Hard disk replacement determination method, device, equipment and readable storage medium

Abstract

The application discloses a hard disk replacement discrimination method, device and equipment and a readable storage medium. The hard disk replacement discrimination method comprises the following steps: obtaining a hard disk running time corresponding to a target hard disk, extracting a health characteristic value and a retirement characteristic value in the hard disk running time; constructing a time-varying matrix of the health characteristic value in the hard disk running time to obtain health characteristic data; constructing a time-varying matrix of the retirement characteristic value in the hard disk running time to obtain retirement characteristic data; and discriminating whether the target hard disk needs to be replaced by performing health state prediction and retirement state prediction on the target hard disk based on a hard disk state prediction model, the health characteristic data and the retirement characteristic data. The application solves the technical problem of low hard disk replacement discrimination accuracy.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology in financial technology (Fintech), and in particular to a hard drive replacement detection method, apparatus, device, and readable storage medium. Background Technology

[0002] With the continuous development of fintech, especially internet fintech, more and more technologies (such as distributed systems and artificial intelligence) are being applied in the financial field. However, the financial industry is also placing higher demands on technology, such as on the distribution of tasks to be completed.

[0003] With the deepening of informatization, the value of data is increasing, and the medium for data storage is becoming more and more important. Currently, in order to ensure the security of data on hard drives, the first step is usually to monitor SMART (Self-Monitoring Analysis and Reporting Technology) data. Then, by setting thresholds or rules, SMART data is detected to determine whether the hard drive has failed and whether it needs to be replaced. However, this method cannot predict hard drive failures in advance, and therefore cannot guarantee that the hard drive will be replaced before it fails. This can easily lead to data loss due to hard drive failure. Therefore, the accuracy of current hard drive replacement determination is low. Summary of the Invention

[0004] The main objective of this application is to provide a hard disk replacement identification method, apparatus, device, and readable storage medium, aiming to solve the technical problem of low accuracy in hard disk replacement identification in the prior art.

[0005] To achieve the above objectives, this application provides a hard drive replacement detection method, which is applied to a hard drive replacement detection device, and includes:

[0006] Obtain the hard drive runtime corresponding to the target hard drive, and extract the health feature value and retirement feature value during the hard drive runtime;

[0007] Construct a matrix of the changes of the health feature values ​​over time during the operation of the hard disk to obtain health feature data;

[0008] Construct a time-varying matrix of the retirement feature values ​​during the hard disk's operating time to obtain retirement feature data;

[0009] Based on the hard drive status prediction model, the health feature data, and the retirement feature data, the system determines whether the target hard drive needs to be replaced by predicting its health status and retirement status.

[0010] This application also provides a hard drive replacement detection device, which is a virtual device and is applied to a hard drive replacement detection equipment. The hard drive replacement detection device includes:

[0011] The extraction module is used to obtain the hard drive running time corresponding to the target hard drive, and extract the health feature value and retirement feature value of the hard drive running time;

[0012] The first construction module is used to construct a time-varying matrix of the health feature values ​​during the hard disk's operating time to obtain health feature data.

[0013] The second construction module is used to construct a time-varying matrix of the retirement feature values ​​during the hard disk's operating time to obtain retirement feature data.

[0014] The discrimination module is used to determine whether the target hard drive needs to be replaced based on the hard drive status prediction model, the health feature data, and the retirement feature data by predicting the health status and retirement status of the target hard drive.

[0015] This application also provides a hard disk replacement identification device, which is a physical device. The hard disk replacement identification device includes: a memory, a processor, and a program for the hard disk replacement identification method stored in the memory and executable on the processor. When the program for the hard disk replacement identification method is executed by the processor, it can implement the steps of the hard disk replacement identification method as described above.

[0016] This application also provides a readable storage medium storing a program that implements a hard disk replacement detection method. When the program is executed by a processor, it implements the steps of the hard disk replacement detection method as described above.

[0017] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the hard disk replacement determination method described above.

[0018] This application provides a hard drive replacement determination method, apparatus, device, readable storage medium, and computer program product. Compared with the existing technology that uses SMART data detection by setting thresholds or rules to determine whether a hard drive has failed and whether it needs to be replaced, this application first obtains the retirement characteristic data and corresponding health characteristic data of the target hard drive. Then, based on the hard drive status prediction model, the health characteristic data, and the retirement characteristic data, it predicts the health status and retirement status of the target hard drive. This achieves the purpose of predicting whether the target hard drive is healthy and whether it needs to be retired. Based on the results of the prediction of the target hard drive's health and whether it needs to be retired, it can comprehensively consider whether to replace the target hard drive before it fails (is damaged), that is, determine whether the target hard drive needs to be replaced. This overcomes the technical defect of the method of detecting SMART data by setting thresholds or rules to determine whether a hard drive has failed and whether it needs to be replaced, which cannot guarantee that the hard drive will be replaced before it may fail, and is prone to data loss due to hard drive failure. This improves the accuracy of hard drive replacement determination. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating the first embodiment of the hard drive replacement determination method of this application;

[0022] Figure 2 This is a schematic diagram of the parameter matrix representing the change of health characteristic values ​​over time during the hard drive's operating period in the hard drive replacement discrimination method of this application;

[0023] Figure 3 This is a schematic diagram of the parameter matrix representing the change of retirement characteristic values ​​over time during the hard drive's operating life in the hard drive replacement discrimination method of this application;

[0024] Figure 4This is a flowchart illustrating the process of predicting the health status of the target hard drive and obtaining the health status prediction result based on a convolutional neural network model and a recurrent neural network model in the hard drive replacement discrimination method of this application.

[0025] Figure 5 This is a flowchart illustrating the second embodiment of the hard disk replacement determination method of this application;

[0026] Figure 6 This is a flowchart illustrating the process of a two-level prediction model based on a health status prediction model and a retirement status prediction model in the hard drive replacement discrimination method of this application, which makes two predictions on whether the target hard drive is healthy and whether it needs to be retired.

[0027] Figure 7 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of this application.

[0028] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0029] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of this application.

[0030] First, it should be understood that when SMART data is detected by setting thresholds or rules to determine whether a hard drive has failed, it is usually because the hard drive has failed first, causing a certain indicator value in the SMART data to not meet the set threshold or rule. Thus, the detection device or user can obtain information that the hard drive has failed. Therefore, when SMART data is detected by setting thresholds or rules to determine whether a hard drive has failed, there is a high probability that the hard drive has actually failed (damaged), which can easily lead to the loss of data stored on the hard drive and cannot guarantee the absolute safety of the data stored on the hard drive.

[0031] This application provides a hard drive replacement determination method. In the first embodiment of the hard drive replacement determination method of this application, refer to... Figure 1 The hard drive replacement detection method includes:

[0032] Step S10: Obtain the hard drive running time corresponding to the target hard drive, and extract the health feature value and retirement feature value during the hard drive running time;

[0033] Step S20: Construct a matrix showing the change of the health feature values ​​over time during the hard disk's operating time to obtain health feature data;

[0034] Step S30: Construct a time-varying matrix of the retirement feature values ​​during the hard disk's operating time to obtain retirement feature data;

[0035] In this embodiment, it should be noted that the health feature data is feature data that measures the health status of the target hard drive during its operating time. Specifically, it can be data on the change of the health feature value of the target hard drive over time during its operating time, such as a parameter matrix or line graph representing the change of the health feature value over time. The health feature corresponding to the health feature value includes at least one of the following: Reallocated Sector Count, Command Timeout, Current Pending Sector Count, Reported Uncorrectable Errors, and Offline Uncorrectable Sector Count. Figure 2 , Figure 2 This is a schematic diagram of a parameter matrix representing the change of health feature values ​​over time during the hard drive's operation. Here, t1 to tn represent the operation time of the target hard drive, which can be in days or hours, etc., and h1 to h5 represent each health feature. Each value in the matrix is ​​a health feature value.

[0036] Additionally, it should be noted that even when a hard drive is healthy and can still function normally, some indicators may show signs of decline, or the hard drive may have exceeded its service life. In such cases, it is necessary to retire the hard drive and replace it. The retirement characteristic data refers to the characteristic data that measures the retirement status of the target hard drive during its operating time. Specifically, it can be data on the changes of the retirement characteristic values ​​of the target hard drive over time, such as a parameter matrix or line graph showing the changes of retirement characteristic values ​​over time. The retirement characteristics corresponding to the retirement characteristic values ​​include at least one of the following: seek error rate, hard drive temperature, cumulative power-on time, and hard drive business complexity. The hard drive temperature refers to the current internal temperature of the hard drive. While excessively high temperatures do not cause data loss, the resulting mechanical deformation can increase seek and read / write error rates, reduce hard drive performance, and thus affect the hard drive's retirement status. The hard drive business complexity refers to the complexity of the business handled by the hard drive. Higher business complexity indicates more frequent read / write operations, and generally, a higher number of erase / write cycles for hard drive storage blocks results in a shorter lifespan, thus affecting the hard drive's retirement status. Figure 3 , Figure 3This is a schematic diagram of a parameter matrix representing the change of retirement feature values ​​over time during the hard drive's operating time. Here, t1 to tn represent the operating time of the target hard drive, which can be in days or hours, etc., and r1 to r5 represent each retirement feature. Each value in the matrix is ​​the retirement feature value.

[0037] Additionally, it should be noted that when constructing training sample data for the hard disk status prediction model, the original decommissioning feature data can be directly copied, and the hard disk business complexity can be adjusted to expand the sample and increase the richness of the training sample data for the hard disk status prediction model. For example, if the decommissioning feature data is matrix A, and the hard disk business complexity in matrix A is 0.1, then after copying matrix A, the hard disk business complexity in the copied matrix A can be adjusted, for example, from 0.1 to 0.2 or from 0.1 to 0.3, etc. The required adjustment value can be set by yourself to obtain a new decommissioning feature data, thereby realizing the sample expansion of the decommissioning feature data.

[0038] Specifically, step S10 includes: obtaining the hard drive running time corresponding to the target hard drive, and periodically extracting the feature values ​​of each health feature during the hard drive running time according to the time sequence of the generation of health feature values ​​and retirement feature values ​​of the target hard drive, to obtain health feature values. Similarly, periodically extracting the feature values ​​of each retirement feature during the hard drive running time, to obtain retirement feature values. The periodic extraction time period can be set to 1 day, 12 hours, or 6 hours, etc. Further, step S20 includes: constructing a time-varying matrix of the health feature values ​​during the hard drive running time according to the time sequence of periodic extraction of each health feature value in the health feature data, to obtain health feature data. This achieves the purpose of constructing health feature data with time information, so that the health feature data has both the distribution information of health feature values ​​and the changing trend information of health feature values. Each column of the time-varying matrix of health feature values ​​during the hard drive running time corresponds to a periodic extraction time period, and each row corresponds to a health feature. Figure 2If the periodic extraction time period is set to 1 day, then t1 can represent the first day of hard drive operation, t2 can represent the second day of hard drive operation, and correspondingly, tn can represent the nth day of hard drive operation. h1 to h5 represent various health characteristics. Further, step S30 includes: constructing a time-varying matrix of the retirement characteristic values ​​during the hard drive's operation time based on the time sequence of periodic extraction of each retirement characteristic value in the retirement characteristic data during the hard drive's operation time, thereby obtaining the retirement characteristic data. Each column of the time-varying matrix of the retirement characteristic values ​​during the hard drive's operation time corresponds to a periodic extraction time period, and each row corresponds to a retirement characteristic. This achieves the goal of constructing retirement characteristic data with time information, ensuring that the retirement characteristic data possesses both the distribution information and the trend information of the retirement characteristic values. Figure 3 If the periodic extraction time period is set to 1 day, then t1 can represent the first day of hard disk operation, t2 can represent the second day of hard disk operation, and correspondingly, tn can represent the nth day of hard disk operation, while r1 to r4 are the respective retirement characteristics.

[0039] Step S40: Based on the hard disk status prediction model, the health feature data, and the retirement feature data, determine whether the target hard disk needs to be replaced by predicting its health status and retirement status.

[0040] In this embodiment, it should be noted that the hard disk status prediction model includes a health status prediction model and a retirement status prediction model. The health status prediction model is used to predict the health status of the target hard disk to predict the probability that the target hard disk is healthy, and the retirement status prediction model is used to predict the retirement status of the target hard disk to predict the probability that the target hard disk needs to be retired.

[0041] Based on the hard drive status prediction model, the health feature data, and the retirement feature data, the system predicts the health status and retirement status of the target hard drive to determine whether it needs to be replaced. Specifically, the health feature data is input into the health status prediction model to predict the health status of the target hard drive, obtaining a health status prediction result. Similarly, the retirement feature data is input into the retirement status prediction model to predict the retirement status of the target hard drive, obtaining a retirement status prediction result. Based on these two predictions, the system determines whether the target hard drive needs to be replaced. Since the retirement feature values ​​include at least one of hard drive temperature and hard drive business complexity, this adds at least one of these two factors to the hard drive replacement decision, fully considering the impact of hard drive temperature and / or business complexity on the hard drive's retirement status, thus improving the accuracy of hard drive replacement determination.

[0042] Furthermore, in another embodiment, the hard disk status prediction model is a multi-classification model, and step S40 includes: merging the health feature data and the retirement feature data to obtain hard disk feature data; inputting the hard disk feature data into the hard disk status prediction model for multi-classification to obtain multi-classification results; based on the multi-classification results, simultaneously predicting the health status and retirement status of the target hard disk to obtain the health probability and retirement probability corresponding to the target hard disk, and then determining whether the target hard disk needs to be replaced based on the health prediction probability and the retirement prediction probability.

[0043] Further, the step of inputting the hard drive feature data into the hard drive status prediction model for multi-classification to obtain multi-classification results includes: extracting features from the hard drive feature data based on the feature extractor of the hard drive status prediction model to obtain hard drive feature extraction results; and then mapping the hard drive feature extraction results into multi-classification results based on the classifier of the hard drive status prediction model. The multi-classification results can exist in matrix or vector form, and can be converted into a multi-classification probability vector through a preset activation function. The multi-classification probability vector consists of a health prediction probability and a retirement prediction probability. The health prediction probability is the probability that the target hard drive is predicted to be healthy based on the multi-classification model, and the retirement prediction probability is the probability that the target hard drive needs to be retired based on the multi-classification model.

[0044] Furthermore, the step of determining whether the target hard drive needs to be replaced based on the health prediction probability and the retirement prediction probability includes:

[0045] Based on preset aggregation rules, the health prediction probability and the retirement prediction probability are aggregated to obtain the aggregated prediction probability. The aggregated prediction probability is the probability that the target hard drive needs to be replaced based on the multi-classification model. If the aggregated prediction probability is greater than the preset hard drive replacement probability threshold, it is determined that the target hard drive needs to be replaced. If the aggregated prediction probability is not greater than the preset hard drive replacement probability threshold, it is determined that the target hard drive does not need to be replaced.

[0046] Furthermore, in step S40, the hard disk status prediction model includes a health status prediction model and a retirement status prediction model.

[0047] The step of determining whether a target hard drive needs to be replaced by predicting its health status and retirement status based on the hard drive status prediction model, the health feature data, and the retirement feature data includes:

[0048] Step S41: By inputting the health feature data into the health status prediction model, the health status of the target hard drive is predicted to obtain the health status prediction result.

[0049] In this embodiment, it should be noted that the health status prediction result can be a prediction of the target hard drive being healthy, and the health status prediction model can be a convolutional neural network model, a decision tree model, or a recurrent neural network model.

[0050] By inputting the health feature data into the health status prediction model, the health status of the target hard drive is predicted, and a health status prediction result is obtained. Specifically, by inputting the health feature data into the health status prediction model, the probability that the target hard drive is healthy is predicted, and a health probability is obtained.

[0051] Further, in step S41, the health status prediction model includes a convolutional neural network model and a recurrent neural network model.

[0052] The step of inputting the health feature data into the health status prediction model to predict the health status of the target hard drive and obtaining the health status prediction result includes:

[0053] Step S411: Input the health feature data into the convolutional neural network model to predict the health status of the target hard disk based on the distribution of the health feature data during the hard disk's operating time, and obtain a first health status prediction result.

[0054] In this embodiment, it should be noted that the convolutional neural network model includes a feature extractor for feature extraction and a classifier for fully connected components. The health feature data includes time-series health feature data, which can be a distribution matrix representing the changes in health feature values ​​over time during disk operation. The convolutional neural network model can be a neural network of types such as RNN, CNN, and ZfNet.

[0055] The health feature data is input into the convolutional neural network model to predict the health status of the target hard drive based on the distribution of the health feature data over the hard drive's operating time, obtaining a first health status prediction result. Specifically, the health feature data is input into the convolutional neural network model, and the feature extractor of the convolutional neural network model extracts features from the time-series health feature data to obtain feature extraction results. Then, the classifier of the convolutional neural network model performs a fully connected operation on the feature extraction results to obtain a fully connected result. Then, through a preset activation function, the fully connected result is mapped to a health probability, and the health probability is used as the first health status prediction result. Since the time-series health feature data is a distribution matrix representing the changes of health feature values ​​over time during the hard drive's operating time, the convolutional neural network model processes the entire distribution matrix of health feature values ​​over time during the hard drive's operating time. This achieves the goal of predicting the health probability of the target hard drive based on the overall distribution of the health feature data over the hard drive's operating time, rather than simply judging the probability of whether the target hard drive is healthy based on certain thresholds or rules. Therefore, the accuracy of health status prediction is improved.

[0056] Step S412: Input the health feature data into the recurrent neural network model to predict the health status of the target hard drive based on the changing trend of the health features, and obtain a second health status prediction result.

[0057] In this embodiment, it should be noted that the input of each time step of the recurrent neural network model is the output of the previous time step and the health feature value of the current time step. The time step can be set to a periodic extraction time period. Therefore, the decision-making process of each time step of the recurrent neural network model is based on the decision result of the previous time step, thereby enabling the prediction of the health status of the target hard drive based on the changing trend of health feature data. The recurrent neural network model includes LSTM network, etc.

[0058] The health feature data is input into the recurrent neural network model to predict the health status of the target hard drive based on the changing trend of the health features, thereby obtaining a second health status prediction result. Specifically, the health feature values ​​in the time-series health feature data are sequentially input into the recurrent neural network model according to their corresponding time order, so as to predict the health probability of the target hard drive being healthy based on the changing trend of the health feature values ​​in the health feature data, thereby obtaining a second health probability, and then using the second health probability as the second health status prediction result.

[0059] Step S413: Aggregate the first health status prediction result and the second health status prediction result to obtain the health status prediction result.

[0060] In this embodiment, specifically, based on preset aggregation rules, the health probability and the second health probability are aggregated to obtain an aggregated health probability, which is then used as the health status prediction result. The preset aggregation rules include summation and weighted averaging, thereby achieving the goal of predicting the health probability of the target hard drive by combining the distribution and changing trends of health feature data over the hard drive's operating time. This provides more decision-making basis for hard drive health status prediction, thus improving the accuracy of hard drive health status prediction. Figure 4 The diagram illustrates a process for predicting the health status of a target hard drive based on a convolutional neural network (CNN) model and a recurrent neural network (RNN) model. The RNN model is the long-short-term network model used in this process. The matrices of the CNN and RNN models are input to represent the time-series health feature data. AVG (Average Generation Group) is used for averaging, which is the preset aggregation rule. Both the CNN and RNN models predict whether the target hard drive is healthy or faulty (unhealthy). Specifically, the health probability of the target hard drive being healthy can be represented by a probability greater than a preset health probability threshold, indicating health, or a probability not greater than the preset health probability threshold, indicating fault. The preset health probability threshold can be set to 50% or 70%, etc.

[0061] Step S42: By inputting the decommissioning feature data into the decommissioning status prediction model, the decommissioning status of the target hard disk is predicted, and the decommissioning status prediction result is obtained.

[0062] In this embodiment, it should be noted that the retirement status prediction model can be a convolutional neural network model or a decision tree model, etc., and the retirement feature data includes time-series retirement feature data, wherein the time-series retirement feature data can be a distribution matrix representing the changes of retirement feature values ​​over time during the hard disk's operating time.

[0063] By inputting the retirement feature data into the retirement status prediction model, the retirement status of the target hard drive is predicted, and a retirement status prediction result is obtained. Specifically, the retirement feature data is input into the retirement status prediction model, and features are extracted from the time-series retirement feature data based on the feature extractor of the retirement status prediction model to obtain a second feature extraction result. Then, based on the classifier of the retirement status prediction model, a fully connected function is applied to the second feature extraction result to obtain a fully connected result. Then, the fully connected result is mapped to a retirement probability through a preset activation function, and the retirement probability is used as the retirement status prediction result. Since the time-series retirement feature data is a distribution matrix representing the change of retirement feature values ​​over time during the hard drive's operating time, the retirement status prediction model processes the distribution matrix of retirement feature values ​​over time as a whole, thereby achieving the purpose of predicting the retirement probability of the target hard drive based on the overall distribution of the retirement feature data over the hard drive's operating time.

[0064] Step S43: Aggregate the health status prediction result and the retirement status prediction result to obtain the hard disk status prediction result;

[0065] In this embodiment, based on a preset aggregation rule, the health prediction probability corresponding to the health status prediction result and the retirement probability corresponding to the retirement status prediction result are aggregated to obtain the hard drive replacement probability corresponding to the target hard drive. The hard drive replacement probability is used as the hard drive status prediction result. The preset aggregation rule includes summation and weighted average, etc. The health prediction probability is the probability of predicting that the target hard drive is healthy, which can be the health probability, the second health probability, or the aggregated health probability. The retirement probability is the probability of predicting that the target hard drive needs to be retired, and the hard drive replacement probability is the probability of predicting that the target hard drive needs to be replaced.

[0066] Step S44: Based on the hard drive status prediction result, determine whether the target hard drive needs to be replaced.

[0067] In this embodiment, based on the hard drive status prediction result, it is determined whether the target hard drive needs to be replaced. Specifically, the hard drive replacement probability is compared with a preset hard drive replacement probability threshold. If the hard drive replacement probability is greater than the preset hard drive replacement probability threshold, it is determined that the target hard drive needs to be replaced. If the hard drive replacement probability is not greater than the preset hard drive replacement probability threshold, it is determined that the target hard drive does not need to be replaced. This achieves a comprehensive consideration of whether to replace the hard drive based on the health status and retirement status of the target hard drive, providing more decision-making basis for determining whether to replace the hard drive, and thus improving the accuracy of hard drive replacement determination based on the advance determination of whether to replace the hard drive.

[0068] Furthermore, the hard disk status prediction model includes a decommissioning status prediction model.

[0069] Before the step of determining whether the target hard drive needs to be replaced by predicting its health status and retirement status based on the hard drive status prediction model, the health feature data, and the retirement feature data, the hard drive replacement determination method further includes:

[0070] Step Z10: Obtain the retirement status prediction model to be trained, and extract the hard drive temperature and hard drive business complexity of the target hard drive during the hard drive's operating time.

[0071] In this embodiment, it should be noted that the retirement state prediction model to be trained is an untrained retirement state prediction model, the hard disk temperature is the current internal temperature of the hard disk, and the hard disk business complexity represents the complexity of the business undertaken by the hard disk. The specific reasons why the hard disk temperature and the hard disk business complexity affect the retirement state of the hard disk can be referred to the detailed explanation below steps S10 to S30, which will not be repeated here.

[0072] The steps of extracting the hard disk temperature and hard disk business complexity of the training target hard disk during the hard disk operation time include: obtaining the hard disk operation time corresponding to the training target hard disk, and periodically extracting the hard disk temperature and hard disk business complexity during the hard disk operation time according to the time sequence in which the hard disk temperature and hard disk business complexity are generated during the hard disk operation time of the training target hard disk.

[0073] Step Z20: Construct a matrix showing the time-varying changes of the hard disk temperature and the hard disk business complexity during the hard disk's operating time to obtain training data for hard disk retirement features;

[0074] In this embodiment, a matrix showing the time-varying changes of the hard drive temperature and the hard drive business complexity during the hard drive's operating time is constructed to obtain hard drive retirement feature training data. Specifically, based on the time sequence of periodic extraction of the hard drive temperature and the hard drive business complexity during the hard drive's operating time, a matrix showing the time-varying changes of the hard drive temperature and the hard drive business complexity during the hard drive's operating time is constructed to obtain hard drive retirement feature training data.

[0075] Furthermore, when periodically extracting the hard disk temperature and hard disk business complexity of the training target hard disk during the hard disk operation time, other retirement feature values ​​generated by the training target hard disk during the hard disk operation time can also be periodically extracted, and a time-varying matrix of the hard disk temperature, the hard disk business complexity, and the other retirement feature values ​​during the hard disk operation time can be constructed to obtain hard disk retirement feature training data. The other retirement feature values ​​include seek error rate and cumulative power-on time, etc.

[0076] Step Z30: Based on the hard drive retirement feature training data and the hard drive retirement label corresponding to the training target hard drive, iteratively train and optimize the retirement status prediction model to be trained to obtain the retirement status prediction model.

[0077] In this embodiment, based on the hard drive retirement feature training data and the hard drive retirement label corresponding to the training target hard drive, the retirement state prediction model to be trained is iteratively trained and optimized to obtain the retirement state prediction model. Specifically, by inputting the hard drive retirement feature training data into the retirement state prediction model to be trained, the retirement state of the training target hard drive is predicted to obtain the retirement state training prediction result. Then, based on the difference between the retirement state training prediction result and the hard drive retirement label, the model loss corresponding to the retirement state prediction model to be trained is calculated, and it is then determined whether the model loss has converged. If the model loss has converged, the retirement state prediction model to be trained is used as the retirement state prediction model. If the model loss fails to converge, the decommissioning state prediction model is updated using a preset model update method based on the gradient calculated from the model loss, and the process returns to the execution step: extracting the hard drive temperature and hard drive business complexity during the training target hard drive's operating time. The preset model update method includes gradient descent and gradient ascent methods, thereby achieving the goal of constructing a decommissioning state prediction model based on hard drive temperature and hard drive business complexity. This allows the decommissioning state prediction model to accurately predict whether a hard drive needs to be decommissioned based on hard drive temperature and hard drive business complexity, adding two criteria for hard drive replacement judgment: hard drive temperature and hard drive business complexity, thus improving the accuracy of hard drive replacement judgment.

[0078] This application provides a hard drive replacement determination method. Compared to existing technologies that use thresholds or rules to detect SMART data to determine whether a hard drive has failed and whether it needs replacement, this application first obtains the retirement characteristic data and health characteristic data of the target hard drive. Then, based on the hard drive status prediction model, the health characteristic data, and the retirement characteristic data, it predicts the health status and retirement status of the target hard drive. This achieves the purpose of predicting whether the target hard drive is healthy and whether it needs to be retired. Based on the results of the predictions of the target hard drive's health and retirement status, it can comprehensively consider whether to replace the target hard drive before it fails (is damaged), thus determining whether the target hard drive needs to be replaced. This overcomes the technical defect of using thresholds or rules to detect SMART data to determine whether a hard drive has failed and whether it needs replacement, which cannot guarantee that the hard drive will be replaced before it fails and is prone to data loss due to hard drive failure. This improves the accuracy of hard drive replacement determination.

[0079] Furthermore, referring to Figure 5 Based on the first embodiment of this application, in another embodiment of this application, the hard disk status prediction model includes a health status prediction model and a retirement status prediction model.

[0080] The step of determining whether a target hard drive needs to be replaced by predicting its health status and retirement status based on the hard drive status prediction model, the health feature data, and the retirement feature data includes:

[0081] Step A10: By inputting the health feature data into the health status prediction model, the health status of the target hard drive is predicted to obtain the health status prediction result.

[0082] In this embodiment, the health status prediction result can be a prediction of the target hard drive being healthy, and the health status prediction model can be a convolutional neural network model, a decision tree model, or a recurrent neural network model.

[0083] Specifically, by inputting the health feature data into the health status prediction model, the probability that the target hard drive is healthy is predicted, and the health probability is obtained. The specific process of inputting the health feature data into the health status prediction model to predict the probability that the target hard drive is healthy and obtaining the health probability can be referred to the specific steps in steps S411 to S413, and will not be repeated here.

[0084] Step A20: If the health status prediction result is unhealthy, then it is determined that the target hard drive needs to be replaced.

[0085] In this embodiment, it should be noted that if the health probability is not greater than the preset health probability threshold, the health status prediction result is unhealthy, and the target hard drive is directly determined to need to be replaced. Therefore, there is no need to predict the retirement status of the target hard drive. That is, there is no need to call the retirement status prediction model to predict whether the target hard drive needs to be retired, which saves model running time and improves the efficiency of model replacement judgment.

[0086] Step A30: If the health status prediction result is healthy, then by inputting the retirement feature data into the retirement status prediction model, the retirement status of the target hard disk is predicted to obtain the retirement status prediction result.

[0087] In this embodiment, it should be noted that if the health probability is greater than the preset health probability threshold, the health status prediction result is healthy, and then it is further determined whether the target hard drive needs to be retired, so as to further determine whether the hard drive needs to be replaced, rather than directly determining that the hard drive does not need to be replaced, so as to improve the accuracy of hard drive replacement determination, and at the same time, it realizes the prediction of whether the hard drive needs to be updated in advance when the target hard drive is healthy.

[0088] If the health status prediction result is healthy, then the retirement status prediction of the target hard drive is performed by inputting the retirement feature data into the retirement status prediction model to obtain the retirement status prediction result. Specifically, if the health status prediction result is healthy, then the retirement probability of the target hard drive needing to be retired is predicted by inputting the retirement feature data into the retirement status prediction model, and the retirement probability is used as the retirement status prediction result. The specific process of predicting the retirement probability of the target hard drive needing to be retired by inputting the retirement feature data into the retirement status prediction model can be referred to the specific content in step S22, and will not be repeated here.

[0089] Step A40: Based on the retirement status prediction result, determine whether the target hard drive needs to be replaced.

[0090] In this embodiment, based on the retirement status prediction result, it is determined whether the target hard drive needs to be replaced. Specifically, the retirement probability is compared with a preset retirement probability threshold. If the retirement probability is greater than the preset retirement probability threshold, it is determined that the target hard drive needs to be replaced; if the retirement probability is not greater than the preset retirement probability threshold, it is determined that the target hard drive does not need to be replaced. This achieves a two-level prediction model based on a health status prediction model and a retirement status prediction model, performing dual predictions on the target hard drive's health and whether it needs to be retired, thus determining whether the target hard drive needs to be replaced. Therefore, instead of simply determining whether to replace the hard drive by setting thresholds or rules, it is possible to determine whether to replace the hard drive in advance, while also improving the accuracy of hard drive replacement determination. Figure 6 The diagram illustrates a two-tiered prediction model based on a health status prediction model and a decommissioning status prediction model. This model performs two-way predictions on the target hard drive: whether it is healthy and whether it needs to be decommissioned. The convolutional neural network model predicting health or failure is the same as the convolutional neural network model in the health status prediction model; the long-short time-series network model predicting health or failure is the same as the recurrent neural network model in the health status prediction model; and the convolutional neural network model predicting service or decommissioning is the decommissioning status prediction model. t1 to tn represent the time periods for periodic extraction, h1 to h5 represent health features, and r1 to r4 represent decommissioning features. The feature is that AVG is an average, which is the preset aggregation rule, to predict whether the target hard drive is healthy or faulty. Specifically, it can be represented by the health probability of the target hard drive being healthy. For example, if the health probability is greater than a preset probability value, it is healthy; if it is not greater than the probability value, it is faulty. The preset probability value can be set to 50% or 70%, etc. Similarly, it predicts whether the target hard drive is in service or retired. Specifically, it can be represented by the retirement probability of the target hard drive needing to be retired. For example, if the retirement probability is greater than a preset retirement probability threshold, it needs to be retired; if it is not greater than the preset retirement probability threshold, it does not need to be retired. The preset retirement probability threshold can be set to 50% or 70%, etc.

[0091] This application provides a hybrid neural network-based hard drive replacement determination method. Specifically, after acquiring the retirement feature data and corresponding health feature data of the target hard drive, the health feature data is input into a health status prediction model to predict the health status of the target hard drive, obtaining a health status prediction result. If the health status prediction result is unhealthy, the target hard drive is determined to need replacement; if the health status prediction result is healthy, the retirement feature data is input into the retirement status prediction model to predict the retirement status of the target hard drive, obtaining a retirement status prediction result. Based on the retirement status prediction result, it is determined whether the target hard drive needs replacement. In other words, it achieves the goal of... When the health status prediction result is healthy, it further determines whether the target hard drive needs to be retired, in order to further determine whether the hard drive needs to be replaced, rather than directly determining that the hard drive does not need to be replaced. This improves the accuracy of hard drive replacement determination. At the same time, it also achieves the prediction of whether the hard drive needs to be updated in advance when the target hard drive is healthy, further improving the accuracy of hard drive replacement determination. Therefore, it overcomes the technical defect of detecting SMART data by setting thresholds or rules to determine whether the hard drive has failed, and thus determining whether the failed hard drive needs to be replaced. This method cannot guarantee that the hard drive will be replaced before it may fail, and is prone to data loss due to hard drive failure. This improves the accuracy of hard drive replacement determination.

[0092] Reference Figure 7 , Figure 7 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of this application.

[0093] like Figure 7 As shown, the hard disk replacement detection device may include: a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002. The communication bus 1002 is used to establish communication between the processor 1001 and the memory 1005. The memory 1005 may be a high-speed RAM or a stable, non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0094] Optionally, the hard drive replacement detection device may also include a rectangular user interface, a network interface, a camera, RF (Radio Frequency) circuitry, sensors, hard drive circuitry, a WiFi module, etc. The rectangular user interface may include a display screen and an input submodule such as a keyboard. Optionally, the rectangular user interface may also include a standard wired interface or a wireless interface. The network interface may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface).

[0095] Those skilled in the art will understand that Figure 7 The hard drive replacement detection device structure shown does not constitute a limitation on the hard drive replacement detection device. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0096] like Figure 7 As shown, the memory 1005, as a computer storage medium, may include an operating system, a network communication module, and a hard disk replacement detection program. The operating system is a program that manages and controls the hardware and software resources of the hard disk replacement detection device, supporting the operation of the hard disk replacement detection program and other software and / or programs. The network communication module is used to enable communication between the various components within the memory 1005, as well as communication with other hardware and software in the hard disk replacement detection system.

[0097] exist Figure 7 In the hard disk replacement identification device shown, the processor 1001 is used to execute the hard disk replacement identification program stored in the memory 1005 to implement the steps of the hard disk replacement identification method described above.

[0098] The specific implementation of the hard drive replacement detection device in this application is basically the same as the embodiments of the hard drive replacement detection method described above, and will not be repeated here.

[0099] This application embodiment also provides a hard disk replacement detection device, which is applied to a hard disk replacement detection device, and the hard disk replacement detection device includes:

[0100] The extraction module is used to obtain the hard drive running time corresponding to the target hard drive, and extract the health feature value and retirement feature value of the hard drive running time;

[0101] The first construction module is used to construct a time-varying matrix of the health feature values ​​during the hard disk's operating time to obtain health feature data.

[0102] The second construction module is used to construct a time-varying matrix of the retirement feature values ​​during the hard disk's operating time to obtain retirement feature data.

[0103] The discrimination module is used to determine whether the target hard drive needs to be replaced based on the hard drive status prediction model, the health feature data, and the retirement feature data by predicting the health status and retirement status of the target hard drive.

[0104] Optionally, the hard disk status prediction model includes a health status prediction model and a retirement status prediction model, and the discrimination module is further used for:

[0105] By inputting the health feature data into the health status prediction model, the health status of the target hard drive is predicted, and the health status prediction result is obtained.

[0106] If the health status prediction result is unhealthy, then the target hard drive needs to be replaced.

[0107] If the health status prediction result is healthy, then by inputting the retirement feature data into the retirement status prediction model, the retirement status prediction of the target hard drive is performed to obtain the retirement status prediction result.

[0108] Based on the predicted retirement status, it is determined whether the target hard drive needs to be replaced.

[0109] Optionally, the hard disk status prediction model includes a health status prediction model and a retirement status prediction model, and the discrimination module is further used for:

[0110] By inputting the health feature data into the health status prediction model, the health status of the target hard drive is predicted, and the health status prediction result is obtained.

[0111] By inputting the decommissioning feature data into the decommissioning status prediction model, the decommissioning status of the target hard disk is predicted, and the decommissioning status prediction result is obtained.

[0112] The health status prediction results and the retirement status prediction results are aggregated to obtain the hard disk status prediction results;

[0113] Based on the hard drive status prediction results, it is determined whether the target hard drive needs to be replaced.

[0114] Optionally, the health status prediction model includes a convolutional neural network model and a recurrent neural network model, and the discrimination module is further used for:

[0115] The health feature data is input into the convolutional neural network model to predict the health status of the target hard disk based on the distribution of the health feature data during the hard disk's operating time, thereby obtaining a first health status prediction result.

[0116] The health feature data is input into the recurrent neural network model to predict the health status of the target hard drive based on the changing trend of the health features, and a second health status prediction result is obtained.

[0117] The first health status prediction result and the second health status prediction result are aggregated to obtain the health status prediction result.

[0118] Optionally, the health feature data includes time-series health feature data during the hard disk's operating time, the first health status prediction result includes a health probability, and the discrimination module is further configured to:

[0119] Based on the feature extractor in the convolutional neural network model, the time-series health feature data is mapped to feature extraction results;

[0120] Based on the classifier in the convolutional neural network model, the feature extraction results are converted into the health probability.

[0121] Optionally, the hard disk status prediction model includes a retirement status prediction model, and the hard disk replacement discrimination device is further used for:

[0122] Obtain the retirement status prediction model to be trained, and extract the hard drive temperature and hard drive business complexity of the target hard drive during the hard drive's operating time.

[0123] Construct a matrix showing the time-varying changes of the hard drive temperature and the hard drive business complexity during the hard drive's operating time to obtain training data for hard drive retirement characteristics;

[0124] Based on the hard drive retirement feature training data and the hard drive retirement label corresponding to the training target hard drive, the retirement status prediction model to be trained is iteratively trained and optimized to obtain the retirement status prediction model.

[0125] Optionally, the acquisition module further includes:

[0126] The retirement characteristic values ​​include at least one of hard drive temperature and hard drive business complexity.

[0127] The specific implementation of the hard disk replacement discrimination device in this application is basically the same as the embodiments of the hard disk replacement discrimination method described above, and will not be repeated here.

[0128] This application provides a readable storage medium that stores one or more programs, which can be executed by one or more processors to implement the steps of the hard disk replacement determination method described above.

[0129] The specific implementation of the readable storage medium in this application is basically the same as the embodiments of the above-described hard disk replacement discrimination method, and will not be described again here.

[0130] This application provides a computer program product, which includes one or more computer programs. The one or more computer programs can be executed by one or more processors to implement the steps of the hard disk replacement determination method described above.

[0131] The specific implementation of the computer program product in this application is basically the same as the embodiments of the above-mentioned hard disk replacement discrimination method, and will not be described again here.

[0132] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent scope of this application.

Claims

1. A method for determining hard drive replacement, characterized in that, The hard drive replacement detection method includes: Obtain the hard drive runtime corresponding to the target hard drive, and extract the health feature value and retirement feature value during the hard drive runtime; A matrix is ​​constructed to show the time-varying changes of the health feature values ​​during the hard disk's operating time, thereby obtaining health feature data; the health feature data is feature data that measures the health status of the target hard disk during the hard disk's operating time. A time-varying matrix of the retirement feature values ​​is constructed to obtain retirement feature data; the retirement feature data is feature data that measures the retirement status of the target hard disk during its operating time. Based on the hard drive status prediction model, the health feature data, and the retirement feature data, the target hard drive is evaluated for health status and retirement status to determine whether it needs to be replaced. The hard drive status prediction model includes a health status prediction model and a retirement status prediction model. The step of determining whether a target hard drive needs to be replaced by predicting its health status and retirement status based on the hard drive status prediction model, the health feature data, and the retirement feature data includes: By inputting the health feature data into the health status prediction model, the health status of the target hard drive is predicted, and the health status prediction result is obtained. If the health status prediction result is unhealthy, then the target hard drive needs to be replaced. If the health status prediction result is healthy, then by inputting the retirement feature data into the retirement status prediction model, the retirement status prediction of the target hard drive is performed to obtain the retirement status prediction result. Based on the predicted retirement status, it is determined whether the target hard drive needs to be replaced.

2. The hard drive replacement determination method as described in claim 1, characterized in that, The hard drive status prediction model includes a health status prediction model and a retirement status prediction model. The step of determining whether a target hard drive needs to be replaced by predicting its health status and retirement status based on the hard drive status prediction model, the health feature data, and the retirement feature data includes: By inputting the health feature data into the health status prediction model, the health status of the target hard drive is predicted, and the health status prediction result is obtained. By inputting the decommissioning feature data into the decommissioning status prediction model, the decommissioning status of the target hard disk is predicted, and the decommissioning status prediction result is obtained. The health status prediction results and the retirement status prediction results are aggregated to obtain the hard disk status prediction results; Based on the hard drive status prediction results, it is determined whether the target hard drive needs to be replaced.

3. The hard drive replacement determination method as described in claim 1 or 2, characterized in that, The health status prediction model includes a convolutional neural network model and a recurrent neural network model. The step of inputting the health feature data into the health status prediction model to predict the health status of the target hard drive and obtaining the health status prediction result includes: The health feature data is input into the convolutional neural network model to predict the health status of the target hard disk based on the distribution of the health feature data during the hard disk's operating time, thereby obtaining a first health status prediction result. The health feature data is input into the recurrent neural network model to predict the health status of the target hard drive based on the changing trend of the health features, and a second health status prediction result is obtained. The first health status prediction result and the second health status prediction result are aggregated to obtain the health status prediction result.

4. The hard drive replacement determination method as described in claim 3, characterized in that, The health feature data includes time-series health feature data during the hard disk's operating time, and the first health status prediction result includes a health probability. The step of inputting the health feature data into the convolutional neural network model to predict the health status of the target hard drive based on the distribution of the health feature data during the hard drive's operating time, and obtaining a first health status prediction result, includes: Based on the feature extractor in the convolutional neural network model, the time-series health feature data is mapped to feature extraction results; Based on the classifier in the convolutional neural network model, the feature extraction results are converted into the health probability.

5. The hard drive replacement determination method as described in claim 1, characterized in that, The retirement characteristic values ​​include at least one of hard drive temperature and hard drive business complexity.

6. The hard drive replacement determination method as described in claim 1, characterized in that, The hard drive status prediction model includes a retirement status prediction model. Before the step of determining whether the target hard drive needs to be replaced by predicting its health status and retirement status based on the hard drive status prediction model, the health feature data, and the retirement feature data, the hard drive replacement determination method further includes: Obtain the retirement status prediction model to be trained, and extract the hard drive temperature and hard drive business complexity of the target hard drive during the hard drive's operating time. Construct a matrix showing the time-varying changes of the hard drive temperature and the hard drive business complexity during the hard drive's operating time to obtain training data for hard drive retirement characteristics; Based on the hard drive retirement feature training data and the hard drive retirement label corresponding to the training target hard drive, the retirement status prediction model to be trained is iteratively trained and optimized to obtain the retirement status prediction model.

7. A hard drive replacement detection device, characterized in that, The hard drive replacement detection device includes: The extraction module is used to obtain the hard drive running time corresponding to the target hard drive, and extract the health feature value and retirement feature value of the hard drive running time; The first construction module is used to construct a matrix showing the change of the health feature values ​​over time during the hard disk's operating time, thereby obtaining health feature data; the health feature data is feature data that measures the health status of the target hard disk during the hard disk's operating time. The second construction module is used to construct a time-varying matrix of the retirement feature values ​​during the hard disk's operating time to obtain retirement feature data; the retirement feature data is feature data that measures the retirement status of the target hard disk during the hard disk's operating time. The discrimination module is used to determine whether the target hard drive needs to be replaced based on the hard drive status prediction model, the health feature data, and the retirement feature data by predicting the health status and retirement status of the target hard drive. The discrimination module is specifically used to predict the health status of the target hard drive by inputting the health feature data into a health status prediction model, and obtain a health status prediction result; if the health status prediction result is unhealthy, it is determined that the target hard drive needs to be replaced; if the health status prediction result is healthy, the target hard drive is predicted to be retired by inputting the retirement feature data into a retirement status prediction model, and obtain a retirement status prediction result; based on the retirement status prediction result, it is determined whether the target hard drive needs to be replaced; the hard drive status prediction model includes the health status prediction model and the retirement status prediction model.

8. A hard drive replacement detection device, characterized in that, The hard drive replacement detection device includes: a memory, a processor, and a program stored in the memory for implementing the hard drive replacement detection method. The memory is used to store the program that implements the hard disk replacement detection method; The processor is used to execute a program that implements the hard disk replacement determination method, so as to implement the steps of the hard disk replacement determination method as described in any one of claims 1 to 6.

9. A readable storage medium, characterized in that, The readable storage medium stores a program that implements a hard disk replacement detection method, which is executed by a processor to implement the steps of the hard disk replacement detection method as described in any one of claims 1 to 6.

Images