a work order in the operation and maintenance system for replacing a hard drive due to a hard drive failure. The hard drive failure work order includes information such as the device identification number of the hard drive, the hard drive identification number, the time of failure reporting, and the type of failure. Hard drive failure alarms and error messages can be understood as alarms and error messages generated by various applications when using the hard drive, indicating that the hard drive has failed or is abnormal.

[0049] The third log data can be understood as the monitoring log data generated by the normal hard drive's own monitoring system. The fourth log data can be the alarm and error logs generated by the application during the use of the normal hard drive.

[0050] It is understandable that the first and third log data can be understood as monitoring log data generated by the hard drive's own monitoring system, while the second and fourth log data can be understood as log data generated by the application during the use of the hard drive.

[0051] It can also be understood that when an application uses the hard drive, it generates alarms or error logs related to the hard drive, indicating that the hard drive is malfunctioning. This malfunction is more severe than the hard drive failure detected by the hard drive's own monitoring system. Therefore, by simultaneously using the first and third log data as negative samples and the second and fourth log data as positive samples to train the fault prediction model, the correlation between samples in the training sample set can be enhanced, which is beneficial to improving the accuracy and recall of the fault prediction model. Accuracy and recall are two common metrics for evaluating model performance; for details, please refer to existing technologies, which will not be elaborated here for simplicity.

[0052] It should be understood that since the first and third log data belong to different data types than the second and fourth log data, features from the first and second log data can be fused to form negative samples, and features from the third and fourth log data can be fused to form positive samples. For ease of understanding, examples illustrating the composition of the corresponding positive and negative samples are provided below.

[0053] For example, the first log data includes features such as raw read error rate, spin-up time, reallocated sectors count, seek error rate, seek time performance, spin retry count, current pending sector count, head flying hours, reported uncorrected errors, high fly writes, and located cycle count. The second log data includes features such as the number of read / write errors on the hard drive. The data corresponding to the features of a specific hard drive in the first and second log data are merged to form a negative sample for that hard drive. Similarly, the third and fourth log data are similar to the first log data, except that the first and second log data are log data corresponding to faulty hard drives, while the third and fourth log data are log data for normal hard drives.

[0054] The hard disk failure prediction method disclosed herein enhances the correlation between samples in the training sample set by simultaneously using monitoring log data generated by its own monitoring system and alarm and error log data generated by the application during hard disk use, which is beneficial to improving the accuracy and recall of the failure prediction model.

[0055] Based on any of the above embodiments, in this embodiment...

[0056] The first log data and the third log data belong to SAMRT log data;

[0057] The second log data and the fourth log data belong to the Message log data.

[0058] SAMRT log data consists of monitoring logs generated by the hard drive's own monitoring system. Message log data consists of logs generated by the system on which the hard drive resides.

[0059] As mentioned above, the first and third log data can both be understood as monitoring log data generated by the hard drive's own monitoring system, while the second and fourth log data can both be understood as log data generated by the application during hard drive usage. Therefore, in one embodiment, the first and third log data belong to SAMRT log data; the second and fourth log data belong to Message log data.

[0060] The hard disk failure prediction method disclosed herein enhances the correlation between samples in the training sample set by simultaneously using monitoring log data (SAMRT log data) generated by its own monitoring system and alarm and error log data (Message log data) generated by the application during the use of the hard disk, which is beneficial to improving the accuracy and recall of the failure prediction model.

[0061] Based on any of the above embodiments, in this embodiment...

[0062] The fault prediction model is dynamically updated according to a preset update strategy.

[0063] Among them, the preset update strategy is used to clarify the conditions for triggering updates to the fault prediction model.

[0064] It is understandable that in existing technologies, because log data is updated in real time and the status of the hard drive changes in real time, the fault prediction model needs to be updated in real time. However, the fault prediction model is trained offline and then run online to predict hard drive failures. This method suffers from a decrease in prediction accuracy due to the inability to update the model in a timely manner. Therefore, the fault prediction model can be dynamically updated according to a preset update strategy.

[0065] The hard disk failure prediction method disclosed herein enables the failure prediction model to predict hard disk failures more accurately and in a timely manner through a preset update strategy, allowing time for hard disk repair or replacement and ensuring the normal operation of hard disk-based services.

[0066] Based on any of the above embodiments, in this embodiment...

[0067] The preset update strategies include: timed task-triggered updates, random-triggered updates, and automatic updates triggered when fault prediction results deviate from evaluation indicators.

[0068] The scheduled update function triggers an update to the hard drive failure prediction model after a predetermined time period. The random update function updates the hard drive failure model randomly. The automatic update function, triggered when the failure prediction result deviates from the evaluation metric, can be understood as: triggering an update to the hard drive failure model when the difference between the evaluation metric and the target value exceeds a first preset threshold.

[0069] It is understandable that users can use one or more of the three preset update strategies mentioned above according to their own needs.

[0070] The hard disk failure prediction method disclosed herein provides three preset update strategies, allowing users to use different model update strategies according to their needs, thereby reducing the problem of decreased model prediction accuracy caused by the inability of the prediction model to be updated in a timely manner.

[0071] Based on any of the above embodiments, in this embodiment, the step of updating according to a preset update strategy further includes:

[0072] In the feature pool, feature analysis and selection are performed on the data obtained by data fusion based on the negative samples of the faulty hard drive and the positive samples of the normal hard drive, and data with correlation exceeding a set threshold are extracted as training dataset.

[0073] Based on the training dataset, different algorithm models in the algorithm pool are trained, and the optimal algorithm model is extracted as the updated fault prediction model; wherein, the different algorithm models belong to the classification algorithms in machine learning.

[0074] Data fusion can be understood as treating positive and negative samples as a whole for analysis and feature selection.

[0075] As mentioned earlier, a positive sample or a negative sample can be composed of at least one feature, and the correlation between features is different. Features with high correlation are beneficial to improving the prediction accuracy of the model. Therefore, in this embodiment, before training the fault prediction model, feature data with correlation exceeding a set threshold are selected as training datasets to train the algorithm model in order to improve the prediction accuracy of the model.

[0076] It is understandable that, since log data changes in real time, the positive and negative samples used to train the fault prediction model will also change in real time with the log data. Therefore, the features whose correlation exceeds the set threshold may be different each time, and the corresponding data may also be different. As a result, the feature data contained in the final training dataset may also be different.

[0077] It is also understandable that, since the training dataset changes in real time, multiple different algorithm models can be used to train the training dataset each time the fault prediction model needs to be updated, and the algorithm model with the best prediction result can be selected. For example, the algorithm model with the largest F-score in model cross-validation can be selected as the updated fault prediction model.

[0078] The algorithm model can be a machine learning algorithm, such as support vector machine, decision tree, random forest, reinforcement learning, XGBoost, or deep learning algorithms. Wherein, TP represents the number of cases where a hard drive is predicted to be faulty and actually is; FP represents the number of cases where a hard drive is predicted to be faulty but actually is normal; and FN represents the number of cases where a hard drive is predicted to be normal but actually is faulty.

[0079] The hard disk failure prediction method disclosed herein improves the prediction accuracy of the algorithm model by analyzing the correlation between features corresponding to positive and negative samples and selecting features with high correlation as the training dataset. Furthermore, by selecting the model with the best training results from multiple different algorithm models as the updated failure prediction model, the method can select the algorithm model with the best prediction performance from multiple different algorithm models based on real-time changes in the training dataset, thereby achieving high prediction accuracy during the prediction process.

[0080] Figure 2 This is the second flowchart illustrating the hard disk failure prediction method provided in this disclosure. Figure 2 The hard disk fault prediction method provided in this disclosure is illustrated by the process of training a fault prediction model and the process of predicting hard disk faults based on the trained fault prediction model. A detailed description follows. Figure 2 The flowchart.

[0081] Step 210: Extract negative sample data and positive sample data from the SMART log and Message log of the faulty hard drive and the SMART log and Message log of the normal hard drive, respectively.

[0082] Specifically, step 210 may include steps 2101 to 2103.

[0083] Step 2101: Retrieve work orders for replacing HDD hard drives caused by HDD hard drive failure from the fault pool of the operation and maintenance system.

[0084] The work order includes: the device identifier (device SN) of the hard drive, the hard drive identifier (hard drive parts_sn or slot number), the fault reporting time, and the fault type.

[0085] Step 2102: Using a period of time before the hard drive failure reporting time as the time interval of the failure occurrence, extract the SMART log corresponding to the failure hard drive identifier from the database, pull the Message log corresponding to the failure hard drive identifier from the host, count the failure disk alarm and error logs in the Message log, merge the counts of the failure disk alarm and error logs in the SMART log corresponding to the failure hard drive identifier and the Message log corresponding to the failure hard drive identifier as the failure log, and label the failure log as a negative sample.

[0086] The time period can be set independently, such as one month, 15 days, or one week.

[0087] Step 2103: Extract the SMART logs corresponding to normal HDD hard drives from the database, pull the Message logs of the corresponding drives from the host, count the alarm and error logs of the corresponding hard drives in the Message logs (count 0 if none), merge the SMART logs of normal drives with the SMART log fault counts into normal logs, and label the normal logs as positive samples.

[0088] It is understandable that there is no specific order of execution between steps 2102 and 2103.

[0089] Step 220: Merge positive and negative samples, perform data preprocessing, select features in the feature pool, and send the selected features into the algorithm pool.

[0090] Specifically, data feature correlation analysis can be performed using principal component analysis (PCA) to obtain feature data with correlation exceeding a set threshold as the training dataset.

[0091] Step 230: Train multiple different algorithm models on the selected feature data in the algorithm pool, and determine the algorithm model with the best prediction result from the multiple different algorithm models as the fault prediction model.

[0092] Step 240: The fault prediction model is launched online. It predicts the HDD hard drive fault status in real time through real-time SMART and Message logs and outputs the hard drive fault prediction results.

[0093] Figure 3 This is the third flowchart illustrating the hard disk failure prediction method provided in this disclosure. Figure 3 The hard disk failure prediction method provided in this disclosure is explained by describing the process of updating the failure prediction model and the process of predicting hard disk failures based on the updated failure prediction model. The following details the process. Figure 3 The flowchart.

[0094] Step 310: Trigger an update of the fault prediction model using one or more of the following triggering methods: scheduled task-triggered update, random-triggered update, or automatic update triggered when the fault prediction result deviates from the evaluation index.

[0095] Step 320: After starting the model update, add SMART log data and Message logs generated in real time later, which were not included in the previous training dataset, to the training dataset before the model update.

[0096] It is understandable that log data is generated in real time, so the previously generated log data was not included in the training dataset. Therefore, it is necessary to add the newly generated log data to the original training dataset to obtain a new dataset.

[0097] Step 330: By filtering the new dataset in the feature pool, feature data with relevance exceeding a set threshold are obtained as the new training dataset.

[0098] Step 340: Train multiple different algorithm models in the algorithm pool with the new training dataset, and determine the algorithm model with the best prediction performance from the multiple different algorithm models as the fault prediction model.

[0099] Step 350: The fault prediction model is launched online. It predicts the HDD hard drive failure status in real time through real-time SMART and Message logs and outputs the hard drive failure prediction results.

[0100] The hard disk failure prediction device provided in this disclosure is described below. The hard disk failure prediction device described below can be referred to in correspondence with the hard disk failure prediction method described above.

[0101] Figure 4 This is a schematic diagram of the hard disk failure prediction device provided in this disclosure, such as... Figure 4 As shown, the hard disk failure prediction device provided in this embodiment includes:

[0102] The first acquisition module 410 is used to acquire real-time log data associated with the hard disk status;

[0103] The first acquisition module 420 is used to input the real-time log data into the fault prediction model to obtain hard disk fault prediction results;

[0104] The fault prediction model is trained based on negative samples of faulty hard drives and positive samples of normal hard drives. The sampling time interval of the negative samples of faulty hard drives is the time period between the time point one predetermined sampling period earlier than the fault reporting time point and the fault reporting time point.

[0105] The hard disk failure prediction device disclosed herein uses features reflecting the hard disk status in log data from a period of time prior to the failure reporting point as negative samples of the failed hard disk to train the failure prediction model. This allows the failure prediction model to learn the features of the hard disk before failure during training, thereby enabling the model to provide a fault warning when predicting real-time log data associated with the hard disk status. This warning is given when the corresponding hard disk in the real-time log data is about to fail or has a minor failure, facilitating timely repair or replacement of the hard disk and ensuring the normal operation of hard disk-based services.

[0106] Based on any of the above embodiments, in this embodiment, the real-time log data includes:

[0107] The log data includes first log data based on the fault information corresponding to the hard drive failure work order, second log data based on the fault information corresponding to the alarm and error information of the faulty hard drive, third log data corresponding to the normal hard drive, and fourth log data corresponding to the normal hard drive.

[0108] Furthermore, during the training process of the fault prediction model, the negative samples of the faulty hard drive include the first log data and the second log data; the positive samples of the normal hard drive include the third log data and the fourth log data.

[0109] The hard disk failure prediction device disclosed herein enhances the correlation between samples in the training sample set by simultaneously using monitoring log data generated by its own monitoring system and alarm and error log data generated by the application during the use of the hard disk, which is beneficial to improving the accuracy and recall of the failure prediction model.

[0110] Based on any of the above embodiments, in this embodiment...

[0111] The first log data and the third log data belong to SAMRT log data;

[0112] The second log data and the fourth log data belong to the Message log data.

[0113] The hard disk failure prediction device disclosed herein enhances the correlation between samples in the training sample set by simultaneously using monitoring log data (SAMRT log data) generated by its own monitoring system and alarm and error log data (Message log data) generated by the application during the use of the hard disk, which is beneficial to improving the accuracy and recall of the failure prediction model.

[0114] Based on any of the above embodiments, in this embodiment...

[0115] The fault prediction model is dynamically updated according to a preset update strategy.

[0116] The hard disk failure prediction device disclosed herein enables the failure prediction model to predict hard disk failures more accurately and in a timely manner through a preset update strategy, allowing time for hard disk repair or replacement and ensuring the normal operation of hard disk-based services.

[0117] Based on any of the above embodiments, in this embodiment, the preset update strategy includes: timed task-triggered update, random-triggered update, and automatic update triggered when the fault prediction result deviates from the evaluation index.

[0118] The hard disk failure prediction device disclosed herein provides three preset update strategies, allowing users to use different model update strategies according to their needs, thereby reducing the problem of decreased model prediction accuracy caused by the inability to update the prediction model in a timely manner.

[0119] Based on any of the above embodiments, in this embodiment, the device further includes:

[0120] The first processing module is used to perform feature analysis and selection on the data obtained by data fusion based on the negative samples of the faulty hard drive and the positive samples of the normal hard drive in the feature pool, and extract the data with correlation exceeding a set threshold as the training dataset.

[0121] The second processing module is used to train different algorithm models in the algorithm pool based on the training dataset, and extract the optimal algorithm model as the updated fault prediction model; wherein, the different algorithm models belong to the classification algorithms in machine learning.

[0122] The hard disk fault prediction device disclosed herein improves the prediction accuracy of the fault prediction model by analyzing the correlation between features corresponding to positive and negative samples and selecting features with high correlation as training datasets to train the algorithm model. Furthermore, by selecting the model with the best training results from multiple different algorithm models as the updated fault prediction model, it can select the algorithm model with the best prediction performance from multiple different algorithm models based on real-time changes in the training dataset, thereby enabling the fault prediction model to achieve high prediction accuracy during the prediction process.

[0123] Figure 5 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 5As shown, the electronic device may include a processor 510, a communications interface 520, a memory 530, and a communication bus 540, wherein the processor 510, communications interface 520, and memory 530 communicate with each other via the communication bus 540. The processor 510 can call logical instructions in the memory 530 to execute a hard disk fault prediction method. This method includes: acquiring real-time log data associated with the hard disk status; inputting the real-time log data into a fault prediction model to obtain a hard disk fault prediction result; wherein the fault prediction model is trained based on negative samples of faulty hard disks and positive samples of normal hard disks; and the sampling time interval of the negative samples of faulty hard disks is the time period between a time point one predetermined sampling period earlier than the fault reporting time point and the fault reporting time point.

[0124] Furthermore, the logical instructions in the aforementioned memory 530 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this disclosure, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this disclosure. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0125] On the other hand, this disclosure also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the hard disk failure prediction method provided by this disclosure. The method includes: acquiring real-time log data associated with the hard disk status; inputting the real-time log data into a failure prediction model to obtain a hard disk failure prediction result; wherein the failure prediction model is trained based on negative samples of faulty hard disks and positive samples of normal hard disks; and the sampling time interval of the negative samples of faulty hard disks is the time period between a time point one predetermined sampling period earlier than the failure reporting time point and the failure reporting time point.

[0126] In another aspect, this disclosure also provides a non-transitory computer-readable storage medium having a computer program stored thereon. When executed by a processor, the computer program is implemented to perform the methods described above to provide the hard disk fault prediction method provided by this disclosure. The method includes: acquiring real-time log data associated with the hard disk status; inputting the real-time log data into a fault prediction model to obtain a hard disk fault prediction result; wherein the fault prediction model is trained based on negative samples of faulty hard disks and positive samples of normal hard disks; and the sampling time interval of the negative samples of faulty hard disks is the time period between a time point one predetermined sampling period earlier than the fault reporting time point and the fault reporting time point.

[0127] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0128] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0129] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this disclosure, and are not intended to limit them. Although this disclosure 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. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this disclosure.

Claims

1. A hard disk failure prediction method, characterized in that, include: Obtain real-time log data associated with hard drive status; The real-time log data is input into the fault prediction model to obtain hard disk fault prediction results; The fault prediction model is trained based on negative samples of faulty hard drives and positive samples of normal hard drives; and the sampling time interval of the negative samples of faulty hard drives is the time period between the time point one predetermined sampling period earlier than the fault reporting time point and the fault reporting time point. The real-time log data includes: The log includes first log data based on the fault information corresponding to the hard drive failure work order, second log data based on the fault information corresponding to the alarm and error information of the faulty hard drive, third log data corresponding to the normal hard drive, and fourth log data corresponding to the normal hard drive. Furthermore, during the training process of the fault prediction model, the negative samples of the faulty hard drive include the first log data and the second log data; the positive samples of the normal hard drive include the third log data and the fourth log data. The first log data and the third log data are monitoring log data generated by the hard drive's own monitoring system, while the second log data and the fourth log data are log data generated by the application during the use of the hard drive. The fault prediction model is dynamically updated according to a preset update strategy; The step of updating according to a preset update strategy further includes: In the feature pool, feature analysis and selection are performed on the data obtained by data fusion based on the negative samples of the faulty hard drive and the positive samples of the normal hard drive, and data with correlation exceeding a set threshold are extracted as training dataset. Based on the training dataset, different algorithm models in the algorithm pool are trained, and the optimal algorithm model is extracted as the updated fault prediction model; wherein, the different algorithm models belong to the classification algorithms in machine learning.

2. The hard disk failure prediction method according to claim 1, characterized in that, The first log data and the third log data belong to SAMRT log data; The second log data and the fourth log data belong to the Message log data.

3. The hard disk failure prediction method according to claim 1, characterized in that, The preset update strategies include: timed task-triggered updates, random-triggered updates, and automatic updates triggered when fault prediction results deviate from evaluation indicators.

4. A hard disk failure prediction device, characterized in that, include: The first acquisition module is used to acquire real-time log data associated with the hard disk status; The first acquisition module is used to input the real-time log data into the fault prediction model to obtain hard disk fault prediction results; The fault prediction model is trained based on negative samples of faulty hard drives and positive samples of normal hard drives; and the sampling time interval of the negative samples of faulty hard drives is the time period between the time point one predetermined sampling period earlier than the fault reporting time point and the fault reporting time point. The real-time log data includes: The log includes first log data based on the fault information corresponding to the hard drive failure work order, second log data based on the fault information corresponding to the alarm and error information of the faulty hard drive, third log data corresponding to the normal hard drive, and fourth log data corresponding to the normal hard drive. Furthermore, during the training process of the fault prediction model, the negative samples of the faulty hard drive include the first log data and the second log data; the positive samples of the normal hard drive include the third log data and the fourth log data. The first log data and the third log data are monitoring log data generated by the hard drive's own monitoring system, while the second log data and the fourth log data are log data generated by the application during the use of the hard drive. The fault prediction model is dynamically updated according to a preset update strategy; The device further includes: The first processing module is used to perform feature analysis and selection on the data obtained by data fusion based on the negative samples of the faulty hard drive and the positive samples of the normal hard drive in the feature pool, and extract the data with correlation exceeding a set threshold as the training dataset. The second processing module is used to train different algorithm models in the algorithm pool based on the training dataset, and extract the optimal algorithm model as the updated fault prediction model; wherein, the different algorithm models belong to the classification algorithms in machine learning.

5. An electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the hard disk failure prediction method as described in any one of claims 1 to 3.

6. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the hard disk failure prediction method as described in any one of claims 1 to 3.

7. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the hard disk failure prediction method as described in any one of claims 1 to 3.

Images