Data label correction method, device, equipment, storage medium and program product
By determining the feature distance and health labels of stored data pairs in faulty hard drives, formulating label correction strategies, correcting the health labels of stored data, and retraining the model, the problem of insufficient accuracy of fault prediction models is solved, and timely and accurate hard drive fault detection is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- INDUSTRIAL AND COMMERCIAL BANK OF CHINA
- Filing Date
- 2023-03-22
- Publication Date
- 2026-07-21
AI Technical Summary
Existing fault prediction models lack accuracy in hard drive fault detection, resulting in untimely or inaccurate hard drive fault alarms.
By determining the feature distance and health labels between data pairs stored in the faulty hard drive, a label correction strategy is formulated to correct the health labels of the stored data, and the corrected data is used for retraining the fault prediction model.
It improves the accuracy of the fault prediction model, ensures the timeliness and precision of hard drive fault detection, and updates the model accuracy in real time.
Smart Images
Figure CN116225808B_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device, storage medium, and program product for correcting data tags. Background Technology
[0002] With the development of computer technology in various fields, storage systems have been applied in various sectors. Storage systems, such as hard drives, are the most important component of data centers in various fields.
[0003] Currently, to ensure normal data storage, during the process of using hard drives for data storage, it is necessary to use a fault prediction model to predict faults in the stored data in real time and label each piece of stored data with a health status tag. For example, data with abnormal detection results is labeled as faulty, and data with normal detection results is labeled as healthy. When a faulty health status tag is detected in the stored data on the hard drive, a hard drive fault alarm message will be output to remind maintenance personnel to replace the hard drive.
[0004] Since the timeliness and accuracy of hard drive fault alarms depend entirely on the accuracy of the training results of the fault prediction model, improving the accuracy of the fault prediction model is an urgent problem to be solved in the field of hard drive fault detection. Summary of the Invention
[0005] Therefore, it is necessary to provide a method, apparatus, device, storage medium, and program product for correcting data tags that can improve the accuracy of fault prediction models, in order to address the above-mentioned technical problems.
[0006] Firstly, this application provides a method for correcting data labels. The method includes:
[0007] If the fault prediction model detects a faulty hard drive, it determines the first feature distance between the two stored data in each data pair in the faulty hard drive, and the second feature distance between each stored data in each data pair and the reference data of each data pair; wherein, any two stored data in the faulty hard drive constitute a data pair; the reference data of each data pair is any other stored data in the faulty hard drive other than the data pair.
[0008] Based on the first feature distance and the second feature distance of each data pair, as well as the health label of each stored data, determine the label correction strategy for each stored data.
[0009] Based on the label correction strategy for each stored data, the health labels of each stored data are corrected; the stored data with corrected health labels are then used to retrain the fault prediction model.
[0010] In one embodiment, determining a first characteristic distance between two stored data items contained in each data pair in the faulty hard disk, and a second characteristic distance between each stored data item in each data pair and a reference data item of each data pair, includes:
[0011] Map the stored data in the faulty hard drive to the same space to obtain the spatial characteristics of the stored data in the same space;
[0012] Based on the spatial characteristics of the two stored data contained in each data pair, determine the first feature distance of each data pair;
[0013] Based on the spatial characteristics of each stored data in each data pair and the spatial characteristics of the reference data of each data pair, determine the second feature distance between each stored data in each data pair and the reference data of each data pair.
[0014] In one embodiment, a label correction strategy for each stored data is determined based on a first feature distance and a second feature distance between each data pair, and the health label of each stored data, including:
[0015] Based on the first feature distance and the second feature distance of each data pair, as well as the consistency of the health labels of the two stored data contained in each data pair, the adjacency relationship between the two stored data contained in each data pair is determined.
[0016] Based on the adjacency relationship between the two stored data contained in each data pair, and the health label of each stored data, determine the label correction strategy for each stored data.
[0017] In one embodiment, the second feature distance includes a first sub-distance and a second sub-distance; based on the first feature distance and the second feature distance of each data pair, and the consistency of the health labels of the two stored data contained in each data pair, the adjacency relationship between the two stored data contained in each data pair is determined, including:
[0018] For each data pair, if the sum of the first sub-distance and the second sub-distance of the data pair is greater than the first feature distance, and the health labels of the two stored data contained in the data pair are consistent, then the adjacency relationship between the two stored data contained in the data pair is determined to be adjacency.
[0019] In one embodiment, a label correction strategy for each stored data is determined based on the adjacency relationship between the two stored data contained in each data pair and the health label of each stored data, including:
[0020] Based on the adjacency relationship between the two stored data contained in each data pair, determine the adjacent data corresponding to each stored data;
[0021] Based on the health label of each stored data and the health labels of its adjacent data, a label correction strategy is determined for each stored data.
[0022] In one embodiment, a label correction strategy for each stored data is determined based on the health label of each stored data and the health labels of its adjacent data, including:
[0023] Based on the health label of each stored data and the health labels of its adjacent data, the label accuracy category of each stored data is determined; the label accuracy categories include: correct, incorrect, and pending.
[0024] A label correction strategy is determined for each stored data based on the label accuracy category of each stored data.
[0025] In one embodiment, the label accuracy category of each stored data is determined based on the health label of each stored data and the health labels of its adjacent data, including:
[0026] Based on the health label of each stored data and the health label of the adjacent data corresponding to each stored data, determine the proportion of adjacent data with the same label for each stored data.
[0027] The label accuracy category of each stored data is determined based on the proportion of adjacent data with the same label.
[0028] In one embodiment, a label correction strategy is determined for each stored data based on the label accuracy category of each stored data, including:
[0029] Based on the label accuracy category of each stored data, determine whether the proportion of stored data with incorrect label accuracy category is less than a preset proportion threshold;
[0030] If so, a label correction strategy is determined for each stored data based on the label accuracy category of each stored data.
[0031] Secondly, this application also provides a data tag correction device. The device includes:
[0032] The distance determination module is used to determine, if the fault prediction model detects a faulty hard disk, the first feature distance between the two stored data in each data pair in the faulty hard disk, and the second feature distance between each stored data in each data pair and the reference data of each data pair; wherein, any two stored data in the faulty hard disk constitute a data pair; the reference data of each data pair is any other stored data in the faulty hard disk other than the data pair.
[0033] The strategy determination module is used to determine the label correction strategy for each stored data based on the first feature distance and the second feature distance of each data pair, as well as the health label of each stored data.
[0034] The label correction module is used to correct the health labels of each stored data according to the label correction strategy of each stored data; the stored data with corrected health labels are used to retrain the fault prediction model.
[0035] Thirdly, this application also provides a computer device. The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to perform the following steps:
[0036] If the fault prediction model detects a faulty hard drive, it determines the first feature distance between the two stored data in each data pair in the faulty hard drive, and the second feature distance between each stored data in each data pair and the reference data of each data pair; wherein, any two stored data in the faulty hard drive constitute a data pair; the reference data of each data pair is any other stored data in the faulty hard drive other than the data pair.
[0037] Based on the first feature distance and the second feature distance of each data pair, as well as the health label of each stored data, determine the label correction strategy for each stored data.
[0038] Based on the label correction strategy for each stored data, the health labels of each stored data are corrected; the stored data with corrected health labels are then used to retrain the fault prediction model.
[0039] Fourthly, this application also provides a computer-readable storage medium. This computer-readable storage medium stores a computer program thereon, which, when executed by a processor, performs the following steps:
[0040] If the fault prediction model detects a faulty hard drive, it determines the first feature distance between the two stored data in each data pair in the faulty hard drive, and the second feature distance between each stored data in each data pair and the reference data of each data pair; wherein, any two stored data in the faulty hard drive constitute a data pair; the reference data of each data pair is any other stored data in the faulty hard drive other than the data pair.
[0041] Based on the first feature distance and the second feature distance of each data pair, as well as the health label of each stored data, determine the label correction strategy for each stored data.
[0042] Based on the label correction strategy for each stored data, the health labels of each stored data are corrected; the stored data with corrected health labels are then used to retrain the fault prediction model.
[0043] Fifthly, this application also provides a computer program product. This computer program product includes a computer program that, when executed by a processor, performs the following steps:
[0044] If the fault prediction model detects a faulty hard drive, it determines the first feature distance between the two stored data in each data pair in the faulty hard drive, and the second feature distance between each stored data in each data pair and the reference data of each data pair; wherein, any two stored data in the faulty hard drive constitute a data pair; the reference data of each data pair is any other stored data in the faulty hard drive other than the data pair.
[0045] Based on the first feature distance and the second feature distance of each data pair, as well as the health label of each stored data, determine the label correction strategy for each stored data.
[0046] Based on the label correction strategy for each stored data, the health labels of each stored data are corrected; the stored data with corrected health labels are then used to retrain the fault prediction model.
[0047] The aforementioned data label correction method, apparatus, device, storage medium, and program product, for a faulty hard drive output by a fault prediction model, acquires all stored data in the faulty hard drive and groups all the stored data into multiple data pairs, with each data pair corresponding to a reference data. Then, it determines a first feature distance between the two stored data in each data pair, and a second feature distance between each stored data in each data pair and its corresponding reference data. Furthermore, in determining the label correction strategy for each stored data, it considers not only the relationship between the first and second feature distances but also the health label of each stored data, making the determined label correction strategy for each stored data more closely matched to the stored data, thus providing a guarantee for subsequent correction of the health labels of the stored data based on the label correction strategy. Furthermore, the correction results after correcting the health labels of each stored data based on the label correction strategy will be correspondingly more accurate. In addition, the stored data with corrected health labels will be used to retrain the fault prediction model. This solution can simultaneously correct the health labels of the storage data of the predicted faulty hard drives in real time while the fault prediction model is detecting hard drive faults. Based on the correct health labels of the storage data, the fault prediction model can be retrained in real time, thereby continuously updating the accuracy of the fault prediction model while it is in use, and thus improving the accuracy of the fault prediction model in predicting faulty hard drives. Attached Figure Description
[0048] Figure 1This is an application environment diagram of a data tag correction method provided in this embodiment;
[0049] Figure 2 A flowchart illustrating the first data tag correction method provided in this embodiment;
[0050] Figure 3 This is a flowchart illustrating a method for determining the label correction strategy for each stored data in this embodiment.
[0051] Figure 4 This is a flowchart illustrating the second data tag correction method provided in this embodiment;
[0052] Figure 5 This is a structural block diagram of the first type of data tag correction device provided in this embodiment;
[0053] Figure 6 This is a structural block diagram of the second type of data tag correction device provided in this embodiment;
[0054] Figure 7 This is a structural block diagram of the third type of data tag correction device provided in this embodiment;
[0055] Figure 8 This is an internal structural diagram of a computer device provided in this embodiment. Detailed Implementation
[0056] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.
[0057] The data tag correction method provided in this application embodiment can be applied to, for example, Figure 1 In the application environment shown, in one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows. Figure 1 As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores data related to data tag correction. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a data tag correction method.
[0058] In one embodiment, such as Figure 2 As shown, a method for correcting data labels is provided, which can be applied to... Figure 1 Taking a computer as an example, the explanation includes the following steps:
[0059] S201, if the fault prediction model detects a faulty hard disk, then determine the first feature distance between the two stored data in each data pair in the faulty hard disk, and the second feature distance between each stored data in each data pair and the reference data of each data pair.
[0060] In this embodiment, any two stored data items in the faulty hard drive constitute a data pair; the reference data for each data pair is any other stored data item in the faulty hard drive besides the data pair itself; the fault prediction model is a pre-trained model used to detect faults in the stored data on the hard drive. The first feature distance characterizes the distance between the two stored data items contained in a data pair. The second feature distance characterizes the distance between each stored data item in each data pair and the corresponding reference data. It is understood that since the faulty hard drive in this embodiment is the faulty hard drive output by the fault prediction model, the faulty hard drive in this embodiment must contain stored data with an abnormal health label. Furthermore, the health labels of the stored data in the faulty hard drive are predicted and labeled by the fault prediction model.
[0061] Specifically, in this embodiment, for the faulty hard drive output by the fault prediction model, all the stored data in the faulty hard drive are first obtained, then each data pair is determined from all the stored data, and the reference data corresponding to each data pair is determined. Based on each data pair and its corresponding reference data, the first feature distance between the two stored data contained in each data pair and the second feature distance between each stored data in each data pair and the reference data of each data pair are determined.
[0062] Optionally, in this embodiment, the process of determining each data pair from all stored data in the faulty hard disk can be as follows: randomly select any two stored data points as a data pair each time, until all two stored data points in the faulty hard disk have formed a data pair. Furthermore, for each data pair, arbitrarily select one of the other stored data points in the faulty hard disk (excluding that data pair) as the reference data corresponding to that data pair.
[0063] Optionally, one possible method for determining the first feature distance between the two stored data items in each data pair in the faulty hard drive, and the second feature distance between each stored data item in each data pair and the reference data of each data pair, can be as follows: Quantize each stored data item according to a preset quantization rule to obtain word vectors for each stored data item; further, calculate the first Euclidean distance between the two word vectors based on the word vectors of the two stored data items in each data pair, and use the first Euclidean distance as the first feature distance. Calculate the second Euclidean distance between the first stored data item and the reference data based on the word vectors of the reference data of each data pair, and the word vectors of the two stored data items in the data pair (such as the first stored data item and the second stored data item); calculate the third Euclidean distance between the second stored data item and the reference data; and use the second region distance and the third Euclidean distance as the second feature distance.
[0064] Alternatively, another possible approach to determining the first and second feature distances is to map each stored data in the faulty hard drive to the same space, obtaining the spatial characteristics of each stored data in the same space; based on the spatial characteristics of the two stored data contained in each data pair, determine the first feature distance of each data pair; and based on the spatial characteristics of each stored data in each data pair and the spatial characteristics of the reference data of each data pair, determine the second feature distance between each stored data in each data pair and the reference data of each data pair. Specifically, this can be done by mapping each stored data in the faulty hard drive to a unified real number space or by mapping each stored data in the faulty hard drive to a matrix space; there is no limitation on this. If each stored data in the faulty hard drive is mapped to a unified real number space, then the spatial characteristic of each stored data in that space is the coordinate point of that stored data in that real number space; if each stored data in the faulty hard drive is mapped to a matrix space, then the spatial characteristic of each stored data in that space is the vector of that stored data in that matrix space.
[0065] For each data pair, after obtaining the spatial characteristics of each stored data (such as the first and second stored data) and the spatial characteristics of the reference data in the data pair using the above method, the distance between the first and second stored data can be calculated based on the spatial characteristics of the first and second stored data. This distance is then used as the first characteristic distance of the data pair. Further, based on the spatial characteristics of the first and second stored data and the reference data, a first distance is calculated between the first and reference data, and a second distance is calculated between the second and reference data. These two distances are then used as the second characteristic distance of the data pair. In the above embodiment, before determining the first and second characteristic distances, each stored data in the faulty hard drive is mapped to the same space. The first and second characteristic distances are determined based on the spatial characteristics of each stored data in the same space. This allows the determined first and second characteristic distances to more accurately characterize the adjacency relationships between stored data, providing a basis for subsequently determining the label correction strategy for each stored data based on the first and second characteristic distances of each data pair and the health labels of each stored data.
[0066] S202, based on the first feature distance and the second feature distance of each data pair, and the health label of each stored data, determine the label correction strategy for each stored data.
[0067] The health label for each piece of stored data is the fault prediction result assigned to the stored data by the fault prediction model after predicting faults. For example, if the fault prediction result is healthy, the health label for the stored data is healthy; if the fault prediction result is fault, the health label for the stored data is fault.
[0068] Optionally, in this embodiment, the first feature distance and the second feature distance of each data pair can be compared. Based on the comparison result and the health label of each stored data, a label correction strategy for each stored data is determined. For example, for each data pair, if the first feature distance of the data pair is less than the second feature distance (it should be noted that if the second feature distance has two values, then the first feature distance must be less than the sum of the two second feature distances), then the two stored data contained in the data pair are connected. Connecting the two stored data in all data pairs that meet the above conditions yields an undirected graph corresponding to the stored data. Based on the health label of each stored data in the undirected graph and the stored data connected to it, it is determined whether the health label of the stored data is correct, and thus the label correction strategy for the stored data is determined. For example, if the health label of the stored data is correct, the label correction strategy for the stored data is to retain the health label of the stored data; if the health label of the stored data is incorrect, the label correction strategy for the stored data is to reverse-process the health label of the stored data.
[0069] In one embodiment, determining the label correction strategy for each stored data based on the first and second feature distances of each data pair and the health labels of each stored data can be achieved by: determining the adjacency relationship between the two stored data items in each data pair based on the first and second feature distances of each data pair and the consistency of the health labels of the two stored data items contained in each data pair; and determining the label correction strategy for each stored data item based on the adjacency relationship between the two stored data items in each data pair and the health labels of each stored data item. The second feature distance includes a first sub-distance and a second sub-distance. It can be understood that the first sub-distance and the second sub-distance represent the distance between one stored data item in the data pair and its corresponding reference data. In this embodiment, for each data pair, if the sum of the first sub-distance and the second sub-distance of the data pair is greater than the first feature distance, and the health labels of the two stored data items contained in the data pair are consistent, then the adjacency relationship between the two stored data items contained in the data pair is determined to be adjacency. For example, based on the first feature distance and the second feature distance of each data pair, it can be determined whether the two stored data in each data pair might have an adjacency relationship. If so, it is then determined whether the health labels of the two stored data in the data pair are consistent. If so, it is determined that the two stored data in the data pair have an adjacency relationship; otherwise, the two stored data in the data pair do not have an adjacency relationship. Using the relationship between the first feature distance and the second feature distance, as well as whether the health labels of the two stored data in the data pair are consistent, to comprehensively determine whether the two stored data in the data pair have an adjacency relationship makes the adjacency determination process more rigorous. Optionally, the way to determine whether the two stored data in each data pair might have an adjacency relationship is that if the first feature distance of the data pair is less than the second feature distance (it should be noted that if the second feature distance has two values, then the first feature distance needs to be less than the sum of the two second feature distances), then it indicates that the two stored data in the data pair have an adjacency relationship. When a data pair contains two adjacent data items, the label correction strategy for each data item is determined based on the adjacency relationship between the two adjacent data items and the health label of each data item. For example, it is determined whether the consistency of the health label of each data item with its adjacent data meets a preset rule. If so, the health label of the data item is determined to be correct, and the label correction strategy for the data item is then determined. If not, the health label of the data item is determined to be incorrect, and the label correction strategy for the data item is then determined. For example, if the health label of the data item is correct, the label correction strategy for the data item is to retain the health label of the data item; if the health label of the data item is incorrect, the label correction strategy for the data item is to reverse the processing of the health label of the data item.In the above embodiments, based on the adjacency relationship between the two stored data contained in each data pair and the health label of each stored data, a label correction strategy for each stored data is determined, making the label correction strategy for each stored data more accurate and providing a guarantee for the subsequent correction of the health label of the stored data.
[0070] S203, Based on the label correction strategy for each stored data, correct the health labels of each stored data; the corrected health labels of each stored data are used to retrain the fault prediction model.
[0071] Specifically, in this embodiment, the health labels of each stored data are corrected according to the label correction strategy corresponding to each stored data. For example, if the label correction strategy is to retain the health label of the stored data, then the health label of the stored data is not processed, and the stored data can be directly used to retrain the fault prediction model; if the label correction strategy is to reverse the processing of the health label of the stored data, then for the case where the health label of the stored data is healthy, it is corrected to fault; for the case where the health label of the stored data is healthy, it is corrected to healthy.
[0072] After correcting the health labels of each storage data point using a label correction strategy, the fault prediction model is retrained using the corrected health labels. It is understood that the health labels corresponding to the corrected health labels are accurate. Training the fault prediction model based on storage data with accurate health labels can improve the output accuracy of the fault prediction model, that is, it can improve the accuracy of fault prediction.
[0073] In the aforementioned data label correction method, for the faulty hard drive output by the fault prediction model, all storage data in the faulty hard drive is obtained, and all storage data are grouped into multiple data pairs, with each data pair corresponding to a reference data. Then, the first feature distance between the two storage data in each data pair and the second feature distance between each storage data in each data pair and its corresponding reference data are determined. Furthermore, in determining the label correction strategy for each storage data, the relationship between the first and second feature distances is not only considered, but also the health label of each storage data. This ensures that the determined label correction strategy for each storage data is more closely matched to the storage data, providing a guarantee for subsequent correction of the health labels of the storage data based on the label correction strategy. Moreover, the correction results after correcting the health labels of each storage data based on the label correction strategy will be correspondingly more accurate. In addition, the storage data with corrected health labels will be used to retrain the fault prediction model. This solution can simultaneously correct the health labels of the storage data of the predicted faulty hard drives in real time while the fault prediction model is detecting hard drive failures. Based on the correct health labels of the storage data, the fault prediction model can be retrained in real time, thereby continuously updating the accuracy of the fault prediction model while it is in use, and thus improving the accuracy of the fault prediction model in predicting faulty hard drives.
[0074] Furthermore, in the process of determining the correction strategy for the health labels of each stored data, to make the process of determining the label correction strategy more rigorous, in one embodiment, such as... Figure 3 As shown, based on the adjacency relationship between the two stored data items contained in each data pair, and the health label of each stored data item, the label correction strategy for each stored data item is determined, including:
[0075] S301, determine the adjacent data corresponding to each storage data according to the adjacency relationship between the two stored data contained in each data pair.
[0076] If a data pair contains two adjacent data items, then the two adjacent data items in the data pair are considered to be adjacent to each other.
[0077] Specifically, in this embodiment, for each data pair, if the two stored data contained in the data pair are adjacent, then the two stored data contained in the data pair are regarded as a group of adjacent data. In this way, it can be determined whether the two stored data in each data pair are adjacent data.
[0078] S302, determine the label correction strategy for each stored data based on the health label of each stored data and the health labels of the adjacent data corresponding to each stored data.
[0079] Optionally, the method for determining the label correction strategy for each stored data based on its health label and the health labels of its adjacent data can be as follows: First, determine the label accuracy category for each stored data based on its health label and the health labels of its adjacent data. The label accuracy category includes: correct, incorrect, and pending. Then, determine the label correction strategy for each stored data based on its label accuracy category. For example, this can be achieved by first obtaining the health labels of each stored data, and then determining the percentage of adjacent data with the same label for each stored data based on its health label and the health labels of its adjacent data. Finally, determine the label accuracy category for each stored data based on this percentage. The percentage of adjacent data with the same label can be the percentage of data with the same health label as the stored data among all its adjacent data.
[0080] Specifically, in this embodiment, for each piece of stored data, the proportion of adjacent data with the same label is first determined. Then, a threshold for the adjacent label is compared with the proportion of adjacent data with the same label, based on a pre-determined range. If the proportion of adjacent data with the same label is greater than the maximum value of the range, the label accuracy classification result of the stored data is determined to be correct; if the proportion is less than the minimum value, the label accuracy classification result is determined to be incorrect; if the proportion is within the range, the label accuracy classification result is determined to be pending. For example, taking an adjacent label threshold of 30%-70%, if the proportion is greater than 70%, the label accuracy category of the stored data is correct; if it is less than 30%, the label accuracy category is incorrect; and if it is between 30% and 70%, the label accuracy category is pending. For example, if the health label of a certain stored data is "healthy," and 90% of the adjacent data corresponding to that stored data have the same health label, then the health label accuracy classification result of that stored data is determined to be correct. Determining the label accuracy category of each stored data based on the proportion of adjacent data with the same label simplifies the process. In the above embodiment, the label accuracy categories of stored data are divided into three categories instead of just "correct" and "incorrect." Corresponding correction processes are performed on all three types of stored data—correct, incorrect, and pending—making the correction process more rigorous and resulting in more accurate retained data after correction.
[0081] Furthermore, after determining the label accuracy of each stored data item and before determining the label correction strategy for each stored data item, to ensure the correctness of the corrected stored data, in one embodiment, it is also determined whether the proportion of stored data with incorrect label accuracy category is less than a preset proportion threshold based on the label accuracy category of each stored data item; if so, a label correction strategy is determined for each stored data item based on its label accuracy category. For example, in this embodiment, if the proportion of stored data with incorrect label accuracy category is greater than the predicted proportion threshold, it is considered that there is too much erroneous data in the faulty hard drive. To ensure the label accuracy of the corrected stored data, all data in the faulty hard drive will be deleted. That is, the fault prediction model will no longer be trained based on the stored data in the faulty hard drive. Only when the proportion of stored data with incorrect label accuracy category in the faulty hard drive is less than the predicted proportion threshold is the operation of determining the label correction strategy for each stored data item based on its health label and the health labels of its adjacent data executed. This provides a guarantee for improving the label accuracy of the corrected stored data, thereby improving the accuracy of the fault prediction model.
[0082] Furthermore, after determining the label accuracy category of each stored data, the label correction strategy determined for each category of stored data will correct the health label of each stored data. For example, in this embodiment, stored data with a correct label accuracy category may not require label correction; for stored data with an incorrect label accuracy category, the label of the stored data will be corrected. For instance, if the health label of the stored data is "faulty," it will be corrected to "healthy," and if the health label of the stored data is "healthy," it will be corrected to "faulty"; for data with an undetermined label accuracy category, in this embodiment, this type of data can be deleted.
[0083] To facilitate understanding by those skilled in the art, the above-mentioned data label correction method is described in detail, such as... Figure 4 As shown, the method may include:
[0084] S401, if the fault prediction model detects a faulty hard drive, it maps all the stored data in the faulty hard drive to the same space to obtain the spatial characteristics of all the stored data in the same space.
[0085] S402, determine the first feature distance of each data pair based on the spatial characteristics of the two stored data contained in each data pair.
[0086] In this case, any two stored data items in the faulty hard drive constitute a data pair.
[0087] S403, based on the spatial characteristics of each stored data in each data pair and the spatial characteristics of the reference data of each data pair, determine the second feature distance between each stored data in each data pair and the reference data of each data pair.
[0088] The second feature distance includes the first sub-distance and the second sub-distance; the reference data for each data pair is any other stored data in the faulty hard disk other than the data pair.
[0089] S404, for each data pair, if the sum of the first sub-distance and the second sub-distance of the data pair is greater than the first feature distance, and the health labels of the two stored data contained in the data pair are consistent, then the adjacency relationship between the two stored data contained in the data pair is determined to be adjacency.
[0090] S405, determine the adjacent data corresponding to each storage data according to the adjacency relationship between the two stored data contained in each data pair.
[0091] S406, Based on the health label of each stored data and the health label of the adjacent data corresponding to each stored data, determine the proportion of adjacent data with the same label for each stored data.
[0092] S407, determine the label accuracy category of each stored data based on the proportion of adjacent data with the same label.
[0093] The label accuracy categories include: Correct, Incorrect, and Pending.
[0094] S408, based on the label accuracy category of each stored data, determine whether the proportion of stored data with incorrect label accuracy category is less than a preset proportion threshold; if yes, proceed to S409; if no, proceed to S411.
[0095] S409, determine a label correction strategy for each stored data based on the label accuracy category of each stored data.
[0096] S410, corrects the health labels of each stored data according to the label correction strategy of each stored data.
[0097] Among them, the stored data after correcting the health status labels are used to retrain the fault prediction model.
[0098] S411, delete all stored data on the faulty hard drive.
[0099] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.
[0100] Based on the same inventive concept, this application also provides a data tag correction apparatus for implementing the data tag correction method described above. The solution provided by this apparatus is similar to the implementation described in the above method; therefore, the specific limitations in one or more data tag correction apparatus embodiments provided below can be found in the limitations of the data tag correction method described above, and will not be repeated here.
[0101] In one embodiment, such as Figure 5 As shown, a data tag correction device 1 is provided, comprising: a distance determination module 10, a strategy determination module 11, and a tag correction module 12, wherein:
[0102] The distance determination module 10 is used to determine, if the fault prediction model detects a faulty hard disk, a first feature distance between two stored data in each data pair in the faulty hard disk, and a second feature distance between each stored data in each data pair and the reference data of each data pair.
[0103] In this context, any two stored data in the faulty hard drive constitute a data pair; the reference data for each data pair is any other stored data in the faulty hard drive other than the data pair itself.
[0104] The strategy determination module 11 is used to determine the label correction strategy for each stored data based on the first feature distance and the second feature distance of each data pair, as well as the health label of each stored data.
[0105] The label correction module 12 is used to correct the health labels of each stored data according to the label correction strategy of each stored data; the stored data after correcting the health labels are used to retrain the fault prediction model.
[0106] In one embodiment, such as Figure 6As shown, the distance determination module 10 includes a spatial mapping unit 100, a first determination unit 101, and a second determination unit 102. Wherein:
[0107] The space mapping unit 100 is used to map the stored data in the faulty hard disk to the same space, so as to obtain the spatial characteristics of the stored data in the same space.
[0108] The first determining unit 101 is used to determine the first feature distance of each data pair based on the spatial characteristics of the two stored data contained in each data pair.
[0109] The second determining unit 102 is used to determine the second feature distance between each stored data in each data pair and the reference data of each data pair based on the spatial characteristics of each stored data in each data pair and the spatial characteristics of the reference data of each data pair.
[0110] In one embodiment, such as Figure 7 As shown, the strategy determination module 11 includes a third determination unit 110 and a fourth determination unit 111. Wherein:
[0111] The third determining unit 110 is used to determine the adjacency relationship between the two stored data contained in each data pair based on the first feature distance and the second feature distance of each data pair, as well as the consistency of the health labels of the two stored data contained in each data pair.
[0112] The fourth determining unit 111 is used to determine the label correction strategy for each stored data based on the adjacency relationship between the two stored data contained in each data pair and the health label of each stored data.
[0113] In one embodiment, the second feature distance includes a first sub-distance and a second sub-distance. The third determining unit 110 is specifically used to determine the adjacency relationship between the two stored data in the data pair as adjacency for each data pair if the sum of the first sub-distance and the second sub-distance of the data pair is greater than the first feature distance and the health labels of the two stored data contained in the data pair are consistent.
[0114] In one embodiment, the fourth determining unit 111 includes a first determining subunit and a second determining subunit, wherein:
[0115] The first determining subunit is used to determine the adjacent data corresponding to each stored data based on the adjacency relationship between the two stored data contained in each data pair.
[0116] The second determining subunit is used to determine the label correction strategy for each stored data based on the health label of each stored data and the health labels of the adjacent data corresponding to each stored data.
[0117] In one embodiment, the second determining subunit is specifically used to determine the label accuracy category of each stored data based on the health label of each stored data and the health labels of the adjacent data corresponding to each stored data.
[0118] The label accuracy categories include: correct, incorrect, and pending; a label correction strategy is determined for each stored data based on its label accuracy category.
[0119] In one embodiment, the second determining subunit is further configured to determine the same-label adjacency ratio of each stored data based on the health label of each stored data and the health labels of the adjacent data corresponding to each stored data; and to determine the label accuracy category of each stored data based on the same-label adjacency ratio of each stored data.
[0120] In one embodiment, the second determining subunit is further configured to determine whether the proportion of stored data with incorrect label accuracy category is less than a preset proportion threshold based on the label accuracy category of each stored data; if so, then determine a label correction strategy for each stored data based on the label accuracy category of each stored data.
[0121] Each module in the aforementioned data tag correction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.
[0122] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 8 As shown, the computer device includes a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, NFC (Near Field Communication), or other technologies. When executed by the processor, the computer program implements a data tag correction method. The display screen can be an LCD screen or an e-ink display screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.
[0123] Those skilled in the art will understand that Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.
[0124] In one embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:
[0125] If the fault prediction model detects a faulty hard drive, it determines the first feature distance between the two stored data in each data pair in the faulty hard drive, and the second feature distance between each stored data in each data pair and the reference data of each data pair; wherein, any two stored data in the faulty hard drive constitute a data pair; the reference data of each data pair is any other stored data in the faulty hard drive other than the data pair.
[0126] Based on the first feature distance and the second feature distance of each data pair, as well as the health label of each stored data, determine the label correction strategy for each stored data.
[0127] Based on the label correction strategy for each stored data, the health labels of each stored data are corrected; the stored data with corrected health labels are then used to retrain the fault prediction model.
[0128] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0129] Map the stored data in the faulty hard drive to the same space to obtain the spatial characteristics of the stored data in the same space;
[0130] Based on the spatial characteristics of the two stored data contained in each data pair, determine the first feature distance of each data pair;
[0131] Based on the spatial characteristics of each stored data in each data pair and the spatial characteristics of the reference data of each data pair, determine the second feature distance between each stored data in each data pair and the reference data of each data pair.
[0132] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0133] Based on the first feature distance and the second feature distance of each data pair, as well as the consistency of the health labels of the two stored data contained in each data pair, the adjacency relationship between the two stored data contained in each data pair is determined.
[0134] Based on the adjacency relationship between the two stored data contained in each data pair, and the health label of each stored data, determine the label correction strategy for each stored data.
[0135] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0136] For each data pair, if the sum of the first sub-distance and the second sub-distance of the data pair is greater than the first feature distance, and the health labels of the two stored data contained in the data pair are consistent, then the adjacency relationship between the two stored data contained in the data pair is determined to be adjacency.
[0137] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0138] Based on the adjacency relationship between the two stored data contained in each data pair, determine the adjacent data corresponding to each stored data;
[0139] Based on the health label of each stored data and the health labels of its adjacent data, a label correction strategy is determined for each stored data.
[0140] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0141] Based on the health label of each stored data and the health labels of its adjacent data, the label accuracy category of each stored data is determined; the label accuracy categories include: correct, incorrect, and pending.
[0142] A label correction strategy is determined for each stored data based on the label accuracy category of each stored data.
[0143] In one embodiment, when the processor executes the computer program, it further performs the following steps: determining the proportion of adjacent data with the same label for each stored data based on the health label of each stored data and the health labels of adjacent data corresponding to each stored data.
[0144] The label accuracy category of each stored data is determined based on the proportion of adjacent data with the same label.
[0145] In one embodiment, the processor, when executing a computer program, also performs the following steps:
[0146] Based on the label accuracy category of each stored data, determine whether the proportion of stored data with incorrect label accuracy category is less than a preset proportion threshold;
[0147] If so, a label correction strategy is determined for each stored data based on the label accuracy category of each stored data.
[0148] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor:
[0149] If the fault prediction model detects a faulty hard drive, it determines the first feature distance between the two stored data in each data pair in the faulty hard drive, and the second feature distance between each stored data in each data pair and the reference data of each data pair; wherein, any two stored data in the faulty hard drive constitute a data pair; the reference data of each data pair is any other stored data in the faulty hard drive other than the data pair.
[0150] Based on the first feature distance and the second feature distance of each data pair, as well as the health label of each stored data, determine the label correction strategy for each stored data.
[0151] Based on the label correction strategy for each stored data, the health labels of each stored data are corrected; the stored data with corrected health labels are then used to retrain the fault prediction model.
[0152] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0153] Map the stored data in the faulty hard drive to the same space to obtain the spatial characteristics of the stored data in the same space;
[0154] Based on the spatial characteristics of the two stored data contained in each data pair, determine the first feature distance of each data pair;
[0155] Based on the spatial characteristics of each stored data in each data pair and the spatial characteristics of the reference data of each data pair, determine the second feature distance between each stored data in each data pair and the reference data of each data pair.
[0156] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0157] Based on the first feature distance and the second feature distance of each data pair, as well as the consistency of the health labels of the two stored data contained in each data pair, the adjacency relationship between the two stored data contained in each data pair is determined.
[0158] Based on the adjacency relationship between the two stored data contained in each data pair, and the health label of each stored data, determine the label correction strategy for each stored data.
[0159] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0160] For each data pair, if the sum of the first sub-distance and the second sub-distance of the data pair is greater than the first feature distance, and the health labels of the two stored data contained in the data pair are consistent, then the adjacency relationship between the two stored data contained in the data pair is determined to be adjacency.
[0161] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0162] Based on the adjacency relationship between the two stored data contained in each data pair, determine the adjacent data corresponding to each stored data;
[0163] Based on the health label of each stored data and the health labels of its adjacent data, a label correction strategy is determined for each stored data.
[0164] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0165] Based on the health label of each stored data and the health labels of its adjacent data, the label accuracy category of each stored data is determined; the label accuracy categories include: correct, incorrect, and pending.
[0166] A label correction strategy is determined for each stored data based on the label accuracy category of each stored data.
[0167] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0168] Based on the health label of each stored data and the health label of the adjacent data corresponding to each stored data, determine the proportion of adjacent data with the same label for each stored data.
[0169] The label accuracy category of each stored data is determined based on the proportion of adjacent data with the same label.
[0170] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0171] Based on the label accuracy category of each stored data, determine whether the proportion of stored data with incorrect label accuracy category is less than a preset proportion threshold;
[0172] If so, a label correction strategy is determined for each stored data based on the label accuracy category of each stored data.
[0173] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, performs the following steps:
[0174] If the fault prediction model detects a faulty hard drive, it determines the first feature distance between the two stored data in each data pair in the faulty hard drive, and the second feature distance between each stored data in each data pair and the reference data of each data pair; wherein, any two stored data in the faulty hard drive constitute a data pair; the reference data of each data pair is any other stored data in the faulty hard drive other than the data pair.
[0175] Based on the first feature distance and the second feature distance of each data pair, as well as the health label of each stored data, determine the label correction strategy for each stored data.
[0176] Based on the label correction strategy for each stored data, the health labels of each stored data are corrected; the stored data with corrected health labels are then used to retrain the fault prediction model.
[0177] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0178] Map the stored data in the faulty hard drive to the same space to obtain the spatial characteristics of the stored data in the same space;
[0179] Based on the spatial characteristics of the two stored data contained in each data pair, determine the first feature distance of each data pair;
[0180] Based on the spatial characteristics of each stored data in each data pair and the spatial characteristics of the reference data of each data pair, determine the second feature distance between each stored data in each data pair and the reference data of each data pair.
[0181] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0182] Based on the first feature distance and the second feature distance of each data pair, as well as the consistency of the health labels of the two stored data contained in each data pair, the adjacency relationship between the two stored data contained in each data pair is determined.
[0183] Based on the adjacency relationship between the two stored data contained in each data pair, and the health label of each stored data, determine the label correction strategy for each stored data.
[0184] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0185] For each data pair, if the sum of the first sub-distance and the second sub-distance of the data pair is greater than the first feature distance, and the health labels of the two stored data contained in the data pair are consistent, then the adjacency relationship between the two stored data contained in the data pair is determined to be adjacency.
[0186] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0187] Based on the adjacency relationship between the two stored data contained in each data pair, determine the adjacent data corresponding to each stored data;
[0188] Based on the health label of each stored data and the health labels of its adjacent data, a label correction strategy is determined for each stored data.
[0189] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0190] Based on the health label of each stored data and the health labels of its adjacent data, the label accuracy category of each stored data is determined; the label accuracy categories include: correct, incorrect, and pending.
[0191] A label correction strategy is determined for each stored data based on the label accuracy category of each stored data.
[0192] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0193] Based on the health label of each stored data and the health label of the adjacent data corresponding to each stored data, determine the proportion of adjacent data with the same label for each stored data.
[0194] The label accuracy category of each stored data is determined based on the proportion of adjacent data with the same label.
[0195] In one embodiment, when the computer program is executed by a processor, it also performs the following steps:
[0196] Based on the label accuracy category of each stored data, determine whether the proportion of stored data with incorrect label accuracy category is less than a preset proportion threshold;
[0197] If so, a label correction strategy is determined for each stored data based on the label accuracy category of each stored data.
[0198] It should be noted that the information (including but not limited to information on faulty hard drives and tag correction strategies) and data (including but not limited to storage data and data pairs) involved in this application are all information and data authorized by the user or fully authorized by all parties.
[0199] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.
[0200] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.
[0201] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.
Claims
1. A method for correcting data tags, characterized in that, The method includes: If the fault prediction model detects a faulty hard drive, it determines the first feature distance between the two stored data in each data pair of the faulty hard drive, and the second feature distance between each stored data in each data pair and the reference data of each data pair; wherein, any two stored data in the faulty hard drive constitute a data pair; the reference data of each data pair is any other stored data in the faulty hard drive other than the data pair. Based on the first feature distance and the second feature distance of each data pair, and the health label of each stored data, a label correction strategy for each stored data is determined. The health label of each stored data is the fault prediction result labeled for the stored data after the fault prediction model performs fault prediction on the stored data. Based on the label correction strategy for each stored data, the health labels of each stored data are corrected; the stored data with corrected health labels are used to retrain the fault prediction model. The step of determining the label correction strategy for each stored data based on the first feature distance and the second feature distance of each data pair, as well as the health label of each stored data, includes: Based on the first feature distance and the second feature distance of each data pair, as well as the consistency of the health labels of the two stored data contained in each data pair, the adjacency relationship between the two stored data contained in each data pair is determined. Based on the adjacency relationship between the two stored data contained in each data pair, and the health label of each stored data, a label correction strategy for each stored data is determined.
2. The method according to claim 1, characterized in that, Determining the first characteristic distance between the two stored data items contained in each data pair in the faulty hard disk, and the second characteristic distance between each stored data item in each data pair and the reference data of each data pair, includes: Map each stored data in the faulty hard drive to the same space to obtain the spatial characteristics of each stored data in the same space; Based on the spatial characteristics of the two stored data contained in each data pair, determine the first feature distance of each data pair; Based on the spatial characteristics of each stored data in each data pair and the spatial characteristics of the reference data of each data pair, determine the second feature distance between each stored data in each data pair and the reference data of each data pair.
3. The method according to claim 1, characterized in that, The second feature distance includes a first sub-distance and a second sub-distance; determining the adjacency relationship between the two stored data in each data pair based on the first and second feature distances of each data pair, and the consistency of the health labels of the two stored data contained in each data pair, includes: For each data pair, if the sum of the first sub-distance and the second sub-distance of the data pair is greater than the first feature distance, and the health labels of the two stored data contained in the data pair are consistent, then the adjacency relationship between the two stored data contained in the data pair is determined to be adjacency.
4. The method according to claim 1, characterized in that, The step of determining the label correction strategy for each stored data based on the adjacency relationship between the two stored data contained in each data pair and the health label of each stored data includes: Based on the adjacency relationship between the two stored data contained in each data pair, determine the adjacent data corresponding to each stored data; Based on the health label of each stored data and the health labels of its adjacent data, a label correction strategy is determined for each stored data.
5. The method according to claim 4, characterized in that, The step of determining the label correction strategy for each stored data based on the health label of each stored data and the health labels of its adjacent data includes: Based on the health label of each stored data and the health labels of its adjacent data, the label accuracy category of each stored data is determined; wherein, the label accuracy category includes: correct, incorrect, and pending. A label correction strategy is determined for each stored data based on the label accuracy category of each stored data.
6. The method according to claim 5, characterized in that, The step of determining the label accuracy category of each stored data based on the health label of each stored data and the health labels of its adjacent data includes: Based on the health label of each stored data and the health label of the adjacent data corresponding to each stored data, determine the proportion of adjacent data with the same label for each stored data. The label accuracy category of each stored data is determined based on the proportion of adjacent data with the same label.
7. The method according to claim 5, characterized in that, The step of determining a label correction strategy for each stored data based on the label accuracy category of each stored data includes: Based on the label accuracy category of each stored data, determine whether the proportion of stored data with incorrect label accuracy category is less than a preset proportion threshold; If so, a label correction strategy is determined for each stored data based on the label accuracy category of each stored data.
8. A data tag correction device, characterized in that, The device includes: The distance determination module is used to determine, if the fault prediction model detects a faulty hard drive, a first feature distance between two stored data items contained in each data pair in the faulty hard drive, and a second feature distance between each stored data item in each data pair and the reference data of each data pair; wherein, any two stored data items in the faulty hard drive constitute a data pair; and the reference data of each data pair is any other stored data item in the faulty hard drive other than the data pair. The strategy determination module is used to determine the label correction strategy for each stored data based on the first feature distance and the second feature distance of each data pair, as well as the health label of each stored data. The health label of each stored data is the fault prediction result labeled for the stored data after the fault prediction model performs fault prediction on the stored data. The label correction module is used to correct the health labels of each stored data according to the label correction strategy of each stored data; the stored data after correcting the health labels are used to retrain the fault prediction model. The strategy determination module determines the adjacency relationship between the two stored data in each data pair based on the first feature distance and the second feature distance of each data pair, as well as the consistency of the health labels of the two stored data contained in each data pair. Based on the adjacency relationship between the two stored data contained in each data pair, and the health label of each stored data, a label correction strategy for each stored data is determined.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.
11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.