Solid-state hard disk remaining life analysis method and system based on improved classifier model

Abstract

The present invention relates to the field of data analysis technology, and invents a method for analyzing the remaining life of a solid-state hard disk based on an improved classifier model, comprising: performing outlier screening and time series smoothing interpolation operations on hard disk monitoring data to obtain a hard disk data sequence; performing data standardization operations on the hard disk data sequence to obtain a standard data sequence, and performing multi-channel feature extraction and mutual information dimension reduction operations on the standard data sequence to obtain a hard disk attribute feature sequence; performing bidirectional time series encoding and global attention encoding on the hard disk attribute feature sequence to obtain hard disk attribute time series features; performing variational fault classification and fault labeling and time series dilation convolution on the hard disk attribute time series features to obtain standard time series features, and performing life analysis on the standard time series features to obtain the remaining life of the hard disk. The present invention also proposes a solid-state hard disk remaining life analysis system based on an improved classifier model. The present invention can improve the accuracy of life analysis.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and in particular to a method and system for analyzing the remaining life of a solid-state hard disk based on an improved classifier model. Background Art

[0002] A solid-state drive (SSD) is a non-volatile storage device that uses flash memory to store data and does not rely on traditional mechanical components to access data. Compared with mechanical hard drives, solid-state drives have faster read and write speeds, higher shock resistance, and lower power consumption, but their lifespan is limited, mainly due to the number of erase and write times of the flash memory chip. As the storage device gradually ages, the performance of the solid-state drive will decline and eventually reach the end of its lifespan. Therefore, remaining life analysis of solid-state drives is the key to evaluating the health of the hard drive, ensuring data security, and maintaining the normal operation of the storage system.

[0003] Most of the existing methods for analyzing the remaining life of solid-state drives are based on statistical analysis. They usually rely on empirical formulas and statistical analysis to estimate the life of the hard drive through factors such as the number of erase and write cycles, temperature, and failure rate. In actual applications, the accuracy of the life analysis methods based on statistical analysis is low and cannot adapt to the complex operating modes of modern solid-state drives. They are insufficient in capturing complex timing characteristics and long-term trends, which may lead to low accuracy in the remaining life analysis. Summary of the invention

[0004] The present invention provides a solid state hard disk remaining life analysis method and system based on an improved classifier model, the main purpose of which is to solve the problem of low accuracy when performing remaining life analysis.

[0005] To achieve the above object, the present invention provides a method for analyzing the remaining life of a solid state drive based on an improved classifier model, comprising:

[0006] Extracting hard disk monitoring data from the monitoring log of the solid state hard disk, and performing outlier screening and time series smoothing interpolation operations on the hard disk monitoring data to obtain a hard disk data sequence;

[0007] Performing a data standardization operation on the hard disk data sequence to obtain a standard data sequence, and performing multi-channel feature extraction and mutual information dimension reduction operations on the standard data sequence to obtain a hard disk attribute feature sequence, wherein the performing a data standardization operation on the hard disk data sequence to obtain a standard data sequence includes: performing a non-negative value domain transformation on the hard disk data sequence to obtain a non-negative data sequence; initializing a transformation coefficient, and using the following data transformation algorithm to perform a data transformation on the non-negative data sequence according to the transformation coefficient to obtain a transformed data sequence:

[0008]

[0009] in, It means that when the transformation coefficient is When the first The data value of the data, refers to the transform coefficient, is the data index, is the first data; the normal coefficient of the transformed data sequence is calculated according to the following normality check algorithm:

[0010]

[0011] in, refers to the normal coefficient, is the data index, is the total number of the transformed data sequences, is the symbol for pi, is the standard deviation of the transformed data series, It means that when the transformation coefficient is When the first The data value of the data, is the mean of the transformed data sequence; performing maximum likelihood update on the transformation coefficient according to the normal coefficient to obtain a standard transformation coefficient; performing data transformation on the non-negative data sequence according to the standard transformation coefficient using the data transformation algorithm to obtain a standard data sequence;

[0012] Performing bidirectional temporal coding and global attention coding on the hard disk attribute feature sequence to obtain a hard disk attribute temporal feature;

[0013] Performing variational fault classification and fault labeling on the hard disk attribute time series features to obtain labeled time series features;

[0014] The annotated timing features are subjected to timing dilation convolution to obtain standard timing features, and the life of the solid state drive is analyzed according to the annotated timing features and the standard timing features to obtain the remaining life of the hard drive.

[0015] Optionally, the performing outlier screening and time series smoothing interpolation operations on the hard disk monitoring data to obtain a hard disk data sequence includes:

[0016] Sorting the hard disk monitoring data in time sequence to obtain a monitoring data sequence;

[0017] Performing outlier detection on the monitoring data sequence to obtain an outlier data group;

[0018] Selecting outlier data in the outlier data group one by one as target outlier data, and filtering out a neighborhood monitoring data group of the target outlier data from the monitoring data sequence;

[0019] The target primary interpolation data corresponding to the target outlier data is calculated according to the neighborhood monitoring data group using the following time series neighborhood interpolation algorithm:

[0020]

[0021] in, is the target primary interpolation data, is the timestamp of the target outlier data corresponding to the target primary interpolation data in the monitoring data sequence, is the data index, is the total number of data in the neighborhood monitoring data group, is the first The timestamp of each data in the monitoring data sequence, is the first individual data;

[0022] Aggregating target primary interpolation data of all target outlier data in the outlier data group into a primary interpolation data group;

[0023] Using the primary interpolation data group to update and replace the outlier data group in the monitoring data sequence to obtain a primary data sequence;

[0024] The primary data sequence is subjected to polynomial smoothing interpolation to obtain a hard disk data sequence.

[0025] Optionally, performing polynomial smoothing interpolation on the primary data sequence to obtain a hard disk data sequence includes:

[0026] Performing sparsity statistics on the primary data sequence to obtain sequence sparsity;

[0027] extracting a sparse data sequence from the primary data sequence according to the sequence sparsity;

[0028] extracting boundary conditions, continuity conditions and interpolation conditions from the sparse data sequence respectively;

[0029] generating an interpolation equation group according to the boundary condition, the continuity condition and the interpolation condition;

[0030] Performing smooth interpolation on the sparse data sequence according to the interpolation equation group to obtain an interpolation data sequence;

[0031] The interpolation data sequence is used to update the sparse data sequence in the primary data sequence to obtain a hard disk data sequence.

[0032] Optionally, performing multi-channel feature extraction and mutual information dimension reduction operations on the standard data sequence to obtain a hard disk attribute feature sequence includes:

[0033] Performing group convolution on the standard data sequence to obtain a primary attribute feature sequence group;

[0034] Performing deep convolution on the primary attribute feature sequence group to obtain a secondary attribute feature sequence group;

[0035] Calculating the mutual information value between each secondary attribute feature sequence in the secondary attribute feature sequence group;

[0036] According to the mutual information values ​​of all the secondary attribute feature sequences in the secondary attribute feature sequence group, mutual information dimension reduction is performed on the secondary attribute feature sequence group to obtain a hard disk attribute feature sequence.

[0037] Optionally, performing mutual information dimension reduction on the secondary attribute feature sequence group according to the mutual information values ​​of all the secondary attribute feature sequences in the secondary attribute feature sequence group to obtain the hard disk attribute feature sequence includes:

[0038] Generate a mutual information matrix according to the mutual information values ​​of all the secondary attribute feature sequences in the secondary attribute feature sequence group;

[0039] generating a feature sequence graph structure according to the mutual information matrix and the secondary attribute feature sequence group;

[0040] Performing spectral clustering on the characteristic sequence graph structure to obtain attribute characteristic sequence class groups;

[0041] Selecting attribute feature sequence classes in the attribute feature sequence class group one by one as target attribute feature sequence classes, performing eigenvalue decomposition on the target attribute feature sequence classes, and obtaining a target feature sequence vector and a target eigenvalue;

[0042] Filtering a standard attribute feature sequence from the target feature sequence vector according to the target feature value;

[0043] A hard disk attribute feature sequence is generated according to standard attribute feature sequences of all target attribute feature sequence classes in the attribute feature sequence class group.

[0044] Optionally, calculating the mutual information value between each secondary attribute feature sequence in the secondary attribute feature sequence group includes:

[0045] The mutual information value between each secondary attribute feature sequence in the secondary attribute feature sequence group is calculated using the following mutual information algorithm:

[0046]

[0047] in, is the secondary attribute feature sequence in the secondary attribute feature sequence group and secondary attribute feature sequence The mutual information value between is the secondary attribute feature sequence The characteristics of is the secondary attribute feature sequence The characteristics of The characteristics With the characteristics The joint probability distribution between is the preset mutual information coefficient, It is a feature The marginal probability of It is a feature The marginal probability of .

[0048] Optionally, performing variational fault classification and fault labeling on the hard disk attribute time series features to obtain the labeled time series features includes:

[0049] Performing low-dimensional space projection on the hard disk attribute time series features to obtain a time series feature distribution;

[0050] Performing a reparameterization operation on the time series feature distribution to obtain attribute variable features;

[0051] Performing high-dimensional space projection on the attribute variable features to obtain reconstructed hard disk time series features;

[0052] Calculating a variational fault error between the hard disk attribute timing characteristics and the reconstructed hard disk timing characteristics;

[0053] Performing fault classification on the variational fault error to obtain a hard disk fault event;

[0054] The hard disk attribute timing characteristics are fault-labeled according to the hard disk failure event to obtain the labeled timing characteristics.

[0055] Optionally, performing time series dilation convolution on the annotated time series features to obtain standard time series features includes:

[0056] Extracting a hard disk failure event from the annotated time series feature, and extracting an event timestamp from the hard disk failure event;

[0057] Performing autocorrelation analysis on the marked time series features to obtain a time series autocorrelation coefficient;

[0058] Extracting an autocorrelation timestamp from the annotated time series features according to the time series autocorrelation coefficient;

[0059] Generate a dilated timestamp according to the event timestamp and the autocorrelation timestamp;

[0060] Generating a dilated convolution kernel group according to the dilated timestamp, and performing dilated convolution on the annotated time series features using the dilated convolution kernel group to obtain a dilated time series feature group;

[0061] The expanded time series feature group is subjected to feature fusion to obtain a standard time series feature.

[0062] Optionally, performing life analysis on the solid state drive according to the marked timing characteristics and the standard timing characteristics to obtain the remaining life of the solid state drive includes:

[0063] Performing residual connection on the annotated time series features and the standard time series features to obtain residual time series features;

[0064] Performing feature decomposition and feature node extraction on the residual time series features to obtain a time series feature node set;

[0065] Performing independence check on the time series feature node set according to the residual time series feature to obtain a node dependency set;

[0066] Generate a feature dependency structure graph according to the time series feature node set and the node dependency set;

[0067] Greedily iterating the feature dependency structure graph to obtain a standard feature structure graph;

[0068] Performing iterative health status analysis on the residual time series features according to the standard feature structure diagram to obtain a hard disk health status sequence;

[0069] The life span of the solid state hard disk is analyzed according to the hard disk health status sequence to obtain the remaining life span of the hard disk.

[0070] In order to solve the above problems, the present invention also provides a solid state hard disk remaining life analysis system based on an improved classifier model, the system comprising:

[0071] A data interpolation module is used to extract hard disk monitoring data from the monitoring log of the solid state hard disk, and perform abnormal value screening and time series smoothing interpolation operations on the hard disk monitoring data to obtain a hard disk data sequence;

[0072] The data standardization module is used to perform a data standardization operation on the hard disk data sequence to obtain a standard data sequence, and perform a multi-channel feature extraction and mutual information dimension reduction operation on the standard data sequence to obtain a hard disk attribute feature sequence, wherein the data standardization operation on the hard disk data sequence to obtain a standard data sequence includes: performing a non-negative value domain transformation on the hard disk data sequence to obtain a non-negative data sequence; initializing a transformation coefficient, and using the following data transformation algorithm to perform a data transformation on the non-negative data sequence according to the transformation coefficient to obtain a transformed data sequence:

[0073]

[0074] in, It means that when the transformation coefficient is When the first The data value of the data, refers to the transform coefficient, is the data index, is the first data; the normal coefficient of the transformed data sequence is calculated according to the following normality check algorithm:

[0075]

[0076] in, refers to the normal coefficient, is the data index, is the total number of the transformed data sequences, is the symbol for pi, is the standard deviation of the transformed data series, It means that when the transformation coefficient is When the first The data value of the data, is the mean of the transformed data sequence; performing maximum likelihood update on the transformation coefficient according to the normal coefficient to obtain a standard transformation coefficient; performing data transformation on the non-negative data sequence according to the standard transformation coefficient using the data transformation algorithm to obtain a standard data sequence;

[0077] A time series analysis module, used for performing bidirectional time series coding and global attention coding on the hard disk attribute feature sequence to obtain a hard disk attribute time series feature;

[0078] A fault analysis module, used to perform variational fault classification and fault labeling on the hard disk attribute time series characteristics to obtain labeled time series characteristics;

[0079] The life analysis module is used to perform time expansion convolution on the marked timing features to obtain standard timing features, and perform life analysis on the solid state drive based on the marked timing features and the standard timing features to obtain the remaining life of the hard drive.

[0080] The present invention obtains a hard disk data sequence by performing outlier screening and time series smoothing interpolation operations on the hard disk monitoring data, can screen out outliers in the hard disk monitoring data, and expand the hard disk data, thereby solving the problem of small sample size in the hard disk data and improving the accuracy of remaining life analysis. By performing data standardization operations, the dimension and quantity set of the data can be unified, thereby improving the efficiency of feature extraction. By performing multi-channel feature extraction and mutual information dimensionality reduction operations, feature dimensionality reduction can be performed according to the correlation between feature sequences of different attributes, thereby improving the efficiency of feature calculation. By performing bidirectional time series coding and global attention coding, the time series features of each previous and next time steps of the hard disk attribute feature sequence can be extracted, the context sensitivity of the feature can be enhanced, and the complex interdependence between each time point in the sequence can be captured, thereby improving the accuracy of remaining life analysis.

[0081] By performing variational fault classification and fault labeling, the potential faults and possible faults of the hard disk can be flexibly fitted and analyzed in combination with the time series characteristics of the hard disk data, thereby assisting in the analysis of the remaining life of the hard disk and improving the accuracy of the remaining life analysis of the hard disk. By performing time series expansion convolution, global feature extraction can be further achieved, reducing the impact of gradient explosion and improving the accuracy of life analysis. Therefore, the solid-state hard disk remaining life analysis method and system based on the improved classifier model proposed in the present invention can solve the problem of low accuracy when performing remaining life analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0082] Figure 1 A schematic diagram of a process flow of a solid state drive remaining life analysis method based on an improved classifier model provided in one embodiment of the present invention;

[0083] Figure 2 A schematic diagram of a process for performing data standardization operations provided by an embodiment of the present invention;

[0084] Figure 3 A schematic diagram of a flow chart of a method for extracting a hard disk attribute feature sequence provided by an embodiment of the present invention;

[0085] Figure 4 A functional module diagram of a solid state drive remaining life analysis system based on an improved classifier model provided in one embodiment of the present invention;

[0086] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0087] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0088] The embodiment of the present application provides a method for analyzing the remaining life of a solid-state hard disk based on an improved classifier model. The execution subject of the method for analyzing the remaining life of a solid-state hard disk based on the improved classifier model includes but is not limited to at least one of the electronic devices that can be configured to execute the method provided in the embodiment of the present application, such as a server, a terminal, etc. In other words, the method for analyzing the remaining life of a solid-state hard disk based on the improved classifier model can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc. The server can be an independent server, or it can be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and big data and artificial intelligence platforms.

[0089] Reference Figure 1 FIG. 1 is a flow chart of a method for analyzing the remaining life of a solid state drive based on an improved classifier model according to an embodiment of the present invention. In this embodiment, the method for analyzing the remaining life of a solid state drive based on an improved classifier model includes:

[0090] S1. Extracting hard disk monitoring data from the monitoring log of the solid state hard disk, and performing abnormal value screening and time series smoothing interpolation operations on the hard disk monitoring data to obtain a hard disk data sequence.

[0091] In detail, the solid state drive (SSD) is a non-volatile storage device that uses flash memory to store data and does not rely on traditional mechanical components to access data. Compared with mechanical hard drives, solid state drives have faster read and write speeds, higher shock resistance and lower power consumption. The monitoring log refers to a log that records various parameters of the solid state drive.

[0092] Specifically, the hard disk monitoring data refers to the parameter data related to life analysis in the monitoring log, and the hard disk monitoring data includes the total number of written bytes, the number of retainable blocks, the wear leveling count, the average erase count, the consumed usable life, the solid state drive temperature, the uncorrectable error count, the data relocation count, the stability attributes, the active time, the abnormal shutdown count and the number of host read and write commands.

[0093] In detail, the Get-WmiObject command of PowerShell, the smartmontools tool, and the smartctl tool can be used to obtain the monitoring log from the solid state drive, and the hard disk monitoring data can be extracted from the monitoring log of the solid state drive using a regular expression matching method.

[0094] In the embodiment of the present invention, the step of performing outlier screening and time series smoothing interpolation operations on the hard disk monitoring data to obtain a hard disk data sequence includes:

[0095] Sorting the hard disk monitoring data in time sequence to obtain a monitoring data sequence;

[0096] Performing outlier detection on the monitoring data sequence to obtain an outlier data group;

[0097] Selecting outlier data in the outlier data group one by one as target outlier data, and filtering out a neighborhood monitoring data group of the target outlier data from the monitoring data sequence;

[0098] Calculating target primary interpolation data corresponding to the target outlier data according to the neighborhood monitoring data group;

[0099] Gathering target primary interpolation data of all target outlier data in the outlier data group into a primary interpolation data group;

[0100] Using the primary interpolation data group to update and replace the outlier data group in the monitoring data sequence to obtain a primary data sequence;

[0101] The primary data sequence is subjected to polynomial smoothing interpolation to obtain a hard disk data sequence.

[0102] Specifically, the time series sorting refers to sorting each data in the hard disk monitoring data in a time series order according to the timestamp of each data in the hard disk monitoring data, each monitoring data in the monitoring data sequence is grouped according to the data category, and the data in each group of data sequences is sorted in a time series order. Z-Score and density clustering algorithm (Density Based Spatial Clustering of Applications with Noise, referred to as DBSCAN) can be used for outlier detection. Outlier detection can filter out abnormal values ​​and garbled values ​​in the hard disk monitoring data, thereby improving the accuracy of the data set. The neighborhood monitoring data group refers to a data group composed of several monitoring data in the monitoring data sequence that have the same data type as the target outlier data and have a similar time domain distance.

[0103] Specifically, the target primary interpolation data is calculated using the following temporal neighborhood interpolation algorithm:

[0104]

[0105] in, is the target primary interpolation data, is the timestamp of the target outlier data corresponding to the target primary interpolation data in the monitoring data sequence, is the data index, is the total number of data in the neighborhood monitoring data group, is the first The timestamp of each data in the monitoring data sequence, is the first data.

[0106] Specifically, the time series neighborhood interpolation algorithm can perform numerical interpolation according to the time series trend characteristics of the outlier data in the time series sequence, thereby filling the data gaps caused by screening out the outlier data and providing more accurate time series characteristics.

[0107] In detail, the performing polynomial smoothing interpolation on the primary data sequence to obtain the hard disk data sequence includes:

[0108] Performing sparsity statistics on the primary data sequence to obtain sequence sparsity;

[0109] extracting a sparse data sequence from the primary data sequence according to the sequence sparsity;

[0110] extracting boundary conditions, continuity conditions and interpolation conditions from the sparse data sequence respectively;

[0111] generating an interpolation equation group according to the boundary condition, the continuity condition and the interpolation condition;

[0112] Performing smooth interpolation on the sparse data sequence according to the interpolation equation group to obtain an interpolation data sequence;

[0113] The interpolation data sequence is used to update the sparse data sequence in the primary data sequence to obtain a hard disk data sequence.

[0114] Specifically, the sparsity statistics refers to counting the sparsity of data in each time interval in the primary data sequence as the sequence sparsity, and the sparse data sequence refers to the data sequence corresponding to the time interval in which the sparsity of the sequence coefficient in the primary data sequence is greater than a preset sparsity threshold.

[0115] In detail, the boundary conditions include natural boundary conditions, fixed slope conditions and periodic conditions. The continuity conditions refer to the conditions for determining the continuity of the first-order derivatives and second-order derivatives of the corresponding interpolation equation group. The interpolation conditions refer to the conditions that the data points of the interpolation equation group pass through all data points in the sparse data sequence.

[0116] Specifically, a cubic spline interpolation algorithm can be used to generate an interpolation equation group according to the boundary conditions, the continuity conditions and the interpolation conditions. The cubic spline interpolation (Cubic Spline Interpolation) is a mathematical method for constructing a smooth curve through a series of shape value points. The smooth interpolation of the sparse data sequence according to the interpolation equation group refers to determining the total number of data that need to be interpolated according to the sparsity of the sequence, uniformly calculating the interpolation data of the corresponding time series position according to the total number of data using the interpolation equation group, and smoothly interpolating the sparse data sequence using the corresponding interpolation data.

[0117] In an embodiment of the present invention, a hard disk data sequence is obtained by performing outlier screening and time series smoothing interpolation operations on the hard disk monitoring data, which can screen out outliers in the hard disk monitoring data and expand the hard disk data, thereby solving the problem of small sample size in the hard disk data and improving the accuracy of remaining life analysis.

[0118] S2. Performing data standardization operation on the hard disk data sequence to obtain a standard data sequence, and performing multi-channel feature extraction and mutual information dimension reduction operation on the standard data sequence to obtain a hard disk attribute feature sequence.

[0119] In detail, the hard disk data sequence contains sequences of multiple different types of data, and these sequences have different dimensions and quantity sets. In order to more accurately capture data features, data standardization is required. Data standardization refers to unifying different types of data sequences in the hard disk data sequence to the same data level.

[0120] In the embodiment of the present invention, refer to Figure 2 As shown, the data standardization operation is performed on the hard disk data sequence to obtain a standard data sequence, including:

[0121] S21, performing a non-negative value domain transformation on the hard disk data sequence to obtain a non-negative data sequence;

[0122] S22, initializing transformation coefficients, and performing data transformation on the non-negative data sequence according to the transformation coefficients to obtain a transformed data sequence;

[0123] S23, calculating the normal coefficient of the transformed data sequence;

[0124] S24, performing maximum likelihood update on the transformation coefficient according to the normal coefficient to obtain a standard transformation coefficient;

[0125] S25. Use the data transformation algorithm to perform data transformation on the non-negative data sequence according to the standard transformation coefficient to obtain a standard data sequence.

[0126] In detail, the non-negative value domain transformation refers to screening out the minimum negative value from the hard disk data sequence, adding the absolute value of the minimum negative value to all data in the hard disk data sequence, so that all data in the non-negative data sequence are non-negative numbers.

[0127] Specifically, the transformation coefficient refers to a coefficient used for data transformation, and the transformation coefficient can be initialized using a random number algorithm.

[0128] In detail, the non-negative data sequence is transformed using the following data transformation algorithm to obtain a transformed data sequence:

[0129]

[0130] in, It means that when the transformation coefficient is When the first The data value of the data, refers to the transform coefficient, is the data index, is the first individual data;

[0131] The normality coefficient is calculated according to the normality check algorithm as follows:

[0132]

[0133] in, refers to the normal coefficient, is the data index, is the total number of the transformed data sequences, is the symbol for pi, is the standard deviation of the transformed data series, It means that when the transformation coefficient is When the first The data value of the data, is the mean of the transformed data sequence.

[0134] Specifically, the data transformation algorithm can reduce the value range of the non-negative data sequence, thereby achieving data lightweight and improving the efficiency of data calculation. The normality verification algorithm can ensure that the data in the standard data sequence after data transformation is close to the normal distribution, thereby improving the intuitiveness of the data trend. The maximum likelihood update refers to the use of the maximum likelihood estimation algorithm to continuously adjust the transformation coefficient so that the normal coefficient reaches the maximum value, and the transformation coefficient at this time is used as the standard transformation coefficient.

[0135] For details, refer to Figure 3 As shown, the multi-channel feature extraction and mutual information dimension reduction operation are performed on the standard data sequence to obtain a hard disk attribute feature sequence, including:

[0136] S31, performing group convolution on the standard data sequence to obtain a primary attribute feature sequence group;

[0137] S32, performing deep convolution on the primary attribute feature sequence group to obtain a secondary attribute feature sequence group;

[0138] S33, calculating the mutual information value between each secondary attribute feature sequence in the secondary attribute feature sequence group;

[0139] S34. Perform mutual information dimension reduction on the secondary attribute feature sequence group according to the mutual information values ​​of all the secondary attribute feature sequences in the secondary attribute feature sequence group to obtain a hard disk attribute feature sequence.

[0140] Specifically, 1x1 convolution can be used for group convolution, that is, the input channels are divided into several groups, and convolution operations are performed on each group separately to reduce the computational cost. The deep convolution refers to extracting spatial features from each channel without increasing the number of parameters.

[0141] In detail, the mutual information value is calculated using the following mutual information algorithm:

[0142]

[0143] in, is the secondary attribute feature sequence in the secondary attribute feature sequence group and secondary attribute feature sequence The mutual information value between is the secondary attribute feature sequence The characteristics of is the secondary attribute feature sequence The characteristics of The characteristics With the characteristics The joint probability distribution between is the preset mutual information coefficient, It is a feature The marginal probability of It is a feature The marginal probability of .

[0144] Specifically, the mutual information algorithm can control the sensitivity of the mutual information value calculation through the mutual information coefficient, thereby increasing the robustness of the mutual information calculation and improving the accuracy of feature dimensionality reduction.

[0145] In detail, performing mutual information dimension reduction on the secondary attribute feature sequence group according to the mutual information values ​​of all the secondary attribute feature sequences in the secondary attribute feature sequence group to obtain the hard disk attribute feature sequence includes:

[0146] Generate a mutual information matrix according to the mutual information values ​​of all the secondary attribute feature sequences in the secondary attribute feature sequence group;

[0147] generating a feature sequence graph structure according to the mutual information matrix and the secondary attribute feature sequence group;

[0148] Performing spectral clustering on the characteristic sequence graph structure to obtain attribute characteristic sequence class groups;

[0149] Selecting attribute feature sequence classes in the attribute feature sequence class group one by one as target attribute feature sequence classes, performing eigenvalue decomposition on the target attribute feature sequence classes, and obtaining a target feature sequence vector and a target eigenvalue;

[0150] Filtering a standard attribute feature sequence from the target feature sequence vector according to the target feature value;

[0151] A hard disk attribute feature sequence is generated according to standard attribute feature sequences of all target attribute feature sequence classes in the attribute feature sequence class group.

[0152] Specifically, the mutual information matrix is ​​a symmetric matrix constructed based on the mutual information values ​​between each pair of secondary attribute feature sequences in the secondary attribute feature sequence group, and the feature sequence graph structure refers to a graph structure constructed by taking each secondary attribute feature sequence in the secondary attribute feature sequence group as a graph node and taking the corresponding mutual information values ​​in the mutual information matrix as the values ​​of the edges between the graph nodes.

[0153] In detail, Spectral Clustering is a clustering algorithm based on spectral graph theory. Spectral clustering transforms the clustering problem into the optimal partitioning problem of the graph. Spectral clustering first defines an affinity matrix that describes the similarity of paired data points based on a given sample data set, calculates the eigenvalues ​​and eigenvectors of the affinity matrix, and then selects appropriate eigenvectors to cluster different data points.

[0154] Specifically, filtering out a standard attribute feature sequence from the target feature sequence vector according to the target feature value means taking a feature sequence vector whose corresponding target feature value in the target feature sequence vector is greater than a preset threshold as a standard attribute feature sequence, and a hard disk attribute feature sequence can be generated by splicing or aggregation.

[0155] In the embodiments of the present invention, by performing data standardization operations, the dimension and quantity set of the data can be unified, thereby improving the efficiency of feature extraction. By performing multi-channel feature extraction and mutual information dimensionality reduction operations, feature dimensionality reduction can be performed according to the correlation between feature sequences of different attributes, thereby improving the efficiency of feature calculation.

[0156] S3. Perform bidirectional temporal coding and global attention coding on the hard disk attribute feature sequence to obtain hard disk attribute temporal features.

[0157] In detail, the hard disk attribute time series feature refers to a feature that records the time series change relationship of the hard disk attribute feature sequence.

[0158] In the embodiment of the present invention, the step of performing bidirectional temporal coding and global attention coding on the hard disk attribute feature sequence to obtain the hard disk attribute temporal feature includes:

[0159] Performing forward time series feature extraction on the hard disk attribute feature sequence to obtain a forward attribute time series feature sequence;

[0160] Performing backward time series feature extraction on the hard disk attribute feature sequence to obtain a backward attribute time series feature sequence;

[0161] Performing feature concatenation on the forward attribute time series feature sequence and the backward attribute time series feature sequence to obtain a bidirectional attribute time series feature sequence;

[0162] Self-attention encoding is performed on each attribute bidirectional time series feature in the attribute bidirectional time series feature sequence to obtain a hard disk attribute time series feature.

[0163] In detail, the forward long short-term memory layer of the pre-trained bidirectional long short-term memory network (Bidirectional LongShort Term Memory, referred to as BiLSTM) can be used to perform forward time series feature extraction, and the backward long short-term memory layer of the bidirectional long short-term memory network can be used to perform backward time series feature extraction, so as to extract the time series features of each previous and next time steps of the hard disk attribute feature sequence, and the pre-trained Transformer model can be used for self-attention encoding, which can enhance the context sensitivity of the features, capture the complex interdependence between each time point in the sequence, and improve the accuracy of the remaining life analysis.

[0164] In an embodiment of the present invention, by performing bidirectional temporal coding and global attention coding, the temporal features of each previous and next time step of the hard disk attribute feature sequence can be extracted, the context sensitivity of the features can be enhanced, and the complex interdependence between each time point in the sequence can be captured, thereby improving the accuracy of the remaining life analysis.

[0165] S4. Perform variational fault classification and fault labeling on the hard disk attribute time series features to obtain labeled time series features.

[0166] In detail, the annotated time series feature refers to a hard disk attribute time series feature that annotates the type of hard disk failure or possible future failure and the timestamp.

[0167] In the embodiment of the present invention, performing variational fault classification and fault labeling on the hard disk attribute time series feature to obtain the labeled time series feature includes:

[0168] Performing low-dimensional space projection on the hard disk attribute time series features to obtain a time series feature distribution;

[0169] Performing a reparameterization operation on the time series feature distribution to obtain attribute variable features;

[0170] Performing high-dimensional space projection on the attribute variable features to obtain reconstructed hard disk time series features;

[0171] Calculating a variational fault error between the hard disk attribute timing characteristics and the reconstructed hard disk timing characteristics;

[0172] Performing fault classification on the variational fault error to obtain a hard disk fault event;

[0173] The hard disk attribute timing characteristics are fault-labeled according to the hard disk failure event to obtain the labeled timing characteristics.

[0174] Specifically, the low-dimensional space projection refers to compressing the hard disk attribute timing characteristics into a low-dimensional latent space using an encoder composed of multiple convolutional layers, recurrent neural networks, or long-short time series neural networks. The timing feature distribution can be the normal distribution of the hard disk attribute timing characteristics in the low-dimensional space. The reparameterization can ensure the continuity of the latent space by expressing a random variable as a function of another random variable, so that the output reconstructed hard disk timing characteristics are not discrete samples, which is convenient for subsequent fault analysis.

[0175] In detail, the high-dimensional space projection is the inverse operation of the low-dimensional space projection, and the mean square error algorithm or the cosine distance algorithm can be used to calculate the variational fault error between the hard disk attribute timing characteristics and the reconstructed hard disk timing characteristics. The fault classification refers to taking the hard disk attribute timing characteristics whose variational fault errors are greater than a preset error threshold as fault characteristics, performing fault event mapping on the corresponding variational fault errors, and obtaining hard disk fault events.

[0176] In the embodiment of the present invention, by performing variational fault classification and fault labeling, the potential faults and possible faults of the hard disk can be flexibly fitted and analyzed in combination with the timing characteristics of the hard disk data, thereby assisting in the analysis of the remaining life of the hard disk and improving the accuracy of the remaining life analysis of the hard disk.

[0177] S5. Perform time dilation convolution on the annotated timing features to obtain standard timing features, and perform life analysis on the solid state drive according to the annotated timing features and the standard timing features to obtain the remaining life of the hard drive.

[0178] In detail, performing time series dilation convolution on the marked time series features to obtain standard time series features includes:

[0179] Extracting a hard disk failure event from the annotated time series feature, and extracting an event timestamp from the hard disk failure event;

[0180] Performing autocorrelation analysis on the marked time series features to obtain a time series autocorrelation coefficient;

[0181] Extracting an autocorrelation timestamp from the annotated time series features according to the time series autocorrelation coefficient;

[0182] Generate a dilated timestamp according to the event timestamp and the autocorrelation timestamp;

[0183] Generating a dilated convolution kernel group according to the dilated timestamp, and performing dilated convolution on the annotated time series features using the dilated convolution kernel group to obtain a dilated time series feature group;

[0184] The expanded time series feature group is subjected to feature fusion to obtain a standard time series feature.

[0185] In detail, the event timestamp refers to the timestamp in the annotated time series feature corresponding to the hard disk failure event, and autocorrelation analysis can be performed using an autocorrelation function or a fast Fourier transform. Extracting the autocorrelation timestamp from the annotated time series feature based on the time series autocorrelation coefficient refers to taking the number of lag cycles when the time series autocorrelation coefficient in the annotated time series feature is the largest as the autocorrelation timestamp. The event timestamp and the autocorrelation timestamp correspond to the time steps in the annotated time series feature when the feature change amplitude is large or a failure event occurs. The convolution kernel design for time series expansion of these time steps can accurately capture the time series trend change characteristics.

[0186] Specifically, the dilated timestamp is composed of the event timestamp and the autocorrelation timestamp. Generating the dilated convolution kernel group according to the dilated timestamp refers to setting the size of the dilated convolution kernel to a multiple of the dilated timestamp. The feature fusion method can be a weighted feature summation method. By performing feature fusion, time series features of different scales can be fused, thereby improving the global features of the time series features.

[0187] In detail, the performing life analysis on the solid state drive according to the marked timing characteristics and the standard timing characteristics to obtain the remaining life of the solid state drive includes:

[0188] Performing residual connection on the annotated time series features and the standard time series features to obtain residual time series features;

[0189] Performing feature decomposition and feature node extraction on the residual time series features to obtain a time series feature node set;

[0190] Performing independence check on the time series feature node set according to the residual time series feature to obtain a node dependency set;

[0191] Generate a feature dependency structure graph according to the time series feature node set and the node dependency set;

[0192] Greedily iterating the feature dependency structure graph to obtain a standard feature structure graph;

[0193] Performing iterative health status analysis on the residual time series features according to the standard feature structure diagram to obtain a hard disk health status sequence;

[0194] The life span of the solid state hard disk is analyzed according to the hard disk health status sequence to obtain the remaining life span of the hard disk.

[0195] Specifically, the residual connection refers to direct feature fusion of the labeled time series features and the standard time series features. The residual time series features are time series features obtained by fusion of corresponding direct features. By performing residual connection, feature gradient disappearance or gradient explosion can be prevented. The feature decomposition refers to feature dimension decomposition of the residual time series features to obtain time series features of multiple dimensions. The feature node extraction refers to aggregating the time series feature nodes corresponding to the time series features of each dimension into a time series feature node set. The time series feature nodes are features corresponding to each time step of the time series features of each dimension.

[0196] In detail, the Akaike Information Criterion (AIC) or the Bayesian Information Criterion (BIC) can be used for independence verification. The independence verification can capture the dependency relationship between nodes in the time series feature node set, thereby facilitating the generation of subsequent feature dependency structure diagrams.

[0197] Specifically, generating a feature dependency structure graph according to the timing feature node set and the node dependency set refers to connecting the timing feature nodes in the node dependency set whose dependency values ​​are greater than a preset threshold into a feature dependency structure graph, and the greedy iteration refers to iterating the feature dependency structure graph using a greedy algorithm and the result of the independence check, thereby optimizing the graph structure of the feature dependency structure graph. The standard feature structure graph has timing characteristics and can capture the dependency relationship between timing features of different dimensions, thereby improving the accuracy of life prediction. Even if the information of some feature nodes is missing, the standard feature structure graph can still perform probabilistic inference based on the existing information to ensure the continuity and stability of the life analysis.

[0198] In detail, the iterative health status analysis refers to analyzing the health status of the hard disk time by time step by time step by time step iterative calculation method. The time step iterative calculation can be performed by variable elimination or belief propagation method. Each hard disk health status in the hard disk health status sequence represents the health probability of the solid state drive in the future time period.

[0199] Specifically, the life analysis of the solid-state hard disk according to the hard disk health status sequence to obtain the remaining life of the hard disk means taking the timestamp in the hard disk health status sequence in which the health probability is greater than a preset threshold and the hard disk health status does not change in a future time period as the life timestamp, and taking the difference between the life timestamp and the current timestamp as the remaining life of the hard disk.

[0200] In the embodiment of the present invention, by performing time-series dilation convolution, global feature extraction can be further achieved, the impact of gradient explosion can be reduced, and the accuracy of life analysis can be improved.

[0201] The present invention obtains a hard disk data sequence by performing outlier screening and time series smoothing interpolation operations on the hard disk monitoring data, can screen out outliers in the hard disk monitoring data, and expand the hard disk data, thereby solving the problem of small sample size in the hard disk data and improving the accuracy of remaining life analysis. By performing data standardization operations, the dimension and quantity set of the data can be unified, thereby improving the efficiency of feature extraction. By performing multi-channel feature extraction and mutual information dimensionality reduction operations, feature dimensionality reduction can be performed according to the correlation between feature sequences of different attributes, thereby improving the efficiency of feature calculation. By performing bidirectional time series coding and global attention coding, the time series features of each previous and next time steps of the hard disk attribute feature sequence can be extracted, the context sensitivity of the feature can be enhanced, and the complex interdependence between each time point in the sequence can be captured, thereby improving the accuracy of remaining life analysis.

[0202] By performing variational fault classification and fault labeling, the potential faults and possible faults of the hard disk can be flexibly fitted and analyzed in combination with the time series characteristics of the hard disk data, thereby performing auxiliary analysis on the remaining life of the hard disk and improving the accuracy of the remaining life analysis of the hard disk. By performing time series expansion convolution, global feature extraction can be further achieved, reducing the impact of gradient explosion and improving the accuracy of life analysis. Therefore, the solid-state hard disk remaining life analysis method based on the improved classifier model proposed in the present invention can solve the problem of low accuracy when performing remaining life analysis.

[0203] like Figure 4 , which is a functional module diagram of a solid state drive remaining life analysis system based on an improved classifier model provided in one embodiment of the present invention.

[0204] The solid state drive remaining life analysis system 100 based on the improved classifier model of the present invention can be installed in an electronic device. According to the functions implemented, the solid state drive remaining life analysis system 100 based on the improved classifier model can include a data interpolation module 101, a data standardization module 102, a timing analysis module 103, a fault analysis module 104 and a life analysis module 105. The module of the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0205] In this embodiment, the functions of each module / unit are as follows:

[0206] The data interpolation module 101 is used to extract hard disk monitoring data from the monitoring log of the solid state hard disk, and perform abnormal value screening and time series smoothing interpolation operations on the hard disk monitoring data to obtain a hard disk data sequence;

[0207] The data standardization module 102 is used to perform a data standardization operation on the hard disk data sequence to obtain a standard data sequence, and perform a multi-channel feature extraction and mutual information dimension reduction operation on the standard data sequence to obtain a hard disk attribute feature sequence, wherein the data standardization operation on the hard disk data sequence to obtain a standard data sequence includes: performing a non-negative value domain transformation on the hard disk data sequence to obtain a non-negative data sequence; initializing a transformation coefficient, and using the following data transformation algorithm to perform a data transformation on the non-negative data sequence according to the transformation coefficient to obtain a transformed data sequence:

[0208]

[0209] in, It means that when the transformation coefficient is When the first The data value of the data, refers to the transform coefficient, is the data index, is the first data; the normal coefficient of the transformed data sequence is calculated according to the following normality check algorithm:

[0210]

[0211] in, refers to the normal coefficient, is the data index, is the total number of the transformed data sequences, is the symbol for pi, is the standard deviation of the transformed data series, It means that when the transformation coefficient is When the first The data value of the data, is the mean of the transformed data sequence; performing maximum likelihood update on the transformation coefficient according to the normal coefficient to obtain a standard transformation coefficient; performing data transformation on the non-negative data sequence according to the standard transformation coefficient using the data transformation algorithm to obtain a standard data sequence;

[0212] The time series analysis module 103 is used to perform bidirectional time series coding and global attention coding on the hard disk attribute feature sequence to obtain the hard disk attribute time series feature;

[0213] The fault analysis module 104 is used to perform variational fault classification and fault labeling on the hard disk attribute time series characteristics to obtain the labeled time series characteristics;

[0214] The life analysis module 105 is used to perform time dilation convolution on the annotated timing features to obtain standard timing features, and perform life analysis on the solid state drive according to the annotated timing features and the standard timing features to obtain the remaining life of the hard drive.

[0215] In detail, each module described in the solid state drive remaining life analysis system 100 based on the improved classifier model in the embodiment of the present invention is used in the same manner as described above. Figure 1 The same technical means are used as the solid-state hard disk remaining life analysis method based on the improved classifier model described in, and can produce the same technical effects, so they will not be repeated here.

[0216] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, systems and methods can be implemented in other ways. For example, the system embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0217] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0218] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0219] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0220] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any attached figure mark in the claims should not be regarded as limiting the claims involved.

[0221] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0222] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The multiple units or systems stated in the system embodiment can also be implemented by one unit or system through software or hardware. The words first, second, etc. are used to indicate names, and do not indicate any specific order.

[0223] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, ordinary technicians in this field should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A method for analyzing the remaining life of a solid state drive based on an improved classifier model, characterized in that: The method comprises: Extracting hard disk monitoring data from the monitoring log of the solid state hard disk, and performing outlier screening and time series smoothing interpolation operations on the hard disk monitoring data to obtain a hard disk data sequence; Performing a data standardization operation on the hard disk data sequence to obtain a standard data sequence, and performing multi-channel feature extraction and mutual information dimension reduction operations on the standard data sequence to obtain a hard disk attribute feature sequence, wherein the performing a data standardization operation on the hard disk data sequence to obtain a standard data sequence includes: performing a non-negative value domain transformation on the hard disk data sequence to obtain a non-negative data sequence; initializing a transformation coefficient, and using the following data transformation algorithm to perform a data transformation on the non-negative data sequence according to the transformation coefficient to obtain a transformed data sequence: in, It means that when the transformation coefficient is When the first The data value of the data, refers to the transform coefficient, is the data index, is the first data; the normal coefficient of the transformed data sequence is calculated according to the following normality check algorithm: in, refers to the normal coefficient, is the data index, is the total number of the transformed data sequences, is the symbol for pi, is the standard deviation of the transformed data series, It means that when the transformation coefficient is When the first The data value of the data, is the mean of the transformed data sequence; performing maximum likelihood update on the transformation coefficient according to the normal coefficient to obtain a standard transformation coefficient; performing data transformation on the non-negative data sequence according to the standard transformation coefficient using the data transformation algorithm to obtain a standard data sequence; Performing bidirectional temporal coding and global attention coding on the hard disk attribute feature sequence to obtain a hard disk attribute temporal feature; Performing variational fault classification and fault labeling on the hard disk attribute time series features to obtain labeled time series features; The annotated timing features are subjected to timing dilation convolution to obtain standard timing features, and the life of the solid state drive is analyzed according to the annotated timing features and the standard timing features to obtain the remaining life of the hard drive.

2. The method for analyzing the remaining life of a solid state drive based on an improved classifier model according to claim 1, characterized in that: The step of performing outlier screening and time series smoothing interpolation operations on the hard disk monitoring data to obtain a hard disk data sequence includes: Sorting the hard disk monitoring data in time sequence to obtain a monitoring data sequence; Performing outlier detection on the monitoring data sequence to obtain an outlier data group; Selecting outlier data in the outlier data group one by one as target outlier data, and filtering out a neighborhood monitoring data group of the target outlier data from the monitoring data sequence; The target primary interpolation data corresponding to the target outlier data is calculated according to the neighborhood monitoring data group using the following time series neighborhood interpolation algorithm: in, is the target primary interpolation data, is the timestamp of the target outlier data corresponding to the target primary interpolation data in the monitoring data sequence, is the data index, is the total number of data in the neighborhood monitoring data group, is the first The timestamp of each data in the monitoring data sequence, is the first individual data; Gathering target primary interpolation data of all target outlier data in the outlier data group into a primary interpolation data group; Using the primary interpolation data group to update and replace the outlier data group in the monitoring data sequence to obtain a primary data sequence; The primary data sequence is subjected to polynomial smoothing interpolation to obtain a hard disk data sequence.

3. The method for analyzing the remaining life of a solid state drive based on an improved classifier model according to claim 2, characterized in that: The step of performing polynomial smoothing interpolation on the primary data sequence to obtain a hard disk data sequence includes: Performing sparsity statistics on the primary data sequence to obtain sequence sparsity; extracting a sparse data sequence from the primary data sequence according to the sequence sparsity; extracting boundary conditions, continuity conditions and interpolation conditions from the sparse data sequence respectively; generating an interpolation equation group according to the boundary condition, the continuity condition and the interpolation condition; Performing smooth interpolation on the sparse data sequence according to the interpolation equation group to obtain an interpolation data sequence; The interpolation data sequence is used to update the sparse data sequence in the primary data sequence to obtain a hard disk data sequence.

4. The method for analyzing the remaining life of a solid state drive based on an improved classifier model according to claim 1, characterized in that: The performing of multi-channel feature extraction and mutual information dimension reduction operations on the standard data sequence to obtain a hard disk attribute feature sequence includes: Performing group convolution on the standard data sequence to obtain a primary attribute feature sequence group; Performing deep convolution on the primary attribute feature sequence group to obtain a secondary attribute feature sequence group; Calculating the mutual information value between each secondary attribute feature sequence in the secondary attribute feature sequence group; According to the mutual information values ​​of all the secondary attribute feature sequences in the secondary attribute feature sequence group, mutual information dimension reduction is performed on the secondary attribute feature sequence group to obtain a hard disk attribute feature sequence.

5. The method for analyzing the remaining life of a solid state drive based on an improved classifier model according to claim 4, characterized in that: The step of performing mutual information dimension reduction on the secondary attribute feature sequence group according to the mutual information values ​​of all the secondary attribute feature sequences in the secondary attribute feature sequence group to obtain the hard disk attribute feature sequence includes: Generate a mutual information matrix according to the mutual information values ​​of all the secondary attribute feature sequences in the secondary attribute feature sequence group; generating a feature sequence graph structure according to the mutual information matrix and the secondary attribute feature sequence group; Performing spectral clustering on the characteristic sequence graph structure to obtain attribute characteristic sequence class groups; Selecting attribute feature sequence classes in the attribute feature sequence class group one by one as target attribute feature sequence classes, performing eigenvalue decomposition on the target attribute feature sequence classes, and obtaining a target feature sequence vector and a target eigenvalue; Filtering a standard attribute feature sequence from the target feature sequence vector according to the target feature value; A hard disk attribute feature sequence is generated according to standard attribute feature sequences of all target attribute feature sequence classes in the attribute feature sequence class group.

6. The method for analyzing the remaining life of a solid state drive based on an improved classifier model according to claim 4, characterized in that: The calculating the mutual information value between each secondary attribute feature sequence in the secondary attribute feature sequence group includes: The mutual information value between each secondary attribute feature sequence in the secondary attribute feature sequence group is calculated using the following mutual information algorithm: in, is the secondary attribute feature sequence in the secondary attribute feature sequence group and secondary attribute feature sequence The mutual information value between is the secondary attribute feature sequence The characteristics of is the secondary attribute feature sequence The characteristics of The characteristics With the characteristics The joint probability distribution between is the preset mutual information coefficient, It is a feature The marginal probability of It is a feature The marginal probability of .

7. The method for analyzing the remaining life of a solid state drive based on an improved classifier model according to claim 1, characterized in that: The performing variational fault classification and fault labeling on the hard disk attribute time series features to obtain the labeled time series features includes: Performing low-dimensional space projection on the hard disk attribute time series features to obtain a time series feature distribution; Performing a reparameterization operation on the time series feature distribution to obtain attribute variable features; Performing high-dimensional space projection on the attribute variable features to obtain reconstructed hard disk time series features; Calculating a variational fault error between the hard disk attribute timing characteristics and the reconstructed hard disk timing characteristics; Performing fault classification on the variational fault error to obtain a hard disk fault event; The hard disk attribute timing characteristics are fault-labeled according to the hard disk failure event to obtain the labeled timing characteristics.

8. The method for analyzing the remaining life of a solid state drive based on an improved classifier model according to claim 1, characterized in that: The step of performing time series dilation convolution on the marked time series features to obtain standard time series features includes: Extracting a hard disk failure event from the annotated time series feature, and extracting an event timestamp from the hard disk failure event; Performing autocorrelation analysis on the marked time series features to obtain a time series autocorrelation coefficient; Extracting an autocorrelation timestamp from the annotated time series features according to the time series autocorrelation coefficient; Generate a dilated timestamp according to the event timestamp and the autocorrelation timestamp; Generating a dilated convolution kernel group according to the dilated timestamp, and performing dilated convolution on the annotated time series features using the dilated convolution kernel group to obtain a dilated time series feature group; The expanded time series feature group is subjected to feature fusion to obtain a standard time series feature.

9. The method for analyzing the remaining life of a solid state drive based on an improved classifier model according to claim 1, characterized in that: The performing life analysis on the solid state hard disk according to the marked timing characteristics and the standard timing characteristics to obtain the remaining life of the hard disk includes: Performing residual connection on the annotated time series features and the standard time series features to obtain residual time series features; Performing feature decomposition and feature node extraction on the residual time series features to obtain a time series feature node set; Performing independence check on the time series feature node set according to the residual time series feature to obtain a node dependency set; Generate a feature dependency structure graph according to the time series feature node set and the node dependency set; Greedily iterating the feature dependency structure graph to obtain a standard feature structure graph; Performing iterative health status analysis on the residual time series features according to the standard feature structure diagram to obtain a hard disk health status sequence; The life span of the solid state hard disk is analyzed according to the hard disk health status sequence to obtain the remaining life span of the hard disk.

10. A solid state drive remaining life analysis system based on an improved classifier model, characterized in that: The system comprises: A data interpolation module is used to extract hard disk monitoring data from the monitoring log of the solid state hard disk, and perform abnormal value screening and time series smoothing interpolation operations on the hard disk monitoring data to obtain a hard disk data sequence; The data standardization module is used to perform a data standardization operation on the hard disk data sequence to obtain a standard data sequence, and perform a multi-channel feature extraction and mutual information dimension reduction operation on the standard data sequence to obtain a hard disk attribute feature sequence, wherein the data standardization operation on the hard disk data sequence to obtain a standard data sequence includes: performing a non-negative value domain transformation on the hard disk data sequence to obtain a non-negative data sequence; initializing a transformation coefficient, and using the following data transformation algorithm to perform a data transformation on the non-negative data sequence according to the transformation coefficient to obtain a transformed data sequence: in, It means that when the transformation coefficient is When the first The data value of the data, refers to the transform coefficient, is the data index, is the first data; the normal coefficient of the transformed data sequence is calculated according to the following normality check algorithm: in, refers to the normal coefficient, is the data index, is the total number of the transformed data sequences, is the symbol for pi, is the standard deviation of the transformed data series, It means that when the transformation coefficient is When the first The data value of the data, is the mean of the transformed data sequence; performing maximum likelihood update on the transformation coefficient according to the normal coefficient to obtain a standard transformation coefficient; performing data transformation on the non-negative data sequence according to the standard transformation coefficient using the data transformation algorithm to obtain a standard data sequence; A time series analysis module, used for performing bidirectional time series coding and global attention coding on the hard disk attribute feature sequence to obtain a hard disk attribute time series feature; A fault analysis module, used to perform variational fault classification and fault labeling on the hard disk attribute time series characteristics to obtain labeled time series characteristics; The life analysis module is used to perform time expansion convolution on the marked timing features to obtain standard timing features, and perform life analysis on the solid state drive based on the marked timing features and the standard timing features to obtain the remaining life of the hard drive.

Images