An SSD Fault Prediction Method, Device and Medium Based on Multi-Task Learning

Abstract

The present invention provides an SSD fault prediction method, device and medium based on multi-task learning, which relates to the technical field of machine learning. The method includes: preprocessing and normalizing the data in the obtained hard disk logs, obtaining a plurality of initial SSDs and the initial data of each initial SSD, obtaining a list of target age groups and a list of key features corresponding to each target age group, obtaining a list of target SSDs corresponding to each target age group and a list of target key feature values corresponding to each target SSD, obtaining a target multi-task learning model, obtaining a target stacking result, obtaining a first other classifier and a second other classifier, and obtaining a first other prediction result; obtaining a second other prediction result, and based on the target stacking result, the first other prediction result and the second other prediction result, obtaining a final prediction result corresponding to each target SSD, so as to improve the accuracy of SSD fault prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of machine learning, and particularly to an SSD fault prediction method, device and medium based on multi-task learning. Background Art

[0002] With the wide application of solid-state drives (SSDs) in cloud storage systems and data centers, the high performance and low power consumption of SSDs have gradually replaced traditional hard disk drives (HDDs). However, the frequent occurrence of SSD failures has become an important factor restricting the stability and reliability of cloud storage systems. Statistical data shows that approximately 80% of the hardware failures in data centers are directly related to hard disk failures, and the special failure modes of SSDs make it more complex to predict their failures.

[0003] Existing hard disk fault prediction methods usually rely on the Self-Monitoring, Analysis and Reporting Technology (SMART) logs of SSD devices. This log provides a series of parameters reflecting the hard disk operation status, such as read / write error rate, temperature, reallocated sector count, etc. Researchers generally use machine learning or deep learning techniques to extract features from SMART logs and build prediction models. These methods usually adopt the following technical means:

[0004] (1) Static feature selection method: Researchers select several key features by analyzing the statistical distribution and correlation of SMART attributes and build a classification model. Common feature selection methods include Pearson correlation coefficient, Spearman correlation coefficient, random forest, etc.

[0005] (2) Single model prediction: Rely on a single classification model to predict all hard disk data. Most existing studies use traditional machine learning models (such as Support Vector Machine SVM, Gradient Boosting Decision Tree GBDT) or deep learning models (such as Convolutional Neural Network CNN, Recurrent Neural Network RNN) to predict hard disk failures.

[0006] (3) Laboratory data analysis: Many studies conduct load tests on SSDs in a simulated environment, record SMART log data under working conditions, and use the data in the laboratory for model training and performance evaluation.

[0007] Although the above methods have improved the accuracy of hard disk fault prediction to a certain extent, they still face many challenges:

[0008] (1) Feature importance drift problem

[0009] The failure characteristics of SSDs change over time and with the increase in service life, that is, the importance of the characteristics drifts with the aging of the hard disk and the change of the working environment. Most existing methods are based on static feature selection strategies and are difficult to adapt to this dynamic change, thus reducing the long-term prediction performance of the model. As the service life increases, the weights of the failure characteristics change; the reasons are analyzed as follows: at different stages of the hard disk life cycle, the dominant factors of the failure characteristics may be very different. For example, the failures of new hard disks are mostly caused by manufacturing defects, while after long-term use, read / write errors and uncorrectable sector counts become the main failure factors. The data noise and outliers (such as intermittent errors) during long-term operation are not fully processed, further reducing the accuracy of feature selection.

[0010] (2) Limitations of a single model;

[0011] The performance of a single model in dealing with complex failure modes is limited, especially in the case of diverse data distributions, overfitting or underfitting is likely to occur. The reasons are analyzed as follows: complex SSD failure modes (such as multi-stage failures and intermittent errors) require the model to capture multi-level feature relationships, while a single model is difficult to cover all cases. Some machine learning models are sensitive to a small number of samples. For example, the support vector machine (SVM) has poor robustness to sample imbalance and is prone to misjudgment in the case of extremely low actual failure ratios. Summary of the Invention

[0012] In view of the above technical problems, the technical solution adopted by the present invention is: a method for predicting SSD failures based on multi-task learning, the method comprising the following steps:

[0013] S100, preprocess and standardize the data in the obtained hard disk logs to obtain a number of initial SSDs and the initial data of each initial SSD;

[0014] S200, obtain a list of target age groups and a list of key features corresponding to each target age group, the list of target age groups includes a target age groups corresponding to different preset interval ranges of SSD usage duration, and the list of key features includes a number of key features for predicting SSD failures, and the key features are obtained based on the hard disk logs of the SSD;

[0015] S300. Based on the usage duration included in the initial data, obtain the target SSD list corresponding to each target age group and the set of target key feature value lists corresponding to the target SSD list; the target SSD list includes several target SSDs, and the target SSDs in the same target SSD list are the initial SSDs belonging to the same target age group, the set of target key feature value lists includes several target key feature value lists, and the target key feature value list includes the target key feature values of the target SSD corresponding to several specified features; wherein, the specified feature is the key feature corresponding to the target age group to which the target SSD belongs.

[0016] S400. Input the set of target key feature value lists of the target SSD lists corresponding to a target age groups into the a target sub - learning models included in the target multi - task learning model respectively, and obtain the fault prediction results output by each target sub - learning model; and use the a fault prediction results as meta - features to input into the target meta - model for fault prediction again to obtain the target stacking result.

[0017] S500. Obtain the first other classifier and the second other classifier, input the initial data of the target SSD into the first other classifier for fault prediction to obtain the first other prediction result; and input the initial data of the target SSD into the second other classifier for fault prediction to obtain the second other prediction result.

[0018] S600. Based on the target stacking result, the first other prediction result and the second other prediction result, obtain the final prediction result corresponding to each target SSD.

[0019] According to another aspect of the present invention, there is provided a non - transitory computer - readable storage medium, in which at least one instruction or at least one program segment is stored, and the at least one instruction or the at least one program segment is loaded and executed by a processor to implement the foregoing method.

[0020] According to yet another aspect of the present invention, there is provided an electronic device, including a processor and the foregoing non - transitory computer - readable storage medium.

[0021] The present invention has at least the following beneficial effects: In summary, preprocess and standardize the data in the obtained hard disk logs, obtain a number of initial SSDs and the initial data of each initial SSD, obtain a list of target age groups and a list of key features corresponding to each target age group, based on the usage duration of the target SSD, obtain a list of target SSDs corresponding to each target age group and a set of target key feature value lists corresponding to the list of target SSDs, input the set of target key feature value lists of the list of target SSDs corresponding to a target age groups into a target multitask learning model including a target sub-learning models, obtain a fault prediction result output by each target sub-learning model; and use the a fault prediction results as meta-features to input into a target meta-model for fault prediction again, obtain a target stacking result, obtain a first other classifier and a second other classifier, input the initial data of the target SSD into the first other classifier for fault prediction, obtain a first other prediction result; and input the initial data of the target SSD into the second other classifier for fault prediction, obtain a second other prediction result, based on the target stacking result, the first other prediction result and the second other prediction result, obtain a final prediction result corresponding to each target SSD. The present invention classifies and uses different key features based on the hard disk usage duration, improving the accuracy of SSD fault prediction; the present invention performs fault prediction through multiple models, effectively integrating the prediction results of the multitask model and the additional classifier, significantly reducing the overfitting risk of a single model, and improving the robustness of the fault prediction result. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can be obtained based on these drawings.

[0023] Figure 1 It is a flowchart of a method for predicting SSD faults based on multitask learning provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0024] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

[0025] An embodiment of the present invention provides a method for predicting SSD faults based on multitask learning, as Figure 1 shown, the method includes the following steps:

[0026] S100 preprocesses and normalizes the data in the obtained hard disk logs to obtain a number of initial SSDs and the initial data of each initial SSD.

[0027] Specifically, the data in the hard disk logs can be SMART log data; the preprocessing includes deleting attribute data with no recorded values or fixed values; the preprocessing also includes deleting data corresponding to preset initial features with low correlation, and the preset initial features with low correlation can be determined according to the actual needs of users to ensure that data with strong SSD failure correlation is screened out. In an embodiment of the present invention, the quartile method is used to screen out irrelevant abnormal data points, such as negative values or abnormally large values caused by acquisition errors.

[0028] Specifically, the normalization process is to perform normalization processing. Those skilled in the art know that any method of normalizing data in the prior art belongs to the protection scope of the present invention and will not be elaborated here. The data in the obtained hard disk logs is normalized to eliminate the numerical differences between different hard disk models and improve the comparability of the data.

[0029] S200 obtains a list of target age groups and a list of key features corresponding to each target age group. The list of target age groups includes a target age groups corresponding to different preset interval ranges of SSD usage duration. The list of key features includes a number of key features for predicting SSD failures, and the key features are obtained based on the hard disk logs of the SSDs.

[0030] Among them, the list of target age groups K = {K1, K2,..., K r ,..., K a}, K r is the r-th target age group, that is, the preset range of the usage duration of the r-th SSD. The value range of r is from 1 to a, and a is the number of target age groups; the list of key features AK r corresponding to K r ={AK r1 , AK r2 ,..., AK ry ,..., AK rp(r)}, AK ry is the y-th key feature corresponding to the r-th target age group. The value range of y is from 1 to p(r), and p(r) is the number of key features corresponding to the r-th target age group.

[0031] S300. Based on the usage duration included in the initial data, obtain the target SSD list corresponding to each target age group and the set of target key feature value lists corresponding to the target SSD list; the target SSD list includes a number of target SSDs, and the target SSDs in the same target SSD list are the initial SSDs belonging to the same target age group. The set of target key feature value lists includes a number of target key feature value lists, and the target key feature value list includes the target key feature values of the target SSD corresponding to several specified features; wherein, the specified feature is the key feature corresponding to the target age group to which the target SSD belongs.

[0032] Among them, the target SSD list E corresponding to the r-th target age group r ={E r1 , E r2 , …, E rt , …, E rk（r）}, E rt is the t-th target SSD of the r-th target age group, and the value range of t is from 1 to k(r), where k(r) is the number of target SSDs in the r-th target age group. The target key feature value list AE rt corresponding to E rt ={AE rt1 , AE rt2 , …, AE rty , …, AE rtp(r)}, AE rty is the value of the y-th key feature corresponding to the t-th SSD of the r-th age group.

[0033] S400. Respectively input the sets of target key feature value lists of the target SSD lists corresponding to a target age groups into the a target sub-learning models included in the target multi-task learning model to obtain the fault prediction results output by each target sub-learning model; and use the a fault prediction results as meta-features to input into the target meta-model for fault prediction again to obtain the target stacked result.

[0034] Specifically, input E1 to E a into the a target sub-learning models respectively to obtain the list of target prediction results F = {F1, F2, …, F r , …, F a} of the s target sub-learning models. F r is the fault prediction result of E r input into the r-th target self-learning model.

[0035] S500, Obtain the first other classifier and the second other classifier, input the initial data of the target SSD into the first other classifier for fault prediction to obtain the first other prediction result; and input the initial data of the target SSD into the second other classifier for fault prediction to obtain the second other prediction result.

[0036] Among them, the first other classifier is the target GBDT classifier, and the second other classifier is the target XGBoost classifier.

[0037] Specifically, construct an initial GBDT classifier, and use the sample data of the sample SSD and the true fault result of the sample SSD to train the initial GBDT classifier to obtain the target GBDT classifier; construct an initial XGBoost classifier, and use the sample data of the sample SSD and the true fault result of the sample SSD to train the initial XGBoost classifier to obtain the target XGBoost classifier.

[0038] S600, Based on the target stacking result, the first other prediction result and the second other prediction result, obtain the final prediction result corresponding to each target SSD.

[0039] Specifically, based on the target stacking result, the first other prediction result and the second other prediction result, perform voting decision-making to obtain the fault label of each target SSD, so as to obtain the final prediction result corresponding to each target SSD, and improve the robustness and comprehensive performance of the SSD fault prediction method based on multi-task learning of the present invention.

[0040] Among them, the fault label of the target SSD is "1" or "0". Optionally, when the fault label is "1", the target SSD is in a fault state, and when the fault label is "0", the target SSD is in a non-fault state.

[0041] In summary, the data in the obtained hard disk logs is preprocessed and standardized, several initial SSDs and the initial data of each initial SSD are obtained, a list of target age groups and a list of key features corresponding to each target age group are obtained. Based on the usage duration of the target SSD, a list of target SSDs corresponding to each target age group and a set of target key feature value lists corresponding to the list of target SSDs are obtained. The sets of target key feature value lists of the lists of target SSDs corresponding to a target age groups are respectively input into the a target sub-learning models included in the target multi-task learning model to obtain the fault prediction results output by each target sub-learning model; and the a fault prediction results are used as meta-features to input into the target meta-model for fault prediction again to obtain the target stacking result. A first other classifier and a second other classifier are obtained. The initial data of the target SSD is input into the first other classifier for fault prediction to obtain the first other prediction result; and the initial data of the target SSD is input into the second other classifier for fault prediction to obtain the second other prediction result. Based on the target stacking result, the first other prediction result and the second other prediction result, the final prediction result corresponding to each target SSD is obtained. The present invention classifies according to the hard disk usage duration and uses different key features, improving the accuracy of SSD fault prediction; the present invention performs fault prediction through multiple models, effectively integrating the prediction results of the multi-task model and the additional classifier, significantly reducing the overfitting risk of a single model, and improving the robustness of the fault prediction result.

[0042] Furthermore, the present invention jointly learns the fault characteristics of different hard disk models and age groups through a multi-task learning model, makes full use of the correlation between heterogeneous data, and effectively solves the problem of insufficient applicability of existing models in heterogeneous hard disk environments. The modular design of the architecture of the present invention makes it easy to integrate into existing cloud storage monitoring systems and has high practical application value. The multi-task multi-layer stacking module adopts a distributed parallel computing framework, supports the efficient processing of large-scale hard disk data, and enables the present invention to operate stably in an actual production environment. In addition, the method of the present invention does not depend on specific hardware or data sources, can be flexibly deployed in various data centers or storage environments, and reduces the implementation cost. By predicting potential hard disk failures in advance and taking preventive maintenance measures, the present invention can significantly reduce the system downtime and data loss risk caused by hard disk failures.

[0043] Experimental results show that on the SSD dataset, the dynamic feature selection mechanism of the present invention improves the F1 score of the prediction model by more than 13% and the recall rate by 10%-15%. In addition, in a long-term running hard disk environment, the present invention can significantly reduce the false negative rate and false positive rate, ensuring the dynamic capture ability of the model for hard disk fault characteristics.

[0044] On the mixed dataset, the prediction accuracy of the present invention is improved by more than 18% compared with the traditional method. Especially when dealing with hard disk data involving multiple manufacturers and models, it still maintains high stability and accuracy. By sharing hidden layer features among tasks, the present invention enhances the adaptability of the model to data of unknown hard disk models and realizes the efficient migration of the prediction model.

[0045] In the experiment, the false alarm rate of the present invention is reduced to less than 2%, while the existing models are usually between 3% - 5%. The voting integration mechanism effectively integrates the advantages of different classifiers, ensures more stable prediction results, and reduces extreme prediction situations caused by biases of specific models.

[0046] In the test scenario, the prediction accuracy of the present invention reduces the system operation and maintenance cost by about 20% - 30%. It can effectively reduce emergency replacements and data recovery operations caused by sudden hard disk failures, and improve the overall operation efficiency and service quality of the data center.

[0047] Specifically, the target age group and the corresponding list of key features of the target age group are obtained through the following steps:

[0048] S210, Obtain a number of sample SSDs and the sample data of each sample SSD, preprocess and standardize the sample data to obtain a list of sample feature values for each sample SSD. The list of sample feature values includes the values of several preset initial features corresponding to the sample SSD, and the preset initial features are obtained based on the sample data. Specifically, the preset initial features include: usage duration, hard disk model, read / write error rate, temperature, reallocated sector count, etc. The preset initial features are features reflecting the running state of the hard disk.

[0049] S220, Obtain a list of preset usage time periods, and the list of preset usage time periods includes several preset usage time periods, where the time intervals of different preset usage time periods are the same.

[0050] S230, Based on the list of preset usage time periods and the hard disk models of the sample SSDs, divide the sample SSDs to obtain a set of sample age group lists. The set of sample age group lists includes several sample age group lists, and each sample age group list includes several sample SSDs that are in the same preset usage time period and have the same hard disk model;

[0051] S240, Based on the sample age group list, obtain the importance degree of each preset initial feature in each sample age group, sort them from large to small according to the importance degree, and mark the first b preset initial features as the intermediate features of this sample age group, so as to obtain the corresponding list of intermediate features for each sample age group. The list of intermediate features includes several intermediate features.

[0052] Specifically, b can be determined according to actual requirements.

[0053] S250, obtain the similarity coefficient of the intermediate feature lists of any two adjacent sample age groups. If the similarity coefficient is less than the preset similarity threshold, take the two sample age groups as two target age groups respectively, and take the intermediate feature lists corresponding to the two sample age groups as the key feature lists corresponding to the target age groups respectively.

[0054] Specifically, take the Jaccard coefficient of the intermediate feature lists of any two adjacent sample age groups as the similarity coefficient.

[0055] S260, if the similarity coefficient is not less than the preset similarity threshold, merge the two sample age groups into one target age group, and merge the intermediate feature lists of the two sample age groups as the key feature list corresponding to the target age group. By merging the two sample age groups and the two intermediate key feature lists, redundancy is reduced and the accuracy of key feature selection is improved.

[0056] In summary, obtain the sample data of several sample SSDs and each sample SSD, preprocess and standardize the sample data, obtain the sample feature value list of each sample SSD, obtain the preset usage duration interval, divide the sample SSDs based on the preset usage duration interval and the hard disk model to obtain the sample age group list set, based on the sample age group list, obtain the importance degree of each preset initial feature in each sample age group, sort the importance degrees in each sample age group from large to small, and mark the top b preset initial features as the intermediate features of the sample age group, so as to obtain the intermediate feature list corresponding to each sample age group, obtain the similarity coefficient of the intermediate feature lists of any two adjacent sample age groups. If the similarity coefficient is less than the preset similarity threshold, take the two sample age groups as two target age groups respectively, and take the intermediate feature list corresponding to the target age group as the key feature list; if the similarity coefficient is not less than the preset similarity threshold, merge the two sample age groups into one target age group, and merge the intermediate feature lists of the two sample age groups as the key feature list corresponding to the target age group. The present invention screens the preset initial features based on the importance degree, and more accurately obtains the key features.

[0057] Further, S220 obtains the time interval of the preset usage time period through the following steps:

[0058] S221, based on the sample data, obtain the probability density function of the sample data by the kernel density estimation method. Specifically, the probability density function is the function corresponding to the probability density estimation curve.

[0059] S222. Based on the sample data, obtain the sample mutation points of the sample data through a change point detection algorithm.

[0060] Specifically, those skilled in the art know that any change point detection algorithm in the prior art belongs to the protection scope of the present invention. For example, the change point detection algorithm is a method based on machine learning.

[0061] S223. Based on the probability density function and the sample mutation points, determine the time interval of the preset usage period. Specifically, combine the probability density function and the sample mutation points to analyze the specific changes in the distribution of the sample data before and after the change point; determine the time interval of the preset usage period based on the change situation of the sample data. For example, use the usage duration interval between two adjacent change points as the time interval of the preset usage period.

[0062] Further, S230. Obtaining the importance degree of each preset initial feature in each sample age group based on the sample age group list includes:

[0063] S231. Obtain a list of importance degree calculation methods, and the list of importance degree calculation methods includes several importance degree calculation methods.

[0064] Among them, the importance degree calculation methods include: the calculation method of Spearman correlation coefficient, the calculation method of pearson correlation coefficient. Specifically, the importance degree calculation methods also include: the calculation method of J-Index coefficient, the method of calculating feature contribution degree by random forest.

[0065] S232. Obtain the true failure result of the sample SSD, and based on the true failure result of the sample SSD and the list of sample feature values of the sample SSD corresponding to the sample age group list, use the importance degree calculation method to obtain the importance degree of the preset initial feature.

[0066] Specifically, take the average value of the importance degrees obtained by multiple importance degree calculation methods as the importance degree of a preset initial feature. Obtain the importance degree of the preset initial feature through multiple importance degree calculation methods to improve accuracy and reliability.

[0067] Furthermore, the following steps are also included to obtain the target multi-task learning model:

[0068] S001. Based on the sample data of the sample SSD, obtain the intermediate SSD list corresponding to each target age group and the intermediate key feature value list corresponding to each intermediate SSD. The intermediate SSD list includes several intermediate SSDs corresponding to the sample SSDs in the same target age group, and the intermediate key feature value list includes the target key feature values of several candidate features corresponding to the intermediate SSD. Among them, the candidate feature is the key feature corresponding to the target age group of the usage duration of the intermediate SSD.

[0069] S002. Construct an initial multi-task learning model, and the initial multi-task learning model includes a initial sub-learning models.

[0070] S003. Based on the true failure results of the intermediate SSDs and the intermediate key feature value lists corresponding to each intermediate SSD, train the initial multi-task learning model to obtain the target multi-task learning model.

[0071] Specifically, the present invention uses the target multi-task learning model. The core idea is to capture the correlation between different tasks by sharing hidden parameters and improve the generalization ability of the model. The shared hidden layer structure extracts general features, and at the same time retains the exclusive features of each age group according to the key feature list, and trains multiple initial sub-learning models.

[0072] Specifically, during the training process, the cross-entropy loss function and the Adam optimizer are used to jointly optimize the parameters of all tasks, so that the shared hidden layer fully learns general features, and at the same time retains the specific feature representations of each task, and completes the training of the multi-task learning model.

[0073] An embodiment of the present invention also provides a non-transitory computer-readable storage medium, which can be set in an electronic device to store at least one instruction or at least one program related to a method in the method embodiment. The at least one instruction or the at least one program is loaded and executed by the processor to implement the method provided in the above embodiment.

[0074] An embodiment of the present invention also provides an electronic device, including a processor and the foregoing non-transitory computer-readable storage medium.

[0075] Although some specific embodiments of the present invention have been described in detail by way of examples, those skilled in the art should understand that the above examples are only for illustration and not for limiting the scope of the present invention. Those skilled in the art should also understand that various modifications can be made to the embodiments without departing from the scope and spirit of the present invention.

Claims

1. An SSD fault prediction method based on multi-task learning, characterized in that, The method includes the following steps: S100. Preprocess and standardize the data in the obtained hard disk log to obtain a number of initial SSDs and the initial data of each initial SSD; S200. Obtain a target age group list and a corresponding key feature list for each target age group. The target age group list includes a target age groups corresponding to different preset interval ranges of SSD usage duration. The key feature list includes a number of key features for predicting SSD failures, and the key features are obtained based on the hard disk log of the SSD; S300. Based on the usage duration included in the initial data, obtain a target SSD list corresponding to each target age group and a set of target key feature value lists corresponding to the target SSD list. The target SSD list includes a number of target SSDs, and the target SSDs in the same target SSD list are the initial SSDs belonging to the same target age group. The set of target key feature value lists includes a number of target key feature value lists. The target key feature value list includes the target key feature values of the target SSD corresponding to a number of specified features, where the specified features are the key features corresponding to the target age group to which the target SSD belongs; S400. Input the set of target key feature value lists of the target SSD lists corresponding to the a target age groups into the a target sub-learning models included in the target multi-task learning model respectively to obtain the failure prediction results output by each target sub-learning model; and use the a failure prediction results as meta-features to input into the target meta-model for failure prediction again to obtain the target stacked result; S500. Obtain a first other classifier and a second other classifier. Input the initial data of the target SSD into the first other classifier for failure prediction to obtain a first other prediction result; and input the initial data of the target SSD into the second other classifier for failure prediction to obtain a second other prediction result; S600. Based on the target stacked result, the first other prediction result, and the second other prediction result, obtain the final prediction result corresponding to each target SSD; Obtain the target age group and the corresponding key feature list of the target age group through the following steps: S210. Obtain a number of sample SSDs and the sample data of each sample SSD. Preprocess and standardize the sample data to obtain a sample feature value list for each sample SSD. The sample feature value list includes the values of a number of preset initial features corresponding to the sample SSD, and the preset initial features are obtained based on the sample data; S220. Obtain a preset usage time period list, which includes a number of preset usage time periods, where the time intervals of different preset usage time periods are the same; S230. Based on the preset usage time period list and the hard disk model of the sample SSD, divide the sample SSD to obtain a set of sample age group lists, which includes a number of sample age group lists. The sample age group list includes a number of sample SSDs that are in the same preset usage time period and have the same hard disk model; S240. Based on the sample age group list, obtain the importance degree of each preset initial feature in each sample age group, sort them in descending order of importance degree, and mark the top b preset initial features as the intermediate features of the sample age group, so as to obtain the intermediate feature list corresponding to each sample age group. The intermediate feature list includes several intermediate features. S250. Obtain the similarity coefficient of the intermediate feature lists of any two adjacent sample age groups. If the similarity coefficient is less than the preset similarity threshold, take the two sample age groups as two target age groups respectively, and take the intermediate feature lists corresponding to the two sample age groups as the key feature lists corresponding to the target age groups respectively. S260. If the similarity coefficient is not less than the preset similarity threshold, merge the two sample age groups into one target age group, and merge the intermediate feature lists of the two sample age groups as the key feature list corresponding to the target age group.

2. The SSD fault prediction method based on multi-task learning according to claim 1, wherein, S220 obtains the time interval of the preset usage period through the following steps: S221. Based on the sample data, obtain the probability density function of the sample data through the kernel density estimation method. S222. Based on the sample data, obtain the sample mutation points of the sample data through the change point detection algorithm. S223. Based on the probability density function and the sample mutation points, determine the time interval of the preset usage period.

3. The SSD fault prediction method based on multi-task learning according to claim 1, characterized in that Take the Jaccard coefficient of the intermediate feature lists of any two adjacent sample age groups as the similarity coefficient.

4. The SSD fault prediction method based on multi-task learning according to claim 1, characterized in that In S230, obtaining the importance degree of each preset initial feature in each sample age group based on the sample age group list includes: S231. Obtain the importance degree calculation method list, and the importance degree calculation method list includes several importance degree calculation methods. S232. Obtain the true failure results of the sample SSD, and based on the true failure results of the sample SSD and the sample feature value list of the sample SSD corresponding to the sample age group list, use the importance degree calculation method to obtain the importance degree of the preset initial feature.

5. The SSD fault prediction method based on multi-task learning according to claim 4, wherein The importance degree calculation methods include: the calculation method of the Spearman correlation coefficient and the calculation method of the pearson correlation coefficient.

6. The SSD fault prediction method based on multi-task learning according to claim 4, wherein, It also includes the following steps to obtain the target multi-task learning model: S001. Based on the sample data of the sample SSD, obtain the intermediate SSD list corresponding to each target age group and the intermediate key feature value list corresponding to each intermediate SSD. The intermediate SSD list includes several intermediate SSDs corresponding to the sample SSDs in the same target age group. The intermediate key feature value list includes the target key feature values of several candidate features corresponding to the intermediate SSD. Among them, the candidate feature is the key feature corresponding to the target age group corresponding to the usage duration of the intermediate SSD. S002. Construct an initial multi-task learning model, and the initial multi-task learning model includes a initial sub-learning models. S003. Based on the true failure results of the intermediate SSD and the intermediate key feature value list corresponding to each intermediate SSD, train the initial multi-task learning model to obtain the target multi-task learning model.

7. The SSD fault prediction method based on multi-task learning according to claim 1, wherein The first other classifier is the target GBDT classifier, and the second other classifier is the target XGBoost classifier.

8. A non-transitory computer-readable storage medium storing at least one instruction or at least one program segment, characterized in that, The at least one instruction or the at least one program segment is loaded and executed by a processor to implement the SSD fault prediction method based on multi-task learning according to any one of claims 1-7.

9. An electronic device, characterized in that, It includes a processor and the non-transitory computer-readable storage medium described in claim 8.

Images