Disk fault prediction method, device and equipment and readable storage medium

By normalizing and multi-scale convolution of the time series data of disk attributes, combining channel and time attention mechanisms, the calibrated feature vectors are generated for fault prediction, which solves the problem of low prediction accuracy and improves disk failure Accuracy of prediction.

CN119065907APending Publication Date: 2024-12-03JINAN INSPUR DATA TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411376278.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-29
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

When predicting the traditional disk failure prediction model, its features cannot fully represent disk properties, resulting in low accuracy in failure prediction.

Method used

By normalizing the time series data of disk attributes and multi-scale convolution, the channel attention mechanism and time attention mechanism are used to determine the weights, and the calibrated multi-scale feature vectors are generated for fault prediction.

Benefits of technology

Improve the accuracy of the fault prediction model and enhance the ability to identify and predict disk health status.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119065907A_ABST
    Figure CN119065907A_ABST
Patent Text Reader

Abstract

The invention discloses a disk fault prediction method, device and equipment and a readable storage medium, and is applied to the technical field of fault processing, and the method comprises the steps: carrying out the normalization processing of time series data corresponding to each disk attribute, and obtaining the normalized time series data; performing multi-scale convolution on the normalized time sequence data corresponding to each magnetic disk attribute, determining weights corresponding to different magnetic disk attributes by using a channel attention mechanism, and obtaining a second feature vector based on the weights and the first feature vector corresponding to each magnetic disk attribute; determining the weight of the time point by using a time attention mechanism, and adjusting the second feature vector based on the weight of the time point to obtain a third feature vector; and performing prediction based on the third feature vector by using the prediction model to obtain a fault prediction result corresponding to the time sequence data of each disk attribute. According to the invention, the multi-scale third feature vector is used for fault prediction, so that the accuracy of the fault prediction model is improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the technical field of fault processing, and in particular to a disk fault prediction method, device, equipment and readable storage medium. Background Art

[0002] SMART (Disk Monitoring Data) technology is a hard disk self-monitoring, analysis and reporting technology that is widely used in early warning of disk failures. By continuously monitoring indicators such as the number of head errors, operating temperature, and the number of power switches, SMART technology aims to identify signs that may lead to failures in advance. Traditional prediction models use features that cannot fully represent the properties of the disk, resulting in low accuracy in fault prediction.

[0003] It can be seen that how to improve the accuracy of fault prediction is a technical problem that technical personnel in this field urgently need to solve. Summary of the invention

[0004] In view of this, an object of the present invention is to provide a disk fault prediction method, device, equipment and readable storage medium, which solves the technical problem of low fault prediction accuracy in the prior art.

[0005] In order to solve the above technical problems, the present invention provides a disk failure prediction method, comprising:

[0006] Normalize the time series data corresponding to each disk attribute to obtain the normalized time series data corresponding to each disk attribute;

[0007] Performing multi-scale convolution on the normalized time series data corresponding to each disk attribute to obtain a first feature vector corresponding to each disk attribute;

[0008] Determine weights corresponding to different disk attributes by using a channel attention mechanism, and obtain a second feature vector corresponding to each disk attribute based on the weights and the first feature vector corresponding to each disk attribute;

[0009] Determine the weight of the time point by using the time attention mechanism, and adjust the second feature vector corresponding to each disk attribute based on the weight of the time point to obtain a third feature vector;

[0010] A prediction model is used to perform prediction based on the third eigenvector to obtain a fault prediction result corresponding to the time series data of each disk attribute.

[0011] On the one hand, the normalization process is performed on the time series data corresponding to each disk attribute to obtain the normalized time series data corresponding to each disk attribute, including:

[0012] The time series data corresponding to each disk attribute is normalized using the minimum-maximum normalization method to obtain the normalized time series data corresponding to each disk attribute.

[0013] On the one hand, performing multi-scale convolution on the normalized time series data corresponding to each disk attribute to obtain a first feature vector corresponding to each disk attribute includes:

[0014] Define multiple convolution kernels of different sizes, each of which corresponds to a specific scale;

[0015] Applying each convolution kernel to the normalized time series data corresponding to each input disk attribute to obtain feature maps of different scales;

[0016] Feature maps of different scales are spliced ​​to obtain the first feature vector corresponding to each disk attribute.

[0017] On the one hand, the method of using the channel attention mechanism to determine the weights corresponding to different disk attributes, and obtaining the second feature vector corresponding to each disk attribute based on the weights and the first feature vector corresponding to each disk attribute, includes:

[0018] Performing pooling processing on the first feature vector corresponding to each disk attribute to obtain global information corresponding to each disk attribute;

[0019] Determining the weight of each disk attribute through a feature extraction operation based on the global information; wherein the feature extraction operation includes a convolution operation and an activation function processing;

[0020] The recalibrated feature vector corresponding to each disk attribute is obtained by multiplying the weight of each disk attribute by the first feature vector corresponding to each disk attribute;

[0021] The recalibrated feature vector corresponding to each disk attribute is residually connected with the first feature vector to obtain the second feature vector corresponding to each disk attribute.

[0022] On the one hand, the time attention mechanism is used to determine the weight of a time point, and the second feature vector corresponding to each disk attribute is adjusted based on the weight of the time point to obtain a third feature vector, including:

[0023] Using a convolutional layer to perform feature aggregation processing on the second feature vectors of all disk attributes to obtain a feature map of a time domain signal;

[0024] Processing the feature map of the time domain signal using an activation function to obtain the weight of the time point;

[0025] Determine the local time signal in the second feature vector corresponding to each disk attribute based on a clustered signal segment extraction method; wherein the clustered signal segment extraction method is at least one of a time window-based method, an event-triggered method, a data-driven method, and a statistical characteristic-based method;

[0026] Using a convolutional layer to encode feature information between local time signals in the second feature vector corresponding to each disk attribute, to obtain an adjusted second feature vector;

[0027] Multiplying the weight of the time point and the adjusted second eigenvector corresponding to each disk attribute to obtain a recalibrated adjusted second eigenvector corresponding to each disk attribute;

[0028] The recalibrated and adjusted second feature vector corresponding to each disk attribute and the corresponding second feature vector are added together to obtain the third feature vector corresponding to each disk attribute.

[0029] On the one hand, prediction is performed using a prediction model based on the third eigenvector to obtain a fault prediction result corresponding to the time series data of each disk attribute, including:

[0030] Performing global pooling processing on the third feature vector corresponding to each disk attribute by using global pooling in the prediction model to obtain pooling processing prediction data corresponding to each disk attribute;

[0031] Using the fully connected layer in the prediction model to process the pooled prediction data corresponding to each disk attribute, to obtain global feature prediction data corresponding to each disk attribute;

[0032] The global feature prediction data is processed using the activation function in the prediction model to determine the fault prediction result corresponding to each disk attribute.

[0033] On the one hand, prediction is performed using a prediction model based on the third eigenvector to obtain a fault prediction result corresponding to the time series data of each disk attribute, including:

[0034] Based on the third feature vector, the prediction model is used to perform prediction to obtain a classification label fault prediction result for each disk attribute; wherein, when the classification label is a first preset value, it is determined that the current disk attribute is in a healthy state; and when the classification label is a second preset value, it is determined that the current disk attribute is in a fault state.

[0035] The embodiment of the present invention further provides a disk failure prediction device, comprising:

[0036] A normalization processing module is used to perform normalization processing on the time series data corresponding to each disk attribute to obtain the normalized time series data corresponding to each disk attribute;

[0037] A multi-scale processing module, used for performing multi-scale convolution on the normalized time series data corresponding to each disk attribute to obtain a first feature vector corresponding to each disk attribute;

[0038] An attribute weight determination module, used to determine weights corresponding to different disk attributes by using a channel attention mechanism, and obtain a second feature vector corresponding to each disk attribute based on the weights and the first feature vector corresponding to each disk attribute;

[0039] A time point weight determination module, used to determine the weight of a time point by using a time attention mechanism, and to adjust the second feature vector corresponding to each disk attribute based on the weight of the time point to obtain a third feature vector;

[0040] The fault prediction module is used to perform prediction based on the third feature vector using a prediction model to obtain a fault prediction result corresponding to the time series data of each disk attribute.

[0041] The embodiment of the present invention also includes a disk failure prediction device, including:

[0042] Memory for storing computer programs;

[0043] A processor is used to execute the computer program to implement the steps of the disk failure prediction method as described above.

[0044] An embodiment of the present invention further provides a readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the disk failure prediction method described above are implemented.

[0045] The present invention also provides a computer program product, comprising a computer program / instruction, which implements the steps of the above disk failure prediction method when executed by a processor.

[0046] The purpose of the embodiments of the present invention is to provide a disk fault prediction method, device, equipment and readable storage medium, which can solve the technical problem of low fault prediction accuracy.

[0047] To solve the above technical problems, an embodiment of the present invention provides a disk failure prediction method, which may include: normalizing the time series data corresponding to each disk attribute to obtain the normalized time series data corresponding to each disk attribute; performing multi-scale convolution on the normalized time series data corresponding to each disk attribute to obtain the first feature vector corresponding to each disk attribute; using a channel attention mechanism to determine the weights corresponding to different disk attributes, and based on the weights and the first feature vector corresponding to each disk attribute, obtaining the second feature vector corresponding to each disk attribute; using a time attention mechanism to determine the weight of a time point, and adjusting the second feature vector corresponding to each disk attribute based on the weight of the time point to obtain a third feature vector; using a prediction model to perform prediction based on the third feature vector to obtain a failure prediction result corresponding to the time series data of each disk attribute.

[0048] It can be seen from the above technical scheme that the beneficial effects of the present invention are: the present invention uses multi-scale convolution to extract the features of the normalized time series data corresponding to each disk attribute, and obtains the multi-scale features (first feature vectors) corresponding to each disk attribute, thereby using the channel attention mechanism to determine the weights corresponding to different disk attributes, and based on the weights and the first feature vectors corresponding to each disk attribute, obtain the second feature vector corresponding to each disk attribute; use the time attention mechanism to determine the weights of the time points, adjust the second feature vectors corresponding to each disk attribute based on the weights of the time points, obtain the third feature vector, and obtain the multi-scale third feature vector used for fault prediction. Since the calibrated multi-scale features are used for fault prediction, the accuracy of the fault prediction model is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 A flowchart of a fault prediction method provided by an embodiment of the present invention;

[0051] Figure 2 A schematic diagram of a multi-scale convolution structure is provided for an embodiment of the present invention;

[0052] Figure 3 A structural diagram of a channel attention mechanism provided by an embodiment of the present invention;

[0053] Figure 4 A structural diagram of a temporal attention mechanism provided by an embodiment of the present invention;

[0054] Figure 5 An example flowchart of a disk failure prediction method provided by an embodiment of the present invention;

[0055] Figure 6 A structural framework diagram of a disk failure prediction method provided by an embodiment of the present invention;

[0056] Figure 7 A schematic diagram of the structure of a disk failure prediction device provided by an embodiment of the present invention;

[0057] Figure 8 A schematic diagram of the structure of a disk failure prediction device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] The terms "including" and "having" in the specification of the present invention and the above-mentioned drawings, as well as any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device including a series of steps or units is not limited to the listed steps or units, but may include steps or units that are not listed.

[0060] In order to enable those skilled in the art to better understand the solution of the present invention, the present invention is further described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0061] Next, a flowchart of a fault prediction method provided by an embodiment of the present invention is described in detail. Figure 1 A fault prediction method provided in an embodiment of the present invention may include:

[0062] S101, normalizing the time series data corresponding to each disk attribute to obtain the normalized time series data corresponding to each disk attribute.

[0063] The execution subject of the embodiment of the present invention is an electronic device, which can be a computer, a mobile phone, etc. The embodiment does not limit the specific disk attributes, which refer to attributes related to various health indicators and operating status of the hard disk. For example, the disk attribute can be at least two of the read error rate, the number of motor start-up retries, the number of power on / off times, and the relocation magnetic zone count; or the disk attribute in this embodiment can be at least two of the disk startup time, the number of motor start-stop times, and the seek error rate. The time series data in this embodiment can be SMART (Self-Monitoring, Analysis and Reporting Technology, disk monitoring data). The embodiment does not limit the specific number of disk attributes. For example, the number of disk attributes in this embodiment can be 10; the number of disks in this embodiment can also be 16. The time series data in this embodiment is data composed of data at different time points. The embodiment does not limit the specific method of normalization. For example, the normalization method in this embodiment can be one-hot encoding. For categorical variables, one-hot encoding is a common normalization method that converts each category into a binary vector. For example, if there are n categories, each category will be converted into a vector of length n, in which only one position is 1 and the rest are 0. Alternatively, the normalization method in this embodiment can also be min-max normalization, which linearly transforms the minimum and maximum values ​​of each feature. It is understandable that since the value ranges of different attributes of different disks and suppliers vary greatly, and different attributes represent different meanings, it is necessary to normalize the time series data of each disk attribute for preprocessing. When the normalization method is min-max (minimum and maximum normalization), first, use the formula Time series data for each disk attribute , perform min-max normalization and scale to [-1,1], is the normalized data, Indicates the data value corresponding to any disk attribute at the i-th time point.

[0064] It should be further explained that, in order to improve the efficiency of normalization, the above-mentioned normalization processing of the time series data corresponding to each disk attribute to obtain the normalized time series data corresponding to each disk attribute may include: using the minimum and maximum normalization method to normalize the time series data corresponding to each disk attribute to obtain the normalized time series data corresponding to each disk attribute. The minimum and maximum normalization in this embodiment is an effective data preprocessing technology, which scales the data features into a uniform range through a simple linear transformation, thereby improving the efficiency of normalization.

[0065] It should be further explained that, in order to improve the accuracy and efficiency of disk failure prediction, before normalizing the time series data corresponding to each disk attribute to obtain the normalized time series data corresponding to each disk attribute, it may also include: analyzing each disk attribute to determine the accurate value of each disk attribute for failure prediction; based on the accurate value of each disk attribute, selecting a preset number of disk attributes as the disk attributes used for the final prediction. This embodiment does not limit the specific number of the preset number. For example, the preset number in this embodiment may be 10; or the preset number in this embodiment may also be 15. This embodiment selects a preset number of disk attributes as the disk attributes used for the final prediction based on the accurate value of each disk attribute, which means that the accurate values ​​are arranged from high to low, and the preset number of disk attributes before the accurate value are selected as the disk attributes used for the final prediction. For example, the disk attributes used in the final prediction may include Read Error Rate, Power-On Hours, Spin Retry Count, Spin-Up Time, Power Cycle Count, Start / Stop Count, Real Allocated Sectors Count, Seek Error Rate, Power-off Retract Count, Load / Unload Cycle Count, Temperature, Current Pending Sector Count, Uncorrectable Sector Count, and UltraDMA CRC Error Count (CRC error occurs during data transmission). It is understandable that SMART (disk monitoring data) is data recorded continuously over time and can be regarded as time series data. The SMART system continuously monitors the health indicators and operating status of the hard disk, and records the values ​​of these attributes at fixed time intervals (such as every day, every hour, or every time the disk is started). These attributes include but are not limited to disk temperature, power-on time, start / stop count, seek errors, number of remapped sectors, CRC errors, etc., which reflect the state changes of the disk during use. Therefore, when analyzing and predicting disk failures, SMART attributes can be treated and analyzed as time series data.In actual applications, there are many SMART attributes (more than 50), and if all are used, the disk failure prediction efficiency will be low. This embodiment can identify which attributes are most critical for failure prediction by analyzing the predicted accuracy of each disk attribute. Selecting these key attributes as the basis for the final prediction helps to improve the overall prediction accuracy of the model. This method can ensure that the model focuses on the most valuable information, reduce noise interference, and improve prediction efficiency due to fewer disk attributes.

[0066] S102, performing multi-scale convolution on the normalized time series data corresponding to each disk attribute to obtain a first feature vector corresponding to each disk attribute.

[0067] This embodiment does not limit the specific scale in the multi-scale convolution. For example, the scale in this embodiment can be 3 different scales; or the scale in this embodiment can also be 5 different scales. For ease of understanding, for the normalized time series data, , is a collection of time series data with specific attributes, where each It's at the time Normalized value. In order to extract multi-scale features from these data, a parallel multi-scale one-dimensional convolutional neural network can be used. In this network, different convolutional layers use convolution kernels of different sizes, such as , ,and , corresponding to short, medium and long time windows, respectively, to capture features at different time scales. The specific calculation is given by the following formula: , and specific scales , ,in, is a convolution kernel of a specific scale Middle The weight of the position, It's at the time forward Normalized time series data of unit time. In this way, each convolution layer will generate a feature map (or feature vector), which reflects the trend of time series data changes within a certain time window. After completing the convolution operation of all scales, the feature maps of different scales are spliced ​​to form a new feature vector. ; , this multi-channel (one channel for each disk attribute) feature vector contains the characteristic performance of time series data at multiple time scales, which helps the model to more comprehensively understand and predict the health status of the disk. Figure 2 As shown, Figure 2A schematic diagram of a multi-scale convolution structure is provided for an embodiment of the present invention, wherein Conv represents convolution, ReLU (Rectified Linear Unit) is a nonlinear activation function, Max-Pool represents maximum pooling, and Concatenation represents concatenation. For example, the power-on time after min-max processing is used as input data, and three convolution kernels of different sizes, k1, k2, and k3, are used to obtain features of different scales. ReLU (Rectified Linear Unit) is an activation function often used in combination with CNN (Convolutional Neural Network), which has the advantages of fast convergence and small amount of calculation. The calculation method is as shown in the formula , where x represents the normalized data value, as shown. Max Pooling is a downsampling technique used to reduce the spatial size of the feature map, while reducing the amount of calculation and the number of parameters while retaining the most important feature information. After all processing is completed, the features of different scales are concatenated to form a new power-on time feature map (the first feature vector).

[0068] It should be further explained that the multi-scale convolution is performed on the normalized time series data corresponding to each disk attribute to obtain the first feature vector corresponding to each disk attribute, which may include:

[0069] S1021, define multiple convolution kernels of different sizes, each convolution kernel corresponds to a specific scale;

[0070] S1022, applying each convolution kernel to the normalized time series data corresponding to each input disk attribute to obtain feature maps of different scales;

[0071] S1023, concatenating feature maps of different scales to obtain a first feature vector corresponding to each disk attribute.

[0072] The size of the convolution kernel in this embodiment can be 3 different convolution kernels, or the size of the convolution kernel in this embodiment can be 6 different convolution kernels. By using convolution kernels of different sizes, this embodiment enables the model to extract features from different time scales, which helps to capture richer temporal dynamic information; multi-scale feature map splicing provides a more comprehensive first feature vector for each disk attribute, which helps to improve the generalization ability of the model on new data. By considering information at different scales, the model can better adapt to various situations and reduce the risk of overfitting; by splicing feature maps of different scales, a richer and more complex feature representation can be obtained. This method effectively combines information from different time ranges, allowing the model to more accurately distinguish and understand patterns in the data.

[0073] S103, using the channel attention mechanism to determine the weights corresponding to different disk attributes, and based on the weights and the first feature vector corresponding to each disk attribute, obtain a second feature vector corresponding to each disk attribute.

[0074] The channel attention module (CAM) in this embodiment is a technology for enhancing the channel feature representation in a convolutional neural network. This embodiment does not limit the specific method for determining the weights corresponding to different disk attributes based on the channel attention mechanism. For example, this embodiment can determine the weights corresponding to different disk attributes based on the time series data corresponding to each disk attribute using the channel attention mechanism; or this embodiment can also determine the weights corresponding to different disk attributes based on the normalized time series data corresponding to each disk attribute using the channel attention mechanism; or this embodiment can also determine the weights corresponding to different disk attributes based on the first feature vector corresponding to each disk attribute using the channel attention mechanism. This embodiment does not limit the steps of S103 and S104, and S103 can be performed first and then S104. Or this embodiment can also perform S104 first and then S103, at this time, the time attention mechanism is used to determine the weight of the time point, and the first feature vector corresponding to each disk attribute is adjusted based on the weight of the time point to obtain the time feature vector, and then the channel attention mechanism is used to determine the weights corresponding to different disk attributes, and the time and attribute feature vectors corresponding to each disk attribute are obtained based on the weight and the time feature vector corresponding to each disk attribute. It is understandable that different disk attributes carry different information about the health status of the disk. Since different attributes have different indicative effects on disk failure, it is particularly important to identify and highlight those attributes with greater diagnostic value. In order to identify the importance of different disk attributes to disk health, the present invention introduces a channel attention module (CAM). Different weights can be given to different attributes according to their importance in disk health identification. CAM compresses the data of each disk attribute (which can be the initial time series data, the normalized time series data, or the data in the first eigenvector) into a single value through a global average pooling operation to represent the overall importance of the attribute. Subsequently, a series of convolution operations and Sigmoid activation functions are used to generate the weight of each channel. The size of the weight reflects the relative importance of the corresponding attribute in predicting disk failure. Attributes with higher weights mean that they occupy a more important position in fault prediction. CAM improves the network's sensitivity to different features by explicitly modeling the interdependencies between different channels, thereby adaptively distinguishing the importance of features and dynamically adjusting the contribution of each SMART attribute, so that the model can focus more on those attributes that are closely related to disk failures, thereby improving the accuracy of fault prediction and the interpretability of the model. For example, if the temperature rises abnormally over a period of time, the weight of the temperature attribute will be assigned a higher value, and the model will pay more attention to the role of this attribute in fault prediction. The basic structure of CAM is as follows: Figure 3 As shown, Figure 3 A channel attention mechanism structure diagram provided in an embodiment of the present invention. For a multi-channel input (Each of which represents the characteristics of a channel), Defines how to calculate each channel The global average . W is the width of the feature map, Indicates channel In Location The global average pooling operation converts the feature The response values ​​of all positions are averaged to obtain a single value To represent the channel This operation embeds the global temporal information into CAM then uses a simple gating mechanism to fully capture the channel-wise dependencies and generate the channel recalibration vector , .in is the ReLU activation function, , The number of channels is , the convolution kernel size is The convolution of is a Sigmoid function that compresses the dynamic range of the input activation vector into the range of [0, 1]. The value of represents the The channel recalibration vector is used to convert the feature Recalibration is done as a channel recalibration vector M, which is used to recalibrate the feature Y, as shown in the formula: Finally, the obtained feature M fully considers the guidance of global information and can effectively highlight more discriminative feature information. Then, The range is [0,1]. In order to avoid the reduction of deep feature response values ​​and the impact on recognition performance caused by repeated feature recalibration operations, the idea of ​​residual learning is used and residual connections are introduced to improve the feasibility of optimization while retaining the original information. The final output of CAM is: (That is, the recalibrated feature vector is added to the output feature vector). The introduction of CAM enhances the model's ability to capture key fault features, allowing the originally complex SMART data features to be optimized. By multiplying the weights with the original features, the recalibrated features can be more effectively used in the disk failure prediction model to improve prediction performance. In addition, through the residual connection, the original feature information is retained, ensuring the stability of model training and the integrity of the features. When processing SMART data, the channel attention mechanism can automatically identify and amplify those attributes that are closely related to disk failures, thereby improving the accuracy and reliability of the fault prediction model.

[0075] It should be further explained that, in order to improve the stability of the features, the weights corresponding to different disk attributes are determined by using the channel attention mechanism, and based on the weights and the first feature vector corresponding to each disk attribute, the second feature vector corresponding to each disk attribute is obtained, which may include:

[0076] S1031, performing pooling processing on the first feature vector corresponding to each disk attribute to obtain global information corresponding to each disk attribute;

[0077] S1032, determining the weight of each disk attribute through a feature extraction operation based on the global information; wherein the feature extraction operation includes a convolution operation and an activation function processing;

[0078] S1033, multiplying the weight of each disk attribute by the first feature vector corresponding to each disk attribute to obtain a recalibrated feature vector corresponding to each disk attribute;

[0079] S1034, performing residual connection on the recalibrated feature vector corresponding to each disk attribute and the first feature vector to obtain a second feature vector corresponding to each disk attribute.

[0080] This embodiment can use the channel attention mechanism to automatically adjust the weight of each disk attribute according to the degree of influence of the disk attribute on the disk health status, so that the model focuses more on those attributes that have a significant indicative effect on fault prediction, thereby improving the accuracy of the prediction. In addition, through the residual connection, the original feature information is retained, ensuring the stability of the model training and the integrity of the features.

[0081] S104, using the time attention mechanism to determine the weight of the time point, and adjusting the second feature vector corresponding to each disk attribute based on the weight of the time point to obtain a third feature vector.

[0082] The Time Attention Mechanism (TAM) in this embodiment is used to identify and enhance attributes that are critical to fault prediction. This embodiment does not limit the specific method of determining the weights corresponding to different disk attributes based on the time attention mechanism. For example, this embodiment can use the time attention mechanism to determine the weights corresponding to different disk attributes based on the time series data corresponding to each disk attribute; or this embodiment can also use the time attention mechanism to determine the weights corresponding to different disk attributes based on the normalized time series data corresponding to each disk attribute; or this embodiment can also use the time attention mechanism to determine the weights corresponding to different disk attributes based on the first feature vector corresponding to each disk attribute. It can be understood that the time series data contains multiple attributes such as read error rate, temperature, power-on time, etc. These attributes change over time to form a set of time series data. The goal of TAM is to locate the time signal segment related to disk failure by aggregating all excitation mapping feature information across channels. The calculation process is as follows: Assume that the input feature is represented by Y ,in , Corresponding to the jth time signal position, First, TAM has a channel The convolution layer obtains the feature map s of the feature Y in the time domain signal as shown in the formula , The convolution operation aggregates the features of all channels, and then obtains the temporal weight vector through Sigmoid (activation function). ,like As shown, , Indicates the importance of the Jth time series point, the time weight vector It is used to multiply the time weight vector and the input feature to obtain the recalibrated feature map N, as shown in the formula Before recalibrating Y, TAM uses a convolutional layer The feature information between local time signal segments is encoded to prevent excessive focus on related time signal segments. Similar to CAM, TAM also introduces residual connections. The output of TAM is: Through the above steps, TAM dynamically adjusts the contribution of different time signal segments, so that the model can focus more on those time periods that are closely related to disk failures. By assigning different weights to data at different time points, the model can more accurately capture the pattern changes before the failure, and can help the model identify which SMART indicators are more important for failure prediction at a specific time point, and more accurately locate the key information that may cause the failure, thereby improving the accuracy of the prediction. Figure 4A structure diagram of a temporal attention mechanism provided for an embodiment of the present invention, wherein CNN represents a convolutional neural network, a feature map of CxW is composed of time series data of C disk attributes; W on Sigmoid represents the weight of each time point, and the weight of each time point is multiplied (recalibrated) with a multi-channel second feature vector composed of multiple second feature vectors to obtain a calibrated multi-channel second feature vector, and the calibrated multi-channel second feature vector is concatenated with the multi-channel second feature vector to obtain a third feature vector.

[0083] It should be further explained that, in order to improve the accuracy of determining the third eigenvector, the above-mentioned use of the time attention mechanism to determine the weight of the time point, and adjusting the second eigenvector corresponding to each disk attribute based on the weight of the time point to obtain the third eigenvector can include: using a convolution layer to perform feature aggregation processing on the second eigenvectors of all disk attributes to obtain a feature map of the time domain signal; using an activation function to process the feature map of the time domain signal to obtain the weight of the time point; determining the local time signal in the second eigenvector corresponding to each disk attribute based on an aggregated signal segment extraction method; wherein the aggregated signal segment extraction method is at least one of a time window-based method, an event-triggered method, a data-driven method, and a statistical characteristic-based method; using a convolution layer to encode the feature information between the local time signals in the second eigenvector corresponding to each disk attribute to obtain an adjusted second eigenvector; multiplying the weight of the time point and the adjusted second eigenvector corresponding to each disk attribute to obtain a recalibrated adjusted second eigenvector corresponding to each disk attribute; adding the recalibrated adjusted second eigenvector corresponding to each disk attribute and the corresponding second eigenvector to obtain the third eigenvector corresponding to each disk attribute. The method for extracting the aggregated signal segments in this embodiment can be at least one of an event-triggered method, a data-driven method, and a statistical characteristic-based method; or the method for extracting the aggregated signal segments in this embodiment can also be a summary result based on several methods. This embodiment uses the temporal attention mechanism to enable the model to focus on the time period closely related to the disk failure, so that the model can more accurately capture the pattern changes before the failure and improve the accuracy of the prediction.

[0084] S105 , performing prediction using a prediction model based on the third eigenvector to obtain a fault prediction result corresponding to the time series data of each disk attribute.

[0085] This embodiment does not limit the specific prediction model. For example, the prediction model in this embodiment may be a neural network; or the prediction model in the embodiment may be a decision tree; or the prediction model in the embodiment may also be a support vector machine. This embodiment does not limit the specific fault prediction result. For example, the fault prediction result in this embodiment may be normal, or the fault prediction result in this embodiment may also be a fault.

[0086] It should be further explained that the above prediction model is used to perform prediction based on the third eigenvector to obtain the fault prediction results corresponding to the time series data of each disk attribute, which may include:

[0087] S1051, performing global pooling processing on the third feature vector corresponding to each disk attribute by using global pooling in the prediction model to obtain pooling processing prediction data corresponding to each disk attribute;

[0088] S1052, using the fully connected layer in the prediction model to process the pooled prediction data corresponding to each disk attribute to obtain global feature prediction data corresponding to each disk attribute;

[0089] S1053, using the activation function in the prediction model to process the global feature prediction data, and determine the fault prediction result corresponding to each disk attribute.

[0090] The prediction model in this embodiment is a neural network model including global pooling, a fully connected layer and an activation function, or the neural network model may also include a convolution layer. That is, based on the third eigenvector, global pooling, a convolution layer, a fully connected layer and Softmax are used to perform prediction to obtain the fault prediction result of each type of disk attribute.

[0091] It should be further explained that the above-mentioned prediction based on the third eigenvector using the prediction model to obtain the fault prediction result corresponding to the time series data of each disk attribute can include: predicting based on the third eigenvector using the prediction model to obtain the classification label fault prediction result of each disk attribute; wherein the classification label is the first preset value, and the current disk attribute is determined to be in a healthy state; the classification label is the second preset value, and the current disk attribute is determined to be in a fault state. represents a data set, where represents the input features of each observation in the dataset, and is the corresponding classification label, which is used to indicate the status of the input data. Specifically, when When In good health or normal working condition; otherwise, , it means that the device or system corresponding to the input data is in a faulty or abnormal state. In this embodiment, the use of classification labels can simplify complex data into easy-to-understand categories, so that decision makers can quickly identify key issues and trends.

[0092] A disk failure prediction method provided by an embodiment of the present invention may include: S101, normalizing the time series data corresponding to each disk attribute to obtain the normalized time series data corresponding to each disk attribute; S102, performing multi-scale convolution on the normalized time series data corresponding to each disk attribute to obtain the first feature vector corresponding to each disk attribute; S103, using a channel attention mechanism to determine the weights corresponding to different disk attributes, and based on the weights and the first feature vector corresponding to each disk attribute, obtaining the second feature vector corresponding to each disk attribute; S104, using a time attention mechanism to determine the weight of a time point, and adjusting the second feature vector corresponding to each disk attribute based on the weight of the time point to obtain a third feature vector; S105, using a prediction model to perform prediction based on the third feature vector to obtain a failure prediction result corresponding to the time series data of each disk attribute. The present invention utilizes multi-scale convolution to extract the features of normalized time series data corresponding to each disk attribute, and obtains the multi-scale features (first feature vector) corresponding to each disk attribute, thereby utilizing the channel attention mechanism to determine the weights corresponding to different disk attributes, and based on the weights and the first feature vector corresponding to each disk attribute, obtains the second feature vector corresponding to each disk attribute; utilizes the time attention mechanism to determine the weight of the time point, and adjusts the second feature vector corresponding to each disk attribute based on the weight of the time point to obtain the third feature vector, and obtains the multi-scale third feature vector used for fault prediction. Since the calibrated multi-scale features are used for the final fault prediction, the accuracy of the fault prediction model is improved.

[0093] In order to make the present invention easier to understand, please refer to Figure 5 , Figure 5 An example flow chart of a disk failure prediction method provided by an embodiment of the present invention may specifically include:

[0094] S201, obtaining disk monitoring time series data, and obtaining time series data corresponding to various disk attributes according to the disk monitoring time series data.

[0095] S202, preprocessing different disk attributes using a maximum-minimum-maximum normalization method to obtain normalized time series data corresponding to various disk attributes.

[0096] S203, using a parallel multi-scale one-dimensional convolutional neural network to determine the features on different time scales corresponding to the normalized time series data, and concatenate the features on different time scales to obtain a first feature vector corresponding to each disk attribute.

[0097] In this embodiment, each disk attribute corresponds to a channel.

[0098] S204, compressing the time series data of each disk attribute into a single value through a global average pooling operation, and processing the single value using a convolution operation and an activation function to obtain a weight for each disk attribute; the size of the weight reflects the relative importance of the corresponding disk attribute in predicting disk failures.

[0099] S205 , multiplying the weight of each disk attribute by the first feature vectors corresponding to the various disk attributes, and adding the multiplication result to the first feature vectors corresponding to the various disk attributes to obtain a second feature vector.

[0100] S206, using the time attention mechanism to determine the weight of each time point in the second eigenvector, and adding the result obtained by multiplying the weight of the time point and the second eigenvector to the second eigenvector to obtain a third eigenvector.

[0101] S207, performing fault prediction using a neural network prediction model based on the third eigenvector to obtain a corresponding fault prediction result for each disk attribute.

[0102] The overall framework diagram corresponding to this embodiment is as follows Figure 6 As shown, Figure 6 The structural framework diagram of a disk fault prediction method provided by an embodiment of the present invention. As can be seen from the figure, the overall process is to regard different SMART attribute data as different channels, and use multi-scale convolution for each channel to extract multi-scale features from the normalized SMART data. Different attribute fragments are assigned different weights based on the input (multi-scale features) using TAM, and then the channel recalibration vector generated by CAM is used to recalibrate the features of different SMART attributes. The prediction results are finally obtained through global pooling, convolution layer, fully connected layer and Softmax. The prediction results in this embodiment may include faults; fault categories, disk attribute failure probabilities and fault time points.

[0103] The disk failure prediction device provided by an embodiment of the present invention is introduced below. The disk failure prediction device described below and the disk failure prediction method described above can be referred to each other.

[0104] Figure 7 A schematic diagram of the structure of a disk failure prediction device provided by an embodiment of the present invention may include:

[0105] The normalization processing module 100 is used to perform normalization processing on the time series data corresponding to each disk attribute to obtain the normalized time series data corresponding to each disk attribute;

[0106] The multi-scale processing module 200 is used to perform multi-scale convolution on the normalized time series data corresponding to each disk attribute to obtain a first feature vector corresponding to each disk attribute;

[0107] The attribute weight determination module 300 is used to determine the weights corresponding to different disk attributes by using a channel attention mechanism, and obtain a second feature vector corresponding to each disk attribute based on the weights and the first feature vector corresponding to each disk attribute;

[0108] A time point weight determination module 400 is used to determine the weight of a time point by using a time attention mechanism, and to adjust the second feature vector corresponding to each disk attribute based on the weight of the time point to obtain a third feature vector;

[0109] The fault prediction module 500 is used to perform prediction using a prediction model based on the third feature vector to obtain a fault prediction result corresponding to the time series data of each disk attribute.

[0110] Further, based on the above embodiment, the normalization processing module 100 may include:

[0111] The minimum-maximum normalization unit is used to normalize the time series data corresponding to each disk attribute by using the minimum-maximum normalization method to obtain the normalized time series data corresponding to each disk attribute.

[0112] Further, based on any of the above embodiments, the multi-scale processing module 200 may include:

[0113] A convolution kernel determination unit is used to define multiple convolution kernels of different sizes, each of which corresponds to a specific scale;

[0114] A feature map determining unit, used for applying each convolution kernel to the normalized time series data corresponding to each input disk attribute to obtain feature maps of different scales;

[0115] The splicing unit is used to splice feature maps of different scales to obtain the first feature vector corresponding to each disk attribute.

[0116] Further, based on any of the above embodiments, the attribute weight determination module 300 may include:

[0117] A pooling unit, used for performing pooling processing on the first feature vector corresponding to each disk attribute to obtain global information corresponding to each disk attribute;

[0118] A disk attribute weight determination unit, used to determine the weight of each disk attribute through a feature extraction operation based on the global information; wherein the feature extraction operation includes a convolution operation and an activation function processing;

[0119] A recalibration unit, configured to obtain a recalibrated feature vector corresponding to each disk attribute by multiplying the first feature vector corresponding to each disk attribute by the weight of each disk attribute;

[0120] The residual connection unit is used to perform residual connection on the recalibrated feature vector corresponding to each disk attribute and the first feature vector to obtain the second feature vector corresponding to each disk attribute.

[0121] Further, based on any of the above embodiments, the time point weight determination module 400 may include:

[0122] A time domain signal feature map determining unit, configured to perform feature aggregation processing on the second feature vectors of all disk attributes using a convolutional layer to obtain a feature map of the time domain signal;

[0123] A time point weight determination unit, used for processing the feature map of the time domain signal using an activation function to obtain the weight of the time point;

[0124] A local time signal determination unit, used to determine the local time signal in the second feature vector corresponding to each disk attribute based on an aggregated signal segment extraction method; wherein the aggregated signal segment extraction method is at least one of a time window-based method, an event-triggered method, a data-driven method, and a statistical characteristic-based method;

[0125] Based on the time signal adjustment unit, it is used to encode the characteristic information between the local time signals in the second feature vector corresponding to each disk attribute by using a convolution layer to obtain an adjusted second feature vector;

[0126] A time point-based recalibration unit, configured to multiply the weight of the time point by the adjusted second feature vector corresponding to each disk attribute to obtain a recalibrated adjusted second feature vector corresponding to each disk attribute;

[0127] The third feature vector determination unit is used to add the recalibrated and adjusted second feature vector corresponding to each disk attribute and the corresponding second feature vector to obtain the third feature vector corresponding to each disk attribute.

[0128] Further, based on the above embodiment, the above fault prediction module 500 may include:

[0129] A pooling processing prediction data determination unit, used to perform global pooling processing on the third feature vector corresponding to each disk attribute by using global pooling in the prediction model to obtain pooling processing prediction data corresponding to each disk attribute;

[0130] A global feature prediction data determination unit, configured to process the pooled prediction data corresponding to each disk attribute using a fully connected layer in the prediction model to obtain global feature prediction data corresponding to each disk attribute;

[0131] The fault prediction result determination unit is used to process the global feature prediction data using the activation function in the prediction model to determine the fault prediction result corresponding to each disk attribute.

[0132] Further, based on any of the above embodiments, the above fault prediction module 500 may include:

[0133] A classification label determination unit is used to perform prediction based on the third feature vector using the prediction model to obtain a classification label fault prediction result for each disk attribute; wherein, when the classification label is a first preset value, the current disk attribute is determined to be in a healthy state; and when the classification label is a second preset value, the current disk attribute is determined to be in a fault state.

[0134] It should be noted that the order of the modules and units in the above disk failure prediction device can be changed without affecting the logic.

[0135] Figure 7 The description of the features in the corresponding embodiments can be found in Figure 7 The relevant descriptions of the corresponding embodiments will not be repeated here one by one.

[0136] The disk fault prediction device provided by the embodiment of the present invention may include: a normalization processing module 100, which is used to perform normalization processing on the time series data corresponding to each disk attribute to obtain the normalized time series data corresponding to each disk attribute; a multi-scale processing module 200, which is used to perform multi-scale convolution on the normalized time series data corresponding to each disk attribute to obtain a first feature vector corresponding to each disk attribute; an attribute weight determination module 300, which is used to determine the weights corresponding to different disk attributes using a channel attention mechanism, and obtain a second feature vector corresponding to each disk attribute based on the weights and the first feature vector corresponding to each disk attribute; a time point weight determination module 400, which is used to determine the weight of a time point using a time attention mechanism, and adjust the second feature vector corresponding to each disk attribute based on the weight of the time point to obtain a third feature vector; a fault prediction module 500, which is used to perform prediction using a prediction model based on the third feature vector to obtain a fault prediction result corresponding to the time series data of each disk attribute. The present invention utilizes multi-scale convolution to extract the features of normalized time series data corresponding to each disk attribute, and obtains the multi-scale features (first feature vector) corresponding to each disk attribute, thereby utilizing the channel attention mechanism to determine the weights corresponding to different disk attributes, and based on the weights and the first feature vector corresponding to each disk attribute, obtains the second feature vector corresponding to each disk attribute; utilizes the time attention mechanism to determine the weight of the time point, and adjusts the second feature vector corresponding to each disk attribute based on the weight of the time point to obtain the third feature vector, and obtains the multi-scale third feature vector used for fault prediction. Since the calibrated multi-scale features are used for the final fault prediction, the accuracy of the fault prediction model is improved.

[0137] A disk failure prediction device provided by an embodiment of the present invention is introduced below. The disk failure prediction device described below and the disk failure prediction method described above can be referred to each other.

[0138] Figure 8 A schematic diagram of the structure of a disk failure prediction device provided by an embodiment of the present invention is shown in FIG. Figure 8 As shown, the disk failure prediction device includes: a memory 60 for storing a computer program;

[0139] The processor 61 is used to implement the steps of the disk failure prediction method in the above embodiment when executing a computer program.

[0140] The disk failure prediction device provided in this embodiment may include but is not limited to a smart phone, a tablet computer, a laptop computer, or a desktop computer.

[0141] Among them, the processor 61 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 61 can be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), and programmable logic array (PLA). The processor 61 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a central processing unit (CPU); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 61 may be integrated with a graphics processing unit (GPU), and the GPU is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 61 may also include an artificial intelligence (AI) processor, which is used to process computing operations related to machine learning.

[0142] The memory 60 may include one or more computer-readable storage media, which may be non-transitory. The memory 60 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In this embodiment, the memory 60 is at least used to store the following computer program 601, wherein, after the computer program is loaded and executed by the processor 61, it can implement the relevant steps of the disk failure prediction method disclosed in any of the aforementioned embodiments. In addition, the resources stored in the memory 60 may also include an operating system 602 and data 603, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system 602 may include Windows, Unix, Linux, etc. The data 603 may include but is not limited to disk failure prediction data, etc.

[0143] In some embodiments, the disk failure prediction device may further include a display screen 62 , an input and output interface 63 , a communication interface 64 , a power supply 65 , and a communication bus 66 .

[0144] Those skilled in the art will understand that Figure 8 The structure shown in the figure does not constitute a limitation to the disk failure prediction device, and may include more or less components than those shown in the figure.

[0145] It is understandable that if the disk failure prediction method in the above embodiment is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the current technology or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and executes all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), electrically erasable programmable ROM, register, hard disk, removable disk, CD-ROM, magnetic disk or optical disk and other media that can store program codes.

[0146] Based on this, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the disk failure prediction method described above are implemented.

[0147] Based on this, an embodiment of the present invention further provides a computer program product, including a computer program / instruction, which implements the steps of the above disk failure prediction method when executed by a processor.

[0148] The above is a detailed introduction to a disk failure prediction method, device, equipment and readable storage medium provided by an embodiment of the present invention. The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0149] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0150] The above is a detailed introduction to a disk failure prediction method, device, equipment and readable storage medium provided by the present invention. This article uses specific examples to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method and core ideas of the present invention. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, the present invention can also be improved and modified, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

Claims

1. A disk failure prediction method, It is characterized in that include: Normalize the time series data corresponding to each disk attribute to obtain the normalized time series data corresponding to each disk attribute; Performing multi-scale convolution on the normalized time series data corresponding to each disk attribute to obtain a first feature vector corresponding to each disk attribute; Determine weights corresponding to different disk attributes by using a channel attention mechanism, and obtain a second feature vector corresponding to each disk attribute based on the weights and the first feature vector corresponding to each disk attribute; Determine the weight of the time point by using the time attention mechanism, and adjust the second feature vector corresponding to each disk attribute based on the weight of the time point to obtain a third feature vector; A prediction model is used to perform prediction based on the third eigenvector to obtain a fault prediction result corresponding to the time series data of each disk attribute.

2. The disk failure prediction method according to claim 1, It is characterized in that The normalizing process is performed on the time series data corresponding to each disk attribute to obtain the normalized time series data corresponding to each disk attribute, including: The time series data corresponding to each disk attribute is normalized using the minimum-maximum normalization method to obtain the normalized time series data corresponding to each disk attribute.

3. The disk failure prediction method according to claim 1, It is characterized in that The multi-scale convolution is performed on the normalized time series data corresponding to each disk attribute to obtain a first feature vector corresponding to each disk attribute, including: Define multiple convolution kernels of different sizes, each of which corresponds to a specific scale; Applying each convolution kernel to the normalized time series data corresponding to each input disk attribute to obtain feature maps of different scales; Feature maps of different scales are spliced ​​to obtain the first feature vector corresponding to each disk attribute.

4. The disk failure prediction method according to any one of claims 1 to 3, It is characterized in that The method of using the channel attention mechanism to determine weights corresponding to different disk attributes, and obtaining a second feature vector corresponding to each disk attribute based on the weights and the first feature vector corresponding to each disk attribute, includes: Performing pooling processing on the first feature vector corresponding to each disk attribute to obtain global information corresponding to each disk attribute; Determining the weight of each disk attribute through a feature extraction operation based on the global information; wherein the feature extraction operation includes a convolution operation and an activation function processing; The recalibrated feature vector corresponding to each disk attribute is obtained by multiplying the weight of each disk attribute by the first feature vector corresponding to each disk attribute; The recalibrated feature vector corresponding to each disk attribute is residually connected with the first feature vector to obtain the second feature vector corresponding to each disk attribute.

5. The disk failure prediction method according to claim 1, It is characterized in that The method of using the time attention mechanism to determine the weight of a time point, and adjusting the second feature vector corresponding to each disk attribute based on the weight of the time point to obtain a third feature vector includes: Using a convolutional layer to perform feature aggregation processing on the second feature vectors of all disk attributes to obtain a feature map of a time domain signal; Processing the feature map of the time domain signal using an activation function to obtain the weight of the time point; Determine the local time signal in the second feature vector corresponding to each disk attribute based on a clustered signal segment extraction method; wherein the clustered signal segment extraction method is at least one of a time window-based method, an event-triggered method, a data-driven method, and a statistical characteristic-based method; Using a convolutional layer to encode feature information between local time signals in the second feature vector corresponding to each disk attribute, to obtain an adjusted second feature vector; Multiplying the weight of the time point and the adjusted second eigenvector corresponding to each disk attribute to obtain a recalibrated adjusted second eigenvector corresponding to each disk attribute; The recalibrated and adjusted second feature vector corresponding to each disk attribute and the corresponding second feature vector are added together to obtain the third feature vector corresponding to each disk attribute.

6. The disk failure prediction method according to claim 1, It is characterized in that Based on the third feature vector, a prediction model is used to perform prediction to obtain a fault prediction result corresponding to the time series data of each disk attribute, including: Performing global pooling processing on the third feature vector corresponding to each disk attribute by using global pooling in the prediction model to obtain pooling processing prediction data corresponding to each disk attribute; Using the fully connected layer in the prediction model to process the pooled prediction data corresponding to each disk attribute, to obtain global feature prediction data corresponding to each disk attribute; The global feature prediction data is processed using the activation function in the prediction model to determine the fault prediction result corresponding to each disk attribute.

7. The disk failure prediction method according to claim 1, It is characterized in that Based on the third feature vector, a prediction model is used to perform prediction to obtain a fault prediction result corresponding to the time series data of each disk attribute, including: Based on the third feature vector, the prediction model is used to perform prediction to obtain a classification label fault prediction result for each disk attribute; wherein, when the classification label is a first preset value, it is determined that the current disk attribute is in a healthy state; and when the classification label is a second preset value, it is determined that the current disk attribute is in a faulty state.

8. A disk failure prediction device, It is characterized in that include: A normalization processing module is used to perform normalization processing on the time series data corresponding to each disk attribute to obtain the normalized time series data corresponding to each disk attribute; A multi-scale processing module, used for performing multi-scale convolution on the normalized time series data corresponding to each disk attribute to obtain a first feature vector corresponding to each disk attribute; An attribute weight determination module, used to determine weights corresponding to different disk attributes by using a channel attention mechanism, and obtain a second feature vector corresponding to each disk attribute based on the weights and the first feature vector corresponding to each disk attribute; A time point weight determination module, used to determine the weight of a time point by using a time attention mechanism, and to adjust the second feature vector corresponding to each disk attribute based on the weight of the time point to obtain a third feature vector; The fault prediction module is used to perform prediction based on the third feature vector using a prediction model to obtain a fault prediction result corresponding to the time series data of each disk attribute.

9. A disk failure prediction device, It is characterized in that include: Memory for storing computer programs; A processor, configured to execute the computer program to implement the steps of the disk failure prediction method according to any one of claims 1 to 7.

10. A readable storage medium, It is characterized in that The readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the disk failure prediction method according to any one of claims 1 to 7 are implemented.