A hard disk failure prediction method and system based on multi-task learning

Abstract

The application relates to a hard disk fault prediction method and system based on multi-task learning, which comprises the following steps: obtaining hard disk information of multiple hard disks, obtaining a training sample set according to the hard disk information; based on a representation learning network and a metric learning network, a classification and regression dual-task hard disk fault prediction model is obtained by training the training sample set; when it is necessary to predict the fault of a target hard disk, the hard disk information of the target hard disk is taken as a to-be-predicted sample, the to-be-predicted sample and a training sample in the training sample set are formed into a sample input pair; the sample input pair is input into the classification and regression dual-task hard disk fault prediction model, and a fault prediction result of the target hard disk is obtained. The representation learning network and the metric learning network are combined, the representation learning network learns the representation features of the hard disk, so that potential fault signs can be better captured, the metric learning network is used to learn the similarities and differences between hard disks of the same manufacturer and the same type, and the risk of hard disk fault can be more accurately predicted.

Description

Technical Field

[0001] This invention relates to the field of storage technology, and in particular to a hard disk fault prediction method and system based on multi-task learning. Background Technology

[0002] With the development of artificial intelligence, the total amount of global data storage is growing exponentially, placing higher demands on the reliability and availability of storage systems. Hard drives are currently the most widely used storage devices. Hard drive failure is a common problem in computer systems, potentially leading to data loss and system downtime, causing serious losses to businesses and individuals. Therefore, early detection of hard drive health status and early warning before hard drive failures occur can help maintenance personnel replace or repair disks that are about to fail based on model predictions. This effectively avoids the impact of equipment instability on business operations when failures are imminent, alleviates the pressure on maintenance personnel, and reduces the investment in data recovery.

[0003] Currently, most research employs basic machine learning models, such as decision trees and random forests, which are simple to train but have poor generalization capabilities. Some studies use deep learning methods, such as backpropagation neural networks and recurrent neural networks, achieving good results in hard drive fault prediction tasks based on certain signals. However, their prediction performance is poor when faced with small sample hard drive datasets, and they cannot use general models to adapt to different manufacturers and models of hard drives. Summary of the Invention

[0004] This invention provides a hard disk failure prediction method and system based on multi-task learning. It combines representation learning network and metric learning network to train a dual-task hard disk failure prediction model of classification and regression. The representation learning network learns the representational features of the hard disk to better capture potential failure signs, and the metric learning network learns the similarity and differences between hard disks of the same manufacturer and model to more accurately predict the risk of hard disk failure.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0006] Firstly, a hard disk fault prediction method based on multi-task learning is provided, including:

[0007] Obtain hard drive information from multiple hard drives, and then obtain a training sample set based on the hard drive information;

[0008] Based on representation learning network and metric learning network, a dual-task hard disk failure prediction model of classification and regression is obtained by training a training sample set.

[0009] When it is necessary to predict the failure of the target hard drive, the hard drive information of the target hard drive is used as the sample to be predicted, and the sample to be predicted is combined with the training samples in the training sample set to form a sample input pair.

[0010] The sample is input into a dual-task hard disk failure prediction model that combines classification and regression to obtain the failure prediction result for the target hard disk.

[0011] Furthermore, based on representation learning networks and metric learning networks, a dual-task hard disk failure prediction model for classification and regression is trained using a training sample set, including:

[0012] A representation learning network is used to capture the general hard disk information of all training samples in the training sample set. The general hard disk information includes hard disk manufacturer information, hard disk model information and hard disk fault information.

[0013] Hard drive failure information is used as a label for binary classification, and a failure classification task model is trained.

[0014] Based on the triplet sample search strategy and general information of the hard disk, triplet learning is performed on all training samples to construct a training sample pool.

[0015] A metric learning network is used to perform metric learning on the training sample pool, and the triple loss is optimized to train a fault regression task model.

[0016] Based on the fault classification task model and the fault regression task model, a dual-task hard disk fault prediction model of classification and regression is constructed.

[0017] Furthermore, based on the triplet sample search strategy and general disk information, triplet learning is performed on all training samples to construct a training sample pool, including:

[0018] Based on the triplet sample search strategy and general hard drive information, P triplet training samples of different hard drive models from the same hard drive manufacturer are randomly selected from all training samples in the preset training batch.

[0019] For each hard drive model, K positive samples and K negative samples are randomly selected as sample pairs for the ternary training samples. Positive samples represent training samples with the same hard drive fault information as the ternary training samples, and negative samples represent training samples with different hard drive fault information than the ternary training samples.

[0020] A training sample pool is constructed based on the ternary training samples and their corresponding sample pairs.

[0021] Furthermore, a metric learning network is used to perform metric learning on the training sample pool, and the triplet loss is optimized to train a fault regression task model, including:

[0022] A metric learning network is used to perform metric learning on the training sample pool to obtain the distance value between the triple training sample and each training sample in the sample pair.

[0023] Based on the training samples and distance values, triplet pairs are formed, and the triplet loss is calculated.

[0024] The fault regression task model is obtained by training a metric learning network and triplet loss.

[0025] Furthermore, based on the triple training samples and distance values, triples are formed, and the triple loss is calculated, including:

[0026] Based on the triple training samples and distance values, the most difficult positive sample and the most difficult negative sample corresponding to the triple training sample are selected to form a triple; the most difficult positive sample is the positive sample that is farthest from the triple training sample, and the most difficult negative sample is the negative sample that is closest to the triple training sample.

[0027] The triplet loss is calculated using the triplet loss function, which is expressed as follows:

[0028]

[0029] in, Indicates the loss of the triplet. This represents the training sample set of the faulty hard drive. This represents the training sample set for healthy hard drives, where the ternary training samples... a When it is a faulty hard drive, p for Training samples of faulty hard drives. n for Training samples from healthy hard drives. m It represents the margin between the most difficult positive sample and the most difficult negative sample.

[0030] Secondly, a hard disk fault prediction system based on multi-task learning is provided, including:

[0031] The sample acquisition module is used to acquire hard drive information from multiple hard drives and obtain a training sample set based on the hard drive information.

[0032] The model training module is used to train a dual-task hard disk failure prediction model based on a representation learning network and a metric learning network using a training sample set;

[0033] The sample processing module is used to form a sample input pair with the hard drive information of the target hard drive and the training samples in the training sample set when it is necessary to predict the fault of the target hard drive.

[0034] The fault prediction module is used to classify and regress the sample input into a dual-task hard disk fault prediction model to obtain the fault prediction result of the target hard disk.

[0035] Furthermore, the model training module includes:

[0036] Representation learning network unit, triplet sample search strategy unit, metric learning network unit, and model building unit;

[0037] The representation learning network unit is used to capture general hard disk information of all training samples in the training sample set using the representation learning network. The general hard disk information includes hard disk manufacturer information, hard disk model information, and hard disk fault information. The hard disk fault information is used as a label for binary classification to train a fault classification task model.

[0038] The triplet sample search strategy unit is used to learn triplets from all training samples based on the triplet sample search strategy and general information of the hard disk, and construct a training sample pool.

[0039] The metric learning network unit is used to perform metric learning on the training sample pool using a metric learning network and to optimize the triple loss to train a fault regression task model.

[0040] The model building unit is used to construct a dual-task hard disk fault prediction model based on the fault classification task model and the fault regression task model.

[0041] Furthermore,

[0042] The triplet sample search strategy unit is specifically used to randomly select P different hard drive models from the same hard drive manufacturer from all training samples in a preset training batch, based on the triplet sample search strategy and general hard drive information. For each hard drive model's triplet training sample, K positive samples and K negative samples are randomly selected as sample pairs of triplet training samples. Positive samples represent training samples with the same hard drive fault information as the triplet training samples, and negative samples represent training samples with different hard drive fault information than the triplet training samples. A training sample pool is constructed based on the triplet training samples and their corresponding sample pairs.

[0043] Furthermore,

[0044] The metric learning network unit is specifically used to perform metric learning on the training sample pool using a metric learning network, obtain the distance value between the triple training sample and each training sample in the sample pair, form triplet pairs based on the triple training samples and the distance values, and calculate the triplet loss; the fault regression task model is trained based on the metric learning network and the triplet loss.

[0045] Furthermore,

[0046] The metric learning network unit is also used to select the hardest positive sample and the hardest negative sample corresponding to the triple training sample based on the triple training sample and the distance value, forming a triple; the hardest positive sample is the positive sample that is farthest from the triple training sample, and the hardest negative sample is the negative sample that is closest to the triple training sample.

[0047] The triplet loss is calculated using the triplet loss function, which is expressed as follows:

[0048]

[0049] in, Indicates the loss of the triplet. This represents the training sample set of the faulty hard drive. This represents the training sample set for healthy hard drives, where the ternary training samples... a When it is a faulty hard drive, p for Training samples of faulty hard drives. n for Training samples from healthy hard drives. m It represents the margin between the most difficult positive sample and the most difficult negative sample.

[0050] The beneficial effects achieved by this invention are as follows:

[0051] The system acquires hard drive information from multiple hard drives and obtains a training sample set based on this information. A dual-task hard drive failure prediction model (classification and regression) is trained using this training sample set, based on a representation learning network and a metric learning network. When failure prediction is needed for a target hard drive, the target hard drive's information is used as the sample to be predicted, and this sample is paired with training samples from the training sample set. This pair is then input into the dual-task classification and regression hard drive failure prediction model to obtain the failure prediction result for the target hard drive. This approach combines representation learning and metric learning networks. The representation learning network learns the representational features of hard drives to better capture potential failure signs, while the metric learning network learns to measure the similarities and differences between hard drives from the same manufacturer and of the same model, thus more accurately predicting the risk of hard drive failure. Attached Figure Description

[0052] Figure 1 This is a flowchart of the hard disk fault prediction method based on multi-task learning according to the present invention;

[0053] Figure 2 This is a schematic diagram illustrating the training process of the dual-task hard disk fault prediction model of the present invention, which combines classification and regression.

[0054] Figure 3 This is a structural diagram of the hard disk fault prediction system based on multi-task learning according to the present invention;

[0055] Figure 4 This is a structural diagram of the model training module of the present invention. Detailed Implementation

[0056] The present invention will be further described below with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and should not be used to limit the scope of protection of the present invention.

[0057] like Figure 1 As shown, this embodiment of the invention provides a hard disk fault prediction method based on multi-task learning, including:

[0058] 101. Obtain hard drive information from multiple hard drives and use the hard drive information to obtain a training sample set;

[0059] In this embodiment, the hard drive information generally includes information describing the hard drive's identity, such as manufacturer and model. In this invention, the hard drive information also needs to include SMART data, which is highly correlated with hard drive failure. SMART data, short for Self-Monitoring Analysis and Reporting Technology, is a data tool used to record the operating status of hard disk drives (HDDs) or solid state disks (SSDs). Hard drive SMART data can provide the hard drive's health status, performance indicators, and potential failure warnings, helping users or system administrators take measures before hard drive problems occur.

[0060] To be compatible with hard drives from different manufacturers and models, during training, it is necessary to collect hard drive information from as many manufacturers and models as possible on the market, and use the hard drive information of each hard drive as a training sample to form a training sample set.

[0061] 102. Based on representation learning network and metric learning network, a dual-task hard disk failure prediction model of classification and regression is obtained by training a training sample set;

[0062] In this embodiment, the specific training process of the classification and regression dual-task hard disk fault prediction model is as follows: Figure 2 As shown, the steps include:

[0063] 201. A representation learning network is used to capture the common hard disk information of all training samples in the training sample set;

[0064] Representation learning is a key concept in deep learning, involving learning useful representations or features from raw data to better express the data's inherent structure and patterns. Representation learning networks are specifically Long Short-Term Memory (LSTM) networks. LSTM networks are a special type of Recurrent Neural Network (RNN) designed to solve the gradient vanishing or exploding problems that occur in ordinary RNNs when processing long sequences of data. By introducing memory units and gating mechanisms, LSTM networks can capture dependencies in long sequences, making them particularly suitable for tasks such as time series analysis, natural language processing, and speech recognition.

[0065] Therefore, a representation learning network is used to capture the general hard disk information of all training samples in the training sample set. The general hard disk information includes hard disk manufacturer information, hard disk model information and hard disk fault information.

[0066] 202. Hard drive failure information is used as a label for binary classification, and a failure classification task model is trained.

[0067] Hard drive fault information can determine whether the corresponding hard drive is faulty or healthy. Therefore, it is suitable for binary classification. Then, the fault classification task model is obtained by training the representation learning network in step 201.

[0068] 203. Based on the triplet sample search strategy and general disk information, triplet learning is performed on all training samples to construct a training sample pool.

[0069] Assume the preset batch size for training is b This indicates that the preset training batch contains b Based on the triplet sample search strategy and general hard drive information, P different hard drive models from the same hard drive manufacturer X are randomly selected from all training samples in the preset training batch.

[0070] For each hard drive model's ternary training samples, K positive samples and K negative samples are randomly selected as sample pairs for the ternary training samples. Positive samples represent training samples with the same hard drive fault information as the ternary training samples, and negative samples represent training samples with different hard drive fault information. Assuming the ternary training sample is a faulty hard drive, the training samples in the sample pair that are also faulty hard drives are marked as Y=1, representing positive samples; conversely, the training samples in the sample pair that are healthy hard drives are marked as Y=0, representing negative samples. Similarly, if the ternary training sample is a healthy hard drive, the training samples in the sample pair that are also healthy hard drives are marked as Y=1, representing positive samples; and the training samples in the sample pair that are faulty hard drives are marked as Y=0, representing negative samples.

[0071] The number of samples in each preset training batch is obtained by learning from triples. b* P * 2K samples are used to construct the training sample pool. It can be seen that by adopting the triplet sample search strategy, the number of samples in the original training batch is effectively increased, which can alleviate the problem of poor prediction performance when there are few samples on the hard disk in the existing technology.

[0072] 204. A metric learning network is used to perform metric learning on the training sample pool, and the triple loss is optimized to train a fault regression task model.

[0073] The metric learning network also uses an LSTM network. Unlike the representation learning network, the purpose of the metric learning network is to learn the similarity between two different hard disk time series. Instead of treating the hard disk failure prediction task as a classification task, it is regarded as a clustering or ranking task.

[0074] A metric learning network is used to perform metric learning on the training sample pool to obtain the distance value between the triple training sample and each training sample in the sample pair.

[0075] Based on the triple training samples and distance values, the most difficult positive sample and the most difficult negative sample corresponding to the triple training sample are selected to form a triple; the most difficult positive sample is the positive sample that is farthest from the triple training sample, and the most difficult negative sample is the negative sample that is closest to the triple training sample.

[0076] The triplet loss is calculated using the triplet loss function, which is expressed as follows:

[0077]

[0078] in, Indicates the loss of the triplet. This represents the training sample set of the faulty hard drive. This represents the training sample set for healthy hard drives, where the ternary training samples...a When it is a faulty hard drive, p for Training samples of faulty hard drives. n for Training samples from healthy hard drives. m It represents the margin between the most difficult positive sample and the most difficult negative sample;

[0079] The distance between two faulty hard drives of different models from the same manufacturer will be smaller than the distance between a faulty hard drive and a healthy hard drive of the same model, thus learning the fault modes of hard drives from the same manufacturer and distinguishing them from healthy hard drives;

[0080] The fault regression task model is obtained by training a metric learning network and triplet loss.

[0081] 205. Based on the fault classification task model and the fault regression task model, a dual-task hard disk fault prediction model of classification and regression is constructed.

[0082] The models formed in steps 202 and 204 above are combined to obtain a dual-task hard disk failure prediction model of classification and regression.

[0083] 103. When it is necessary to predict the failure of the target hard drive, the hard drive information of the target hard drive is used as the sample to be predicted, and the sample to be predicted is combined with the training samples in the training sample set to form a sample input pair.

[0084] 104. Input the sample into the dual-task hard disk failure prediction model of input classification and regression to obtain the failure prediction result of the target hard disk.

[0085] In this embodiment, the prediction process includes the following steps:

[0086] (1) Form a sample input pair for all training samples that are marked as faulty hard drives from the same manufacturer and model as the sample to be predicted; if any sample in the fault prediction result is predicted as similar, then mark the sample to be predicted as a faulty hard drive. Otherwise, continue to the next step (2);

[0087] (2) To improve decision-making efficiency, given the sparsity of the healthy disk sample status, 10% of healthy disks of the same manufacturer and model are randomly selected. Then, samples of these healthy disks are randomly collected at monthly intervals and paired with the sample to be predicted to form input sample pairs. If all these sample pairs are predicted to be similar, the sample to be predicted is marked as a healthy disk. Otherwise, proceed to the next step (3).

[0088] (3) Compare multiple fault samples from different hard drive models of the same manufacturer with the sample to be predicted; if more than half of the prediction results are similar, then mark the sample to be predicted as a faulty hard drive.

[0089] (4) Compare multiple fault samples from the same manufacturer and the same hard drive model with the sample to be predicted; if there is no hard drive model to which the sample to be predicted belongs in the sample pool, then directly execute (3).

[0090] The beneficial effects of the embodiments of the present invention are as follows:

[0091] The system acquires hard drive information from multiple hard drives and obtains a training sample set based on this information. A dual-task hard drive failure prediction model (classification and regression) is trained using this training sample set, based on a representation learning network and a metric learning network. When failure prediction is needed for a target hard drive, the target hard drive's information is used as the sample to be predicted, and this sample is paired with training samples from the training sample set. This pair is then input into the dual-task classification and regression hard drive failure prediction model to obtain the failure prediction result for the target hard drive. This approach combines representation learning and metric learning networks. The representation learning network learns the representational features of hard drives to better capture potential failure signs, while the metric learning network learns to measure the similarities and differences between hard drives from the same manufacturer and of the same model, thus more accurately predicting the risk of hard drive failure.

[0092] By employing a triplet sample search strategy, the number of samples in the original training batch is effectively increased, which can alleviate the problem of poor prediction performance when using disk datasets with fewer samples in existing technologies.

[0093] Based on the hard disk fault prediction method based on multi-task learning described in the above embodiments, the following embodiments illustrate the hard disk fault prediction system based on multi-task learning.

[0094] like Figure 3 As shown, this embodiment of the invention provides a hard disk fault prediction system based on multi-task learning, comprising:

[0095] The sample acquisition module 301 is used to acquire hard drive information from multiple hard drives and obtain a training sample set based on the hard drive information.

[0096] The model training module 302 is used to train a dual-task hard disk fault prediction model based on a representation learning network and a metric learning network using a training sample set;

[0097] The sample processing module 303 is used to form a sample input pair with the hard disk information of the target hard disk and the training samples in the training sample set when it is necessary to perform fault prediction on the target hard disk.

[0098] The fault prediction module 304 is used to classify and regress the sample input into a dual-task hard disk fault prediction model to obtain the fault prediction result of the target hard disk.

[0099] Combination Figure 3 The embodiments shown are preferred, such as Figure 4 As shown, in some embodiments of the present invention, the model training module 302 includes:

[0100] Representation learning network unit 401, triplet sample search strategy unit 402, metric learning network unit 403, and model building unit 404;

[0101] The representation learning network unit 401 is used to capture the general information of the hard disk of all training samples in the training sample set using the representation learning network. The general information of the hard disk includes the hard disk manufacturer information, hard disk model information and hard disk fault information. The hard disk fault information is used as a label for binary classification to train a fault classification task model.

[0102] The triplet sample search strategy unit 402 is used to perform triplet learning on all training samples based on the triplet sample search strategy and general information of the hard disk to construct a training sample pool.

[0103] The metric learning network unit 403 is used to perform metric learning on the training sample pool using a metric learning network and optimize the triple loss to train a fault regression task model.

[0104] Model building unit 404 is used to construct a dual-task hard disk fault prediction model based on the fault classification task model and the fault regression task model.

[0105] Combination Figure 4 The illustrated embodiments, and preferably, some embodiments of the present invention,

[0106] The triplet sample search strategy unit 402 is specifically used to randomly select P different hard drive models from the same hard drive manufacturer from all training samples in a preset training batch based on the triplet sample search strategy and general hard drive information; for each hard drive model's triplet training sample, randomly select K positive samples and K negative samples as sample pairs of triplet training samples; positive samples represent training samples with the same hard drive fault information as the triplet training samples, and negative samples represent training samples with different hard drive fault information than the triplet training samples; and a training sample pool is constructed based on the triplet training samples and their corresponding sample pairs.

[0107] Combination Figure 4 The illustrated embodiments, and preferably, some embodiments of the present invention,

[0108] The metric learning network unit 403 is specifically used to perform metric learning on the training sample pool using a metric learning network, to obtain the distance value between the triple training sample and each training sample in the sample pair, to form a triplet based on the triple training sample and the distance value, and to calculate the triplet loss; and to train the fault regression task model based on the metric learning network and the triplet loss.

[0109] The metric learning network unit 403 is also used to select the most difficult positive sample and the most difficult negative sample corresponding to the triple training sample based on the triple training sample and the distance value, forming a triple; the most difficult positive sample is the positive sample that is farthest from the triple training sample, and the most difficult negative sample is the negative sample that is closest to the triple training sample.

[0110] The triplet loss is calculated using the triplet loss function, which is expressed as follows:

[0111]

[0112] in, Indicates the loss of the triplet. This represents the training sample set of the faulty hard drive. This represents the training sample set for healthy hard drives, where the ternary training samples... a When it is a faulty hard drive, p for Training samples of faulty hard drives. n for Training samples from healthy hard drives. m It represents the margin between the most difficult positive sample and the most difficult negative sample.

[0113] In summary, the beneficial effects of the hard disk fault prediction system based on multi-task learning in this invention are as follows:

[0114] It combines representation learning networks and metric learning networks. The representation learning network learns the representational features of hard drives to better capture potential failure signs, while the metric learning network learns the similarities and differences between hard drives from the same manufacturer and of the same model to more accurately predict the risk of hard drive failure.

[0115] By employing a triplet sample search strategy, the number of samples in the original training batch is effectively increased, which can alleviate the problem of poor prediction performance when using disk datasets with fewer samples in existing technologies.

[0116] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0117] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0118] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0119] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0120] The above are merely embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of the claims of the present invention pending approval.

Claims

1. A hard disk fault prediction method based on multi-task learning, characterized in that, include: Obtain hard drive information from multiple hard drives, and obtain a training sample set based on the hard drive information; Based on representation learning network and metric learning network, a dual-task hard disk fault prediction model of classification and regression is obtained by training the training sample set. When it is necessary to predict the failure of the target hard drive, the hard drive information of the target hard drive is used as the sample to be predicted, and the sample to be predicted is combined with the training samples in the training sample set to form a sample input pair. The sample input is fed into the classification and regression dual-task hard disk failure prediction model to obtain the failure prediction result of the target hard disk; The dual-task hard disk failure prediction model, based on representation learning network and metric learning network, is trained using the training sample set and includes: A representation learning network is used to capture the general hard disk information of all training samples in the training sample set. The general hard disk information includes hard disk manufacturer information, hard disk model information, and hard disk fault information. The hard drive fault information is used as a label for binary classification, and a fault classification task model is trained. Based on the triplet sample search strategy and the general information of the hard disk, triplet learning is performed on all training samples to construct a training sample pool. A metric learning network is used to perform metric learning on the training sample pool, and the triplet loss is optimized to train a fault regression task model. Based on the fault classification task model and the fault regression task model, a dual-task hard disk fault prediction model of classification and regression is constructed.

2. The hard disk fault prediction method based on multi-task learning according to claim 1, characterized in that, The triplet-based sample search strategy and the general information of the hard disk are used to perform triplet learning on all training samples to construct a training sample pool, including: Based on the triplet sample search strategy and the general information of the hard drive, P triplet training samples of different hard drive models from the same hard drive manufacturer are randomly selected from all training samples in the preset training batch. For each hard drive model's ternary training samples, K positive samples and K negative samples are randomly selected as sample pairs of the ternary training samples; the positive samples represent training samples with the same hard drive fault information as the ternary training samples, and the negative samples represent training samples with different hard drive fault information than the ternary training samples. A training sample pool is constructed based on the ternary training samples and their corresponding sample pairs.

3. The hard disk fault prediction method based on multi-task learning according to claim 2, characterized in that, The step of using a metric learning network to perform metric learning on the training sample pool and optimizing the triplet loss to train a fault regression task model includes: A metric learning network is used to perform metric learning on the training sample pool to obtain the distance value between the triple training sample and each training sample in the sample pair. Based on the triple training samples and the distance values, triples are formed, and the triple loss is calculated. The fault regression task model is obtained by training the metric learning network and the triplet loss.

4. The hard disk fault prediction method based on multi-task learning according to claim 3, characterized in that, The step of forming triplets based on the triplet training samples and the distance values, and calculating the triplet loss, includes: Based on the triple training samples and the distance values, the most difficult positive sample and the most difficult negative sample corresponding to the triple training samples are selected to form a triple; the most difficult positive sample is the positive sample that is farthest from the triple training sample, and the most difficult negative sample is the negative sample that is closest to the triple training sample. The triplet loss is calculated using the triplet loss function, the expression of which is: ； Among them, the Represents the triplet loss, the stated The training sample set representing the faulty hard drive, the The training sample set represents healthy hard drives. When the ternary training sample 'a' is a faulty hard drive, 'p' is the training sample set of healthy hard drives. The training samples of the faulty hard drive, where n is the... The training samples of healthy hard drives, where m is the margin between the most difficult positive sample and the most difficult negative sample, P represents the number of hard drive models, and K represents the number of sample pairs.

5. A hard disk fault prediction system based on multi-task learning, characterized in that, include: The sample acquisition module is used to acquire hard drive information from multiple hard drives and obtain a training sample set based on the hard drive information. The model training module is used to train a dual-task hard disk fault prediction model based on the training sample set, using a representation learning network and a metric learning network. The sample processing module is used to, when it is necessary to predict the failure of the target hard disk, use the hard disk information of the target hard disk as the sample to be predicted, and form a sample input pair with the training samples in the training sample set. The fault prediction module is used to input the sample input pair into the classification and regression dual-task hard disk fault prediction model to obtain the fault prediction result of the target hard disk; The model training module includes: Representation learning network unit, triplet sample search strategy unit, metric learning network unit, and model building unit; The representation learning network unit is used to capture general hard disk information of all training samples in the training sample set using a representation learning network. The general hard disk information includes hard disk manufacturer information, hard disk model information, and hard disk fault information. The hard disk fault information is used as a label for binary classification to train a fault classification task model. The triplet sample search strategy unit is used to perform triplet learning on all training samples based on the triplet sample search strategy and the general information of the hard disk, and construct a training sample pool. The metric learning network unit is used to perform metric learning on the training sample pool using a metric learning network and to optimize the triple loss to train a fault regression task model. The model building unit is used to construct a dual-task hard disk fault prediction model based on the fault classification task model and the fault regression task model.

6. The hard disk fault prediction system based on multi-task learning according to claim 5, characterized in that, The triplet sample search strategy unit is specifically used to randomly select P different hard drive models from the same hard drive manufacturer from all training samples in a preset training batch, based on the triplet sample search strategy and the general information of the hard drive. For each hard drive model's ternary training samples, K positive samples and K negative samples are randomly selected as sample pairs of the ternary training samples; the positive samples represent training samples with the same hard drive fault information as the ternary training samples, and the negative samples represent training samples with different hard drive fault information than the ternary training samples; a training sample pool is constructed based on the ternary training samples and their corresponding sample pairs.

7. The hard disk fault prediction system based on multi-task learning according to claim 6, characterized in that, The metric learning network unit is specifically used to perform metric learning on the training sample pool using a metric learning network, to obtain the distance value between the triple training sample and each training sample in the sample pair, to form triplet groups based on the triple training samples and the distance values, to calculate the triplet loss, and to train a fault regression task model based on the metric learning network and the triplet loss.

8. The hard disk fault prediction system based on multi-task learning according to claim 7, characterized in that, The metric learning network unit is further configured to select the most difficult positive sample and the most difficult negative sample corresponding to the triple training sample based on the triple training sample and the distance value, forming a triple; the most difficult positive sample is the positive sample that is farthest from the triple training sample, and the most difficult negative sample is the negative sample that is closest to the triple training sample. The triplet loss is calculated using the triplet loss function, the expression of which is: ； Among them, the Represents the triplet loss, the stated The training sample set representing the faulty hard drive, the The training sample set represents healthy hard drives. When the ternary training sample 'a' is a faulty hard drive, 'p' is the training sample set of healthy hard drives. The training samples of the faulty hard drive, where n is the... The training samples of healthy hard drives, where m is the margin between the most difficult positive sample and the most difficult negative sample, P represents the number of hard drive models, and K represents the number of sample pairs.

Images