CN116302870A - Mechanical hard disk health assessment method, system and storage medium based on evolutionary diagram - Google Patents

Mechanical hard disk health assessment method, system and storage medium based on evolutionary diagram Download PDF

Info

Publication number
CN116302870A
CN116302870A CN202211611320.1A CN202211611320A CN116302870A CN 116302870 A CN116302870 A CN 116302870A CN 202211611320 A CN202211611320 A CN 202211611320A CN 116302870 A CN116302870 A CN 116302870A
Authority
CN
China
Prior art keywords
data
mechanical hard
target smart
hard disk
target
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202211611320.1A
Other languages
Chinese (zh)
Inventor
钟华
张月坤
刘志元
左大永
程宣
韩强
郑霄峰
覃绘桥
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Jingtou Excellence Technology Development Co ltd
Suzhou Huaqi Intelligent Technology Co ltd
Original Assignee
Beijing Jingtou Excellence Technology Development Co ltd
Suzhou Huaqi Intelligent Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Jingtou Excellence Technology Development Co ltd, Suzhou Huaqi Intelligent Technology Co ltd filed Critical Beijing Jingtou Excellence Technology Development Co ltd
Priority to CN202211611320.1A priority Critical patent/CN116302870A/en
Publication of CN116302870A publication Critical patent/CN116302870A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
    • G06F11/3466—Performance evaluation by tracing or monitoring
    • G06F11/3476—Data logging
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/3003—Monitoring arrangements specially adapted to the computing system or computing system component being monitored
    • G06F11/3034—Monitoring arrangements specially adapted to the computing system or computing system component being monitored where the computing system component is a storage system, e.g. DASD based or network based
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/3003—Monitoring arrangements specially adapted to the computing system or computing system component being monitored
    • G06F11/3037—Monitoring arrangements specially adapted to the computing system or computing system component being monitored where the computing system component is a memory, e.g. virtual memory, cache
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
    • G06F11/3452—Performance evaluation by statistical analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/30—Monitoring
    • G06F11/34—Recording or statistical evaluation of computer activity, e.g. of down time, of input/output operation ; Recording or statistical evaluation of user activity, e.g. usability assessment
    • G06F11/3466—Performance evaluation by tracing or monitoring
    • G06F11/3495—Performance evaluation by tracing or monitoring for systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computing Systems (AREA)
  • Quality & Reliability (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Artificial Intelligence (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Computation (AREA)
  • Data Mining & Analysis (AREA)
  • Computer Hardware Design (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • Probability & Statistics with Applications (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to a mechanical hard disk health assessment method, a system and a storage medium based on an evolutionary diagram. The method comprises the following steps: acquiring original data sets of the mechanical hard disks according to the full-life SMART data of the mechanical hard disks and the corresponding full-life health degree labels; obtaining importance degree sequencing of SMART attributes of the mechanical hard disk according to the original data set, obtaining target SMART attributes meeting preset sequencing requirements according to the importance degree sequencing, and obtaining target SMART data of all the target SMART attributes and corresponding target data sets; according to the multiscale evolution graph network, the target SMART attributes, the target SMART data and the target data set, self-learning the evolutionary relativity among the target SMART attributes, and acquiring a health evaluation model of the mechanical hard disk; and evaluating the health degree of the mechanical hard disk to be detected according to the health degree evaluation model of the mechanical hard disk. The invention can make the mechanical hard disk health degree evaluation accurate and reliable.

Description

Mechanical hard disk health assessment method, system and storage medium based on evolutionary diagram
Technical Field
The invention relates to the technical field of data storage, in particular to a mechanical hard disk health assessment method, a system and a storage medium based on an evolutionary diagram.
Background
The age of mass data has come, the data storage mode is also transferred to the cloud from the local place, and the reliability of a large-scale data storage center cannot be ignored. Mechanical hard disks are an important component of modern data centers, whose data access stability directly affects the reliability of the data storage system.
Currently, mechanisms for improving the reliability of data storage systems are classified into passive fault tolerance and active fault tolerance. The passive error tolerance refers to ensuring the safety of data by a data backup or erasure code mode when a mechanical hard disk fails. The passive fault tolerance mechanism severely increases the operational burden on the data center due to the large amount of data that needs to be backed up. Active fault tolerance is the detection of mechanical hard disk health using various hard disk attribute data provided by Self-monitoring, analysis and reporting techniques (Self-Monitoring Analysis and Reporting Technology, SMART). The existing mechanical hard disk health assessment method is mainly used for fault judgment and early warning through manually setting a fixed threshold, and is simple and direct, but has low accuracy, poor generalization, incapability of quantifying the health degree of the mechanical hard disk and serious influence on the accuracy of mechanical hard disk health assessment.
Disclosure of Invention
Therefore, the technical problem to be solved by the invention is that in the related art, fault judgment and early warning are carried out by manually setting a fixed threshold value so as to realize the evaluation of the health degree of the mechanical hard disk, the accuracy is low, the generalization is poor, the health degree of the mechanical hard disk can not be quantified, and the evaluation accuracy is seriously influenced.
In order to solve the technical problems, the invention provides a mechanical hard disk health assessment method based on an evolutionary diagram, which comprises the following steps:
acquiring original data sets of the mechanical hard disks according to the full-life SMART data of the mechanical hard disks and the corresponding full-life health degree labels;
obtaining importance degree sequencing of SMART attributes of the mechanical hard disk according to the original data set, obtaining target SMART attributes meeting preset sequencing requirements according to the importance degree sequencing, and obtaining target SMART data of all the target SMART attributes and corresponding target data sets;
according to the multiscale evolution graph network, the target SMART attributes, the target SMART data and the target data set, self-learning the evolutionary relativity among the target SMART attributes, and acquiring a health evaluation model of the mechanical hard disk;
and evaluating the health degree of the mechanical hard disk to be detected according to the health degree evaluation model of the mechanical hard disk.
Optionally, the obtaining the original data set of the plurality of mechanical hard disks according to the full life SMART data and the corresponding full life health label of the plurality of mechanical hard disks includes:
collecting full-life SMART data when a plurality of mechanical hard disks run to faults, and constructing full-life health degree labels corresponding to the mechanical hard disks;
combining the full-life SMART data and the full-life health degree label to obtain full-life original data of a plurality of mechanical hard disks;
and constructing a plurality of original data sets of the mechanical hard disk according to the life-span original data.
Optionally, the obtaining the importance ranking of the SMART attribute of the mechanical hard disk according to the original data set, obtaining the target SMART attribute meeting the preset ranking requirement according to the importance ranking, and obtaining the target SMART data of all the target SMART attributes and the corresponding target data set includes:
constructing an initial integrated tree model, and acquiring a training data set serving as input data of the initial integrated tree model according to the original data set; wherein the training data set is a normalized value of each SMART attribute selected from the original data set;
constructing a regression tree learning objective function, and carrying out iterative training on an initial integrated tree model according to a training data set to obtain a mature integrated tree model;
Obtaining importance ranking of all SMART attributes of the mechanical hard disk according to the occurrence times of each SMART attribute in the mature integrated tree model;
selecting a target SMART attribute meeting the preset ordering requirement from all SMART attributes according to the importance ordering;
and carrying out standardization processing according to the target SMART attribute to obtain target SMART data, obtaining a health degree label corresponding to the target SMART data according to the original data set, and combining the target SMART data and the health degree label to obtain a target data set with all the target SMART attribute.
Optionally, the obtaining the health assessment model of the mechanical hard disk according to the multi-scale evolution graph network, the target SMART attribute, the target SMART data and the target data set, and the self-learning of the correlation of the evolution among the target SMART attributes includes:
constructing a multi-scale evolution graph network, and combining the target SMART attributes, the target SMART data and the target data set, and self-learning the evolutionary correlation among the target SMART attributes to extract multi-scale time sequence characterization of the target SMART data;
and carrying out iterative updating on the multi-scale evolutionary graph network according to the multi-scale time sequence representation of the target SMART data to obtain a health evaluation model of the mechanical hard disk.
Optionally, the constructing a multi-scale evolution graph network, combining the target SMART attributes, the target SMART data and the target data set, self-learning the correlation of evolution among the target SMART attributes to extract a multi-scale timing characterization of the target SMART data, includes:
constructing a multi-scale evolutionary graph network, wherein the multi-scale evolutionary graph network comprises a time convolution module, an evolutionary graph self-learning module and an evolutionary graph convolution module;
inputting the target SMART data to a time convolution module, and extracting multi-scale time sequence characterization of the target SMART data along a time dimension;
taking the output data of the time convolution module as the input data of the evolution graph self-learning module, automatically extracting graph structure characterization of evolution among all target SMART attributes along with time, and outputting an evolved adjacency matrix;
and taking the adjacency matrix output by the evolution graph self-learning module and the multi-scale time sequence representation output by the time convolution module as input data of the evolution graph convolution module, and calculating to obtain a new multi-scale time sequence representation considering the correlation among all the target SMART attributes.
Optionally, the temporal convolution module includes an expanded convolution layer and a gating layer;
the inputting the target SMART data to the time convolution module, extracting a multi-scale time sequence representation of the target SMART data along a time dimension, including:
Dividing target SMART data into a plurality of time sequences according to SMART attributes, taking the time sequence of each target SMART attribute as input data of an expansion convolution layer, and performing expansion convolution operation by using convolution kernels with different sizes to obtain time sequence characterization with different time scales;
and taking the time sequence characterization of different time scales output by the expansion convolution layer as input data of the gating layer to obtain multi-scale time sequence characterization of each target SMART attribute output by the time convolution module.
Optionally, the self-learning module of the evolution diagram comprises a dividing layer of the diagram evolution stage and a self-learning layer of the diagram;
the method for automatically extracting the graph structural representation of evolution among the target SMART attributes along with time by taking the output data of the time convolution module as the input data of the evolution graph self-learning module and outputting an evolved adjacency matrix comprises the following steps:
dividing the multi-scale time sequence representation of the output data of the time convolution module into a plurality of sections along the time dimension by a graph evolution stage dividing layer, and aggregating the time sequence representation of each section into one dimension by using an aggregation function to serve as the attribute representation of each evolution stage;
automatically learning the graph structure representation of each evolution stage according to the attribute representation of each evolution stage by a graph learning layer, wherein the graph structure representation of each evolution stage is obtained by the graph structure representation of the previous evolution stage;
And connecting the graph structure representations of the evolution stages in pairs, and deducing an adjacency matrix of each evolution stage by using a multi-layer perceptron so as to obtain the adjacency matrix evolved through the output of the evolution graph self-learning module.
Optionally, the calculating, using the adjacency matrix output by the self-learning module of the evolutionary diagram and the multi-scale time sequence characterization output by the time convolution module as input data of the evolutionary diagram convolution module, obtains a new multi-scale time sequence characterization considering correlation between each target SMART attribute, including:
the method comprises the steps that an adjacency matrix output by a self-learning module of an evolutionary chart and a multi-scale time sequence representation output by a time convolution module are used as input data to be input into an evolutionary chart convolution module;
and carrying out graph convolution operation on the multi-scale time sequence representation of each target SMART attribute by utilizing the graph structure representation provided by the adjacent matrix through an evolutionary graph convolution module to obtain a new multi-scale time sequence representation considering the correlation among each target SMART attribute.
In addition, the invention also provides a mechanical hard disk health evaluation system based on the evolutionary diagram, which comprises:
the SMART data acquisition module is used for acquiring original data sets of the mechanical hard disks according to the full-life SMART data of the mechanical hard disks and the corresponding full-life health degree labels;
The target data acquisition module is used for obtaining importance ranking of SMART attributes of the mechanical hard disk according to the original data set, acquiring target SMART attributes meeting preset ranking requirements according to the importance ranking, and acquiring target SMART data of all the target SMART attributes and corresponding target data sets;
the evaluation model acquisition module is used for acquiring a health evaluation model of the mechanical hard disk according to the multi-scale evolution graph network, the target SMART attributes, the target SMART data and the target data set and by self-learning the correlation of evolution among the target SMART attributes;
and the health degree evaluation module is used for evaluating the health degree of the mechanical hard disk to be detected according to the health degree evaluation model of the mechanical hard disk.
Furthermore, the invention proposes a computer-readable storage medium, in which computer-executable instructions are stored, which when being executed by a processor are adapted to carry out all or part of the method steps of the method for evaluating the health of a mechanical hard disk based on an evolutionary diagram as described above.
The technical scheme provided by the invention has the following advantages:
according to the mechanical hard disk health assessment method based on the evolutionary diagram, time-varying correlation among the SMART attributes is modeled based on the SMART data of the mechanical hard disk, multi-scale time sequence characterization in the original data is mined, and multi-scale time sequence characterization is aggregated by combining the interdependence among the SMART attributes, so that the health degree of the mechanical hard disk is accurately assessed and quantified. The method has the advantages of higher precision, better stability and stronger robustness.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is apparent that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.
FIG. 1 is a schematic flow chart of steps of a mechanical hard disk health evaluation method based on an evolutionary diagram according to an embodiment of the invention;
FIG. 2 is a graph-based attribute importance score for each SMART attribute in a mechanical hard disk health assessment method according to an embodiment of the present invention;
FIG. 3 is a diagram of the result of evaluating the health of a mechanical hard disk according to the method for evaluating the health of a mechanical hard disk based on an evolutionary diagram according to the embodiment of the invention;
FIG. 4 is a comparison chart of performance of evaluating the health of a mechanical hard disk according to the method for evaluating the health of a mechanical hard disk based on an evolutionary diagram according to the embodiment of the invention;
fig. 5 is a schematic block diagram of a mechanical hard disk health evaluation system based on an evolutionary diagram according to an embodiment of the invention.
Detailed Description
The following description of the embodiments of the present invention will be made apparent and fully in view of the accompanying drawings, in which some, but not all embodiments of the invention are shown. The invention will be described in detail hereinafter with reference to the drawings in conjunction with embodiments. It should be noted that, without conflict, the embodiments of the present invention and features of the embodiments may be combined with each other.
In the related art, the mechanical hard disk health assessment method is mainly used for fault judgment and early warning through manually setting a fixed threshold, and the method is simple and direct, but has low accuracy, poor generalization, and cannot quantify the health degree of the mechanical hard disk, so that the accuracy of mechanical hard disk health assessment is seriously affected. In order to solve the technical problems, the invention provides a mechanical hard disk health assessment method and system based on an evolutionary diagram.
The invention provides a mechanical hard disk health assessment method based on an evolutionary diagram, which is used for assessing the health condition of a mechanical hard disk. Specifically, as shown in fig. 1, the mechanical hard disk health assessment method based on the evolutionary diagram may include the following steps:
s100, acquiring original data sets of a plurality of mechanical hard disks according to full-life SMART data of the plurality of mechanical hard disks and corresponding full-life health degree labels;
s200, obtaining importance degree ranking of SMART attributes of the mechanical hard disk according to the original data set, obtaining target SMART attributes meeting preset ranking requirements according to the importance degree ranking, and obtaining target SMART data of all the target SMART attributes and corresponding target data sets;
s300, according to a multi-scale evolution graph network, target SMART attributes, target SMART data and a target data set, self-learning the evolutionary relativity among the target SMART attributes, and acquiring a health evaluation model of the mechanical hard disk;
S400, according to the health degree evaluation model of the mechanical hard disk, evaluating the health degree of the mechanical hard disk to be detected.
Based on the original full-life SMART data of the mechanical hard disk, a deep learning method is adopted to analyze and mine the law contained in each attribute data, the health degree of the mechanical hard disk is estimated and quantized, and the reliability and maintainability of a data storage system are greatly improved. Moreover, data of each SMART attribute in SMART technology has an interdependence relationship, and the observed value of each SMART attribute depends not only on its historical value but also on the current observed values of other SMART attributes, and this interdependence relationship is not constant, and it gradually changes with the deterioration of the health state of the mechanical hard disk.
According to the mechanical hard disk health assessment method based on the evolutionary diagram, time-varying correlation among the SMART attributes is modeled based on the SMART data of the mechanical hard disk, multi-scale time sequence characterization in the original data is mined, and multi-scale time sequence characterization is aggregated by combining the interdependence among the SMART attributes, so that the health degree of the mechanical hard disk is accurately assessed and quantified. The method has the capability of modeling time-varying interdependencies among the SMART attributes, so that the mechanical hard disk health evaluation accuracy is higher, the stability is better, and the robustness is stronger.
Further, in step S100, according to the full life SMART data and the corresponding full life health label of the plurality of mechanical hard disks, an original data set of the plurality of mechanical hard disks is obtained, which specifically includes the following steps:
s110, collecting full-life SMART data when a plurality of mechanical hard disks run to faults, and constructing full-life health degree labels corresponding to the mechanical hard disks.
In particular, the full life SMART data of a plurality of mechanical hard disks can be acquired and obtained in advance when the mechanical hard disks run to faults
Figure BDA0003998925330000071
Wherein S is t Sample SMART data representing a mechanical hard disk at a sample time t;
meanwhile, the corresponding full-life health degree label can be constructed
Figure BDA0003998925330000072
Wherein C is t A sample health label representing a sample time t; c (C) t The calculated expression of (2) is as follows:
C t ＝t/L；
where t represents the sampling time, i.e. the mechanical hard disk from the time of use to the time of sampling SMART data S t The elapsed time of use; l is the total service time of the mechanical hard disk from the time of being put into use to the time of failure, namely the service life of the mechanical hard disk.
And S120, combining the full-life SMART data and the full-life health degree labels to obtain full-life original data of the mechanical hard disks.
The whole service life SMART data of each mechanical hard disk can be obtained
Figure BDA0003998925330000081
And health label->
Figure BDA0003998925330000082
Combining to obtain the life-time original data of the mechanical hard disks +.>
Figure BDA0003998925330000083
S130, constructing an original data set of a plurality of mechanical hard disks according to the original data of the whole service life.
The whole life data of each mechanical hard disk can be obtained
Figure BDA0003998925330000084
Raw data set D constituting these mechanical hard disks O ＝{(S i ,C i ) -a }; wherein S is i SMART data representing a mechanical hard disk at sample time i; c (C) i A health label representing the sampling time i.
In step S200, the importance ranking of the SMART attribute of the mechanical hard disk is obtained according to the original data set, the target SMART attribute meeting the preset ranking requirement is obtained according to the importance ranking, and the target SMART data and the corresponding target data set of all the target SMART attributes are obtained, which specifically includes the following steps:
s210, constructing an initial integrated tree model, and acquiring a training data set serving as input data of the initial integrated tree model according to an original data set; wherein the training dataset is a normalized value for each SMART attribute selected from the original dataset.
In particular, the raw data set D can be obtained from a mechanical hard disk O ＝{(S i ,C i ) Normalized value (normalized_value) of each SMART attribute is selected as training data set D X ＝{(S' i ,C i ) }, wherein
Figure BDA0003998925330000085
F represents the number of SMART attributes provided by the model of mechanical hard disk;
Constructing an initial integrated tree model to train the data set D X ＝{(S' i ,C i ) As input data for the initial integrated tree model; moreover, the initial setThe mathematical expression of the tree-forming model is as follows:
Figure BDA0003998925330000086
in the method, in the process of the invention,
Figure BDA0003998925330000087
representing a set of regression trees; k represents the number of regression trees; q represents the structure of each regression tree; f (·) represents the regression tree input-output function relationship; q (·) represents the functional relationship of each regression tree, which maps the sampled SMART data S to the leaf nodes to which the prediction results correspond; w (w) q(S) A weight representing the q (S) th leaf node; n (N) T Representing the number of leaf nodes of the regression tree.
S220, constructing a regression tree learning objective function, and performing iterative training on the initial integrated tree model according to the training data set to obtain a mature integrated tree model.
Constructing a regression tree learning objective function for learning each regression tree in the initial integrated tree model; moreover, the mathematical expression of the regression tree learning objective function is as follows:
Figure BDA0003998925330000091
where l (·) is a micro-loss function, measure the health prediction
Figure BDA0003998925330000092
And sample health label C i Is a difference in (2); Ω (f) =γn T +0.5λ||w|| 2 A model complexity penalty term used to smooth the weights ultimately learned by the model to prevent overfitting and limit the total number of leaf nodes;
Performing K iterations on the initial integrated tree model according to the training data set, and learning a regression tree model each iteration; the regression tree learning objective function of the kth iteration is as follows:
Figure BDA0003998925330000093
and finally obtaining the mature integrated tree model phi (·) through iterative training.
S230, according to the occurrence times of each SMART attribute in the mature integrated tree model, the importance ranking of all SMART attributes of the mechanical hard disk is obtained.
Specifically, the number of times that each SMART attribute appears in the trained mature integrated tree model phi (·) can be calculated, and the number of times that each SMART attribute appears in the mature integrated tree model phi (·) is used as an importance score value of the SMART attribute, and all SMART attributes provided by the mechanical hard disk are ranked according to the importance score value, so as to obtain the importance ranking of all SMART attributes of the mechanical hard disk.
S240, selecting a target SMART attribute meeting the preset ordering requirement from all SMART attributes according to the importance ordering. The preset ranking requirement may be that the importance score value of the SMART attribute needs to lie within a certain range of values of the importance ranking (e.g., top 40%, top 50%, or top 60%, etc.).
Specifically, according to the importance ranking of the SMART attributes of the mechanical hard disk, the SMART attribute with the importance score value being the first 50% in the importance ranking in all the SMART attributes may be selected as the target SMART attribute.
S250, performing standardization processing according to the target SMART attribute to obtain target SMART data, obtaining a health degree label corresponding to the target SMART data according to the original data set, and combining the target SMART data and the health degree label to obtain a target data set with all the target SMART attributes.
Specifically, the normalization processing can be performed according to the target SMART attribute, that is, the normalized value of the target SMART attribute is obtained to obtain the target SMART data;
original data set D based on mechanical hard disk O ＝{(S i ,C i ) Acquiring a health tag corresponding to the target SMART data, and constructing a target data set combining the target SMART data and the health tag to acquire all target SMART attributes as follows:
D＝{(X i ,Y i ) And } wherein,
Figure BDA0003998925330000101
target SMART data representing a window width P, N representing the number of target SMART attributes used to construct the dataset, +.>
Figure BDA0003998925330000102
A normalized value representing each target SMART attribute for the t sampling time; />
Figure BDA0003998925330000103
Health tag representing last sampling time of each time window as target SMART data X i Is a health label of (a).
In addition, in step S300, according to the multi-scale evolution graph network, the target SMART attribute, the target SMART data and the target data set, the health evaluation model of the mechanical hard disk is obtained by self-learning the correlation of the evolution among the target SMART attributes, which may specifically include the following steps:
S310, constructing a multi-scale evolution graph network, and combining the target SMART attributes, the target SMART data and the target data set, and self-learning the evolutionary correlation among the target SMART attributes to extract multi-scale time sequence characterization of the target SMART data;
and S320, carrying out iterative updating on the multi-scale evolutionary graph network according to multi-scale time sequence characterization of the target SMART data to obtain a health evaluation model of the mechanical hard disk.
Further, in step S310, a multi-scale evolution graph network is constructed, and in combination with the target SMART attributes, the target SMART data, and the target data set, the multi-scale timing characterization of the target SMART data is extracted by self-learning the correlations evolved between the target SMART attributes, which may further include the following steps:
s312, constructing a multi-scale evolution graph network, wherein the multi-scale evolution graph network comprises a time convolution module, an evolution graph self-learning module and an evolution graph rolling module. The input data X of each layer of multi-scale evolution graph network is the output of the previous layer, and the input data of the first layerFor target SMART data X i 。
S314, inputting the target SMART data into a time convolution module, and extracting multi-scale time sequence characterization of the target SMART data along the time dimension.
S316, taking the output data of the time convolution module as the input data of the evolution graph self-learning module, automatically extracting graph structure characterization of evolution among all target SMART attributes along with time, and outputting an evolved adjacency matrix.
S318, taking the adjacency matrix output by the evolution graph self-learning module and the multi-scale time sequence representation output by the time convolution module as input data of the evolution graph convolution module, and calculating to obtain a new multi-scale time sequence representation considering the correlation among the SMART attributes of each target.
Still further, the temporal convolution module includes an expanded convolution layer and a gating layer. Also, in step S314, the target SMART data is input to the temporal convolution module, and the multi-scale timing characterization of the target SMART data is extracted along the temporal dimension, which may further include the steps of:
s3142, dividing target SMART data into a plurality of time sequences according to SMART attributes, using the time sequence of each target SMART attribute as input data of an expansion convolution layer, and performing expansion convolution operation by using convolution kernels with various different sizes to obtain time sequence representations with different time scales.
Specifically, the input data, i.e., target SMART data, may be divided into N time series X by SMART attributes (1) ,X (2) ,…X (N) Time-series of each target SMART attribute
Figure BDA0003998925330000111
As the input data of the expansion convolution layer, using a plurality of convolution kernels with different sizes to perform expansion convolution operation, and capturing time sequence characterization with different time scales;
Wherein, the calculation expression of the expansion convolution operation is as follows:
Figure BDA0003998925330000112
in the method, in the process of the invention,
Figure BDA0003998925330000113
a number a of convolution operation output is represented; />
Figure BDA0003998925330000114
Representing a size of 1 xk j Is a one-dimensional convolution kernel of (a); d, d t Is the expansion rate.
And, cut the output of each expansion convolution operation to the same length according to the convolution operation output of the maximum convolution kernel, and connect it to get the multi-scale time sequence representation of the expansion convolution layer output, the calculation expression is as follows:
Figure BDA0003998925330000115
wherein concat (·) represents a stitching function; w represents the number of convolution kernels of the expanded convolution layer, the w-th convolution kernel having a size of 1 xk w ；P ξ Representing the dimension of the dilated convolution layer output.
S3144, using the time sequence representations of different time scales output by the expansion convolution layer as input data of the gating layer, and obtaining multi-scale time sequence representations of all target SMART attributes output by the time convolution module.
Specifically, taking the multi-scale time sequence representation output by the expansion convolution layer as the input of the gating layer, and controlling the information quantity transferred to the next module by the time convolution module; moreover, the computational expression of the gating layer is as follows:
ξ＝σ(ξ)⊙μ(ξ)；
wherein μ (·) represents the tanh activation function; sigma (·) represents the sigmoid activation function as a gate pair μ (ζ ( ' i) ) Filtering the information of (2); the Hadamard product, corresponding to the element multiplication;
The input data X is calculated by the time convolution module through the steps, and finally the multi-scale time sequence characterization of each target SMART attribute of the output of the time convolution module can be obtained as follows:
Figure BDA0003998925330000121
i.e. < ->
Figure BDA0003998925330000122
Moreover, the evolution diagram self-learning module comprises a diagram evolution stage division layer and a diagram self-learning layer. In step S316, taking the output data of the time convolution module as the input data of the evolution graph self-learning module, automatically extracting graph structure characterizations evolving with time between each target SMART attribute, and outputting an evolving adjacency matrix, the method may further include the steps of:
s3162, dividing the multi-scale time sequence representation of the output data of the time convolution module into a plurality of sections along the time dimension through a graph evolution stage dividing layer, and aggregating the time sequence representation of each section into one dimension by using an aggregation function to serve as the attribute representation of each evolution stage.
Specifically, input data formed by multi-scale timing characterization of output data of a time convolution module by graph evolution stage partitioning
Figure BDA0003998925330000123
Dividing the time dimension into a plurality of segments, and aggregating the time sequence representation of each segment into one dimension by using an aggregation function, wherein the time sequence representation is used as the attribute representation of each evolution stage, and the expression is as follows:
Figure BDA0003998925330000124
Figure BDA0003998925330000125
in the method, in the process of the invention,
Figure BDA0003998925330000126
AGG (·) represents an aggregation function, such as a mean function; d, d a Representing the time interval of each phaseThe method comprises the steps of carrying out a first treatment on the surface of the M is the total number of segments in the evolution phase.
S3164, automatically learning the graph structure representation of each evolution stage according to the attribute representation of each evolution stage through a graph learning layer, wherein the graph structure representation of each evolution stage is obtained by the graph structure representation of the previous evolution stage.
Specifically, the graph structure characterization of each evolution stage can be automatically learned according to the attribute characterization of each evolution stage through the graph learning layer, and the graph structure characterization of each evolution stage is obtained by the graph structure characterization of the previous evolution stage, namely the obtained graph structure characterization of each stage is gradually evolved. The graph learning layer can characterize the attribute of each evolutionary stage by gamma m And the graph structure representation alpha of the previous stage m-1 As an input sequence, obtaining the graph structure representation alpha of the current stage m The mathematical expression is as follows:
r m ＝σ(W r [γ m ,α m-1 ]+b r )；
u m ＝σ(W u [γ m ,α m-1 ]+b u )；
o m ＝σ(W o [γ m ,(r m ⊙α m-1 )]+b o )；
α m ＝u m ⊙α m-1 +(1-u m )⊙o m ；
wherein r is m And u m Respectively representing a forget gate and an update gate; the Hadamard product, corresponding to the element multiplication; w (W) r 、W u And W is o Representing a learnable weight matrix; b r 、b u And b o Representing the bias term.
Through the steps, a series of continuously evolving graph structure characterization is finally obtained
Figure BDA0003998925330000131
Wherein->
Figure BDA0003998925330000132
Representation of the graph structure representing the mth evolutionary phase.
S3166, connecting the graph structure representations of the evolution stages in pairs, and deducing an adjacent matrix of each evolution stage by using a multi-layer perceptron so as to obtain the adjacent matrix evolved through the output of the self-learning module of the evolution graph.
Specifically, the graph structure characterization of each evolution stage is connected in pairs, a multi-layer perceptron is used for automatically deducing an adjacent matrix of each evolution stage, and the mathematical expression is as follows:
Figure BDA0003998925330000133
M m,ij ＝MLP m (concat(α m,(i) ,α m,(j) ))；
Figure BDA0003998925330000134
in the method, in the process of the invention,
Figure BDA0003998925330000135
candidate adjacency matrix representing the mth evolutionary phase +.>
Figure BDA0003998925330000136
Elements of the ith row and the jth column; m is M m,ij Elements representing the ith row and jth column of the mask matrix; />
Figure BDA0003998925330000137
A adjacency matrix representing an mth evolutionary phase;
the output of the evolutionary diagram self-learning module is a series of adjacency matrices a= { a over M evolutionary stages 1 ,A 2 ,…,A m ,…A M -wherein the adjacency matrix a m From the adjacency matrix A of the previous stage m-1 And the current stage input data alpha m Co-determination, i.e. A m From A m-1 Evolved.
In step S318, the adjacency matrix output by the self-learning module of the evolutionary diagram and the multi-scale time sequence representation output by the time convolution module are used as input data of the evolutionary diagram convolution module, and the operation results in a new multi-scale time sequence representation considering the correlation between the target SMART attributes, which may further include the following steps:
s3182, the adjacency matrix output by the evolution graph self-learning module and the multi-scale time sequence representation output by the time convolution module are used as input data to be input into the evolution graph convolution module.
Specifically, the adjacency matrix output by the evolution graph self-learning module and the multi-scale time sequence characterization output by the time convolution module
Figure BDA0003998925330000141
Input data of an information propagation layer serving as an evolutionary graph convolution module; first, a multiscale timing characterization +.>
Figure BDA0003998925330000142
The graph evolution stage is divided according to the division method of the step S3162, and the mathematical expression is as follows:
Figure BDA0003998925330000143
in the method, in the process of the invention,
Figure BDA0003998925330000144
d a representing the time interval of each phase; m is the total number of segments in the evolution phase, m=p ξ /d a 。
S3184, performing graph convolution operation on the multi-scale time sequence representation of each target SMART attribute by using the graph structure representation provided by the adjacent matrix through an evolution graph convolution module to obtain a new multi-scale time sequence representation considering the correlation among each target SMART attribute.
Specifically, the time sequence of each stage is characterized by an evolution graph rolling module
Figure BDA0003998925330000145
Adjacent matrix A corresponding to it m Performing graph convolution operation to obtain a time sequence table integrating attribute correlation informationThe mathematical expression of the graph convolution operation is as follows:
Figure BDA0003998925330000146
Figure BDA0003998925330000151
wherein I represents a unit array;
Figure BDA0003998925330000152
representation->
Figure BDA0003998925330000153
Is a diagonal matrix, and the calculation expression of the ith element on the diagonal is: />
Figure BDA0003998925330000154
W G Representing a self-learnable weight matrix;
the time sequence representation of each stage is integrated with the adjacent matrix of the corresponding stage to obtain the time sequence representation of the integrated attribute correlation information
Figure BDA0003998925330000155
In addition, in S320, the multi-scale evolutionary graph network is iteratively updated according to the multi-scale timing characterization of the target SMART data to obtain the health evaluation model of the mechanical hard disk, which specifically includes the following steps:
s322, integrating the input of the multi-scale evolutionary graph network into the output by adopting residual connection, wherein the output calculation expression is as follows:
Z＝h(X)+H；
wherein X is input data of a multi-scale evolution graph network; h (·) is a 1×1 convolution operation for adjusting the dimension;
s324, repeating the step S310 once, i.e. alternately stacking an evolutionary graph learning structure, and networking the last layer of multi-scale evolutionary graphThe output of (2) is denoted as Z (l) ；
S326, integrating output information of each layer of multi-scale evolutionary graph network through residual connection, and calculating a health value of a mechanical hard disk by using a fully-connected network, wherein the mathematical expression is as follows:
Figure BDA0003998925330000156
where h (·) is a 1×1 convolution operation for adjusting the dimension;
Figure BDA0003998925330000157
representing a health degree evaluation value of the mechanical hard disk;
s328, repeating the steps S310 and S320 based on the Adam optimization algorithm, setting the iteration times E, and iteratively updating parameters of the multi-scale evolutionary graph network to obtain an optimal health evaluation model of the mechanical hard disk, namely minimizing a mean square error objective function:
Figure BDA0003998925330000161
Wherein Y is the health label value of the mechanical hard disk.
In step S400, according to the health evaluation model of the mechanical hard disk, the evaluation of the health of the mechanical hard disk to be detected may specifically include:
according to the health evaluation model of the mechanical hard disk, the target SMART data of the mechanical hard disk is obtained
Figure BDA0003998925330000162
And inputting the health degree to an optimal health degree evaluation model, and evaluating the health degree of the mechanical hard disk to be detected.
The invention can directly extract multi-scale time sequence characterization from the SMART data of the mechanical hard disk, model the gradually evolving interdependence relation of each attribute of the SMART data, and further accurately mine degradation information related to the health degree of the mechanical hard disk. The method overcomes the defect of the lack of capability of modeling the continuously evolved attribute correlation of the existing method, and realizes the health assessment of the mechanical hard disk based on SMART data.
Specifically, using the Seagate ST4000DM000 hard disk as a case, 657 sets of life-time data were used to verify the effectiveness of the method of the present invention. The parameters of the evolution graph network are set as follows: the number of layers of the evolution graph network is 3; the number of training in small batches is 16; the number of iterations was chosen to be 100.
During experimental verification, firstly, quantifying importance of each attribute of SMART data, sorting, selecting an attribute for network training according to an attribute importance quantification result, wherein the attribute importance score of each attribute is shown in table 1, and the attribute importance score of each attribute is shown in fig. 2.
Then, 80% of hard disks (525 hard disk life-span data) are selected for constructing a training set, and the rest 20% of hard disk life-span data (132 hard disk life-span data) are used as a test set for testing the effectiveness of the method. In order to more intuitively embody the generalization performance of the method, the whole life data of the hard disk of the test set are grouped according to the life of the hard disk as shown in table 2, one whole life data of the hard disk is randomly selected from each group of data, the method is used for health evaluation, and the results are visualized as shown in fig. 3 (a), (b), (c) and (d), respectively.
As can be seen from fig. 3, the method of the present invention has a high degree of health evaluation value close to the real health value, which indicates that the method of the present invention can accurately evaluate the health of the mechanical hard disk. To further verify the superiority of the present invention, the method of the present invention was compared with the optimized gradient enhancement method (XGBoost) and the long and short term memory network based method (LSTM), and the three methods were evaluated using a scoring function and a root mean square error prediction performance index, and the results (mean and variance) on four sets of hard disk data in the test set are shown in fig. 4.
As can be seen from FIG. 4, the two performance index values of the method are smaller than those of the other two prediction methods, which indicates that the mechanical hard disk of the method has higher accuracy of health evaluation, better stability and stronger robustness.
TABLE 1
Figure BDA0003998925330000171
TABLE 2
Figure BDA0003998925330000172
Through the mechanical hard disk health degree evaluation result and the performance comparison with the two methods, the method can be found that the accuracy of mechanical hard disk health degree evaluation is effectively improved and more excellent evaluation performance is obtained by modeling the time-varying dependence of each attribute of the SMART and fusing the correlation information of each attribute with the multi-scale time sequence characterization.
The invention also provides a mechanical hard disk health evaluation system 100 based on the evolutionary diagram, which is used for evaluating the health condition of the mechanical hard disk. As shown in fig. 5, the system includes:
the SMART data acquisition module 102 is configured to acquire an original data set of the plurality of mechanical hard disks according to full-life SMART data and corresponding health tags of the plurality of mechanical hard disks;
the target data acquisition module 104 is configured to obtain importance ranking of SMART attributes of the mechanical hard disk according to the original data set, obtain target SMART attributes meeting a preset ranking requirement according to the importance ranking, and obtain target SMART data of all the target SMART attributes and a corresponding target data set;
the evaluation model obtaining module 106 is configured to obtain a health evaluation model of the mechanical hard disk according to the multi-scale evolution graph network, the target SMART attributes, the target SMART data and the target data set, and from learning the correlation of evolution among the target SMART attributes;
The health evaluation module 108 is configured to evaluate the health of the mechanical hard disk to be detected according to the health evaluation model of the mechanical hard disk.
The mechanical hard disk health evaluation system 100 based on the evolutionary diagram in this embodiment corresponds to the mechanical hard disk health evaluation method based on the evolutionary diagram, and the function of each module in the mechanical hard disk health evaluation system 100 based on the evolutionary diagram in this embodiment is described in detail in the corresponding method embodiment, which is not described herein.
Furthermore, the invention proposes a computer-readable storage medium, in which computer-executable instructions are stored, which when being executed by a processor are adapted to carry out all or part of the method steps of the method for evaluating the health of a mechanical hard disk based on an evolutionary diagram as described above.
The present invention may be implemented by implementing all or part of the above-described method flow, or by instructing the relevant hardware by a computer program, which may be stored in a computer readable storage medium, and which when executed by a processor, may implement the steps of the above-described method embodiments. Wherein the computer program comprises computer program code, which may be in the form of source code, object code, executable files or in some intermediate form, etc. The computer readable medium may include: any entity or device capable of carrying computer program code, a recording medium, a U disk, a removable hard disk, a magnetic disk, an optical disk, a computer Memory, a Read-Only Memory (ROM), a random access Memory (RAM, random Access Memory), an electrical carrier signal, a telecommunications signal, a software distribution medium, and so forth. It should be noted that the content of the computer readable medium can be appropriately increased or decreased according to the requirements of the jurisdiction's jurisdiction and the patent practice, for example, in some jurisdictions, the computer readable medium does not include electrical carrier signals and telecommunication signals according to the jurisdiction and the patent practice.
Based on the same inventive concept, the embodiments of the present application further provide an electronic device, including a memory and a processor, where the memory stores a computer program running on the processor, and when the processor executes the computer program, the processor implements all or part of the method steps in the above method.
The processor may be a central processing unit (Central Processing Unit, CPU), other general purpose processors, digital signal processors (Digital Signal Processor, DSP), application specific integrated circuits (Application Specific Integrated Circuit, ASIC), off-the-shelf programmable gate arrays (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or the like. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like, the processor being a control center of the computer device, and the various interfaces and lines connecting the various parts of the overall computer device.
The memory may be used to store computer programs and/or models, and the processor implements various functions of the computer device by running or executing the computer programs and/or models stored in the memory, and invoking data stored in the memory. The memory may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required for at least one function (e.g., a sound playing function, an image playing function, etc.); the storage data area may store data (e.g., audio data, video data, etc.) created according to the use of the handset. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as a hard disk, memory, plug-in hard disk, smart Media Card (SMC), secure Digital (SD) Card, flash Card (Flash Card), at least one disk storage device, flash memory device, or other volatile solid-state storage device.
It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as a method, system, server, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, magnetic disk storage, optical storage, and the like) having computer-usable program code embodied therein.
The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), servers and computer program products according to embodiments of the invention. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
It will be apparent to those skilled in the art that various modifications and variations can be made to the present invention without departing from the spirit or scope of the invention. Thus, it is intended that the present invention also include such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.

Claims (10)

1. A mechanical hard disk health assessment method based on an evolutionary diagram, the method comprising:
acquiring original data sets of the mechanical hard disks according to the full-life SMART data of the mechanical hard disks and the corresponding full-life health degree labels;
obtaining importance degree sequencing of SMART attributes of the mechanical hard disk according to the original data set, obtaining target SMART attributes meeting preset sequencing requirements according to the importance degree sequencing, and obtaining target SMART data of all the target SMART attributes and corresponding target data sets;
according to the multiscale evolution graph network, the target SMART attributes, the target SMART data and the target data set, self-learning the evolutionary relativity among the target SMART attributes, and acquiring a health evaluation model of the mechanical hard disk;
and evaluating the health degree of the mechanical hard disk to be detected according to the health degree evaluation model of the mechanical hard disk.
2. The method for evaluating health of mechanical hard disks based on an evolutionary diagram according to claim 1, wherein the obtaining the raw data sets of the plurality of mechanical hard disks according to the full life SMART data and the corresponding full life health tags of the plurality of mechanical hard disks comprises:
collecting full-life SMART data when a plurality of mechanical hard disks run to faults, and constructing corresponding full-life health degree labels;
Combining the full-life SMART data and the full-life health degree label to obtain full-life original data of a plurality of mechanical hard disks;
and constructing a plurality of original data sets of the mechanical hard disk according to the life-span original data.
3. The method for evaluating health of a mechanical hard disk based on an evolutionary diagram according to claim 1, wherein the obtaining the importance ranking of SMART attributes of the mechanical hard disk according to the original dataset, obtaining the target SMART attributes meeting the preset ranking requirement according to the importance ranking, and obtaining the target SMART data and the corresponding target dataset of all the target SMART attributes includes:
constructing an initial integrated tree model, and acquiring a training data set serving as input data of the initial integrated tree model according to the original data set; wherein the training dataset is the number of each agent selected from the original dataset:
normalized values for the SMART attributes;
constructing a regression tree learning objective function, and carrying out iterative training on an initial integrated tree model according to a training data set to obtain a mature integrated tree model;
obtaining importance ranking of all SMART attributes of the mechanical hard disk according to the occurrence times of each SMART attribute in the mature integrated tree model;
Selecting a target SMART attribute meeting the preset ordering requirement from all SMART attributes according to the importance ordering;
and carrying out standardization processing according to the target SMART attribute to obtain target SMART data, obtaining a health degree label corresponding to the target SMART data according to the original data set, and combining the target SMART data and the health degree label to obtain a target data set with all the target SMART attribute.
4. The method for evaluating health of a mechanical hard disk based on an evolutionary diagram according to claim 1, wherein the obtaining a health evaluation model of a mechanical hard disk based on the evolutionary diagram network, the target SMART attributes, the target SMART data and the target data set, from learning the evolutionary correlations among the target SMART attributes, comprises:
constructing a multi-scale evolution graph network, and combining the target SMART attributes, the target SMART data and the target data set, and self-learning the evolutionary correlation among the target SMART attributes to extract multi-scale time sequence characterization of the target SMART data;
and carrying out iterative updating on the multi-scale evolutionary graph network according to the multi-scale time sequence representation of the target SMART data to obtain a health evaluation model of the mechanical hard disk.
5. The method of claim 4, wherein constructing the multi-scale evolutionary graph network, in combination with the target SMART attributes, the target SMART data, and the target data set, self-learns the evolutionary correlations between the target SMART attributes to extract the multi-scale timing characterization of the target SMART data, comprises:
Constructing a multi-scale evolutionary graph network, wherein the multi-scale evolutionary graph network comprises a time convolution module, an evolutionary graph self-learning module and an evolutionary graph convolution module;
inputting the target SMART data to a time convolution module, and extracting multi-scale time sequence characterization of the target SMART data along a time dimension;
taking the output data of the time convolution module as the input data of the evolution graph self-learning module, automatically extracting graph structure characterization of evolution among all target SMART attributes along with time, and outputting an evolved adjacency matrix;
and taking the adjacency matrix output by the evolution graph self-learning module and the multi-scale time sequence representation output by the time convolution module as input data of the evolution graph convolution module, and calculating to obtain a new multi-scale time sequence representation considering the correlation among all the target SMART attributes.
6. The method of claim 5, wherein the temporal convolution module comprises an expanded convolution layer and a gating layer;
the inputting the target SMART data to the time convolution module, extracting a multi-scale time sequence representation of the target SMART data along a time dimension, including:
dividing target SMART data into a plurality of time sequences according to SMART attributes, taking the time sequence of each target SMART attribute as input data of an expansion convolution layer, and performing expansion convolution operation by using convolution kernels with different sizes to obtain time sequence characterization with different time scales;
And taking the time sequence characterization of different time scales output by the expansion convolution layer as input data of the gating layer to obtain multi-scale time sequence characterization of each target SMART attribute output by the time convolution module.
7. The method for evaluating the health of a mechanical hard disk based on an evolutionary diagram according to claim 5, wherein the evolutionary diagram self-learning module comprises a diagram evolution stage division layer and a diagram self-learning layer;
the method for automatically extracting the graph structural representation of evolution among the target SMART attributes along with time by taking the output data of the time convolution module as the input data of the evolution graph self-learning module and outputting an evolved adjacency matrix comprises the following steps:
dividing the multi-scale time sequence representation of the output data of the time convolution module into a plurality of sections along the time dimension by a graph evolution stage dividing layer, and aggregating the time sequence representation of each section into one dimension by using an aggregation function to serve as the attribute representation of each evolution stage;
automatically learning the graph structure representation of each evolution stage according to the attribute representation of each evolution stage by a graph learning layer, wherein the graph structure representation of each evolution stage is obtained by the graph structure representation of the previous evolution stage;
and connecting the graph structure representations of the evolution stages in pairs, and deducing an adjacency matrix of each evolution stage by using a multi-layer perceptron so as to obtain the adjacency matrix evolved through the output of the evolution graph self-learning module.
8. The method for evaluating health of a mechanical hard disk based on an evolutionary diagram according to claim 5, wherein the computation obtains a new multi-scale timing representation considering correlation between each target SMART attribute by using the adjacency matrix output by the evolutionary diagram self-learning module and the multi-scale timing representation output by the time convolution module as input data of the evolutionary diagram convolution module, and the method comprises:
the method comprises the steps that an adjacency matrix output by a self-learning module of an evolutionary chart and a multi-scale time sequence representation output by a time convolution module are used as input data to be input into an evolutionary chart convolution module;
and carrying out graph convolution operation on the multi-scale time sequence representation of each target SMART attribute by utilizing the graph structure representation provided by the adjacent matrix through an evolutionary graph convolution module to obtain a new multi-scale time sequence representation considering the correlation among each target SMART attribute.
9. A mechanical hard disk health assessment system based on an evolutionary diagram, comprising:
the SMART data acquisition module is used for acquiring original data sets of the mechanical hard disks according to the full-life SMART data of the mechanical hard disks and the corresponding full-life health degree labels;
the target data acquisition module is used for obtaining importance ranking of SMART attributes of the mechanical hard disk according to the original data set, acquiring target SMART attributes meeting preset ranking requirements according to the importance ranking, and acquiring target SMART data of all the target SMART attributes and corresponding target data sets;
The evaluation model acquisition module is used for acquiring a health evaluation model of the mechanical hard disk according to the multi-scale evolution graph network, the target SMART attributes, the target SMART data and the target data set and by self-learning the correlation of evolution among the target SMART attributes;
and the health degree evaluation module is used for evaluating the health degree of the mechanical hard disk to be detected according to the health degree evaluation model of the mechanical hard disk.
10. A computer-readable storage medium, characterized in that the computer-readable storage medium has stored therein computer-executable instructions for implementing all or part of the method steps of the graph-based mechanical hard disk health assessment method according to claims 1-8 when executed by a processor.
CN202211611320.1A 2022-12-14 2022-12-14 Mechanical hard disk health assessment method, system and storage medium based on evolutionary diagram Pending CN116302870A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202211611320.1A CN116302870A (en) 2022-12-14 2022-12-14 Mechanical hard disk health assessment method, system and storage medium based on evolutionary diagram

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202211611320.1A CN116302870A (en) 2022-12-14 2022-12-14 Mechanical hard disk health assessment method, system and storage medium based on evolutionary diagram

Publications (1)

Publication Number Publication Date
CN116302870A true CN116302870A (en) 2023-06-23

Family

ID=86824665

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202211611320.1A Pending CN116302870A (en) 2022-12-14 2022-12-14 Mechanical hard disk health assessment method, system and storage medium based on evolutionary diagram

Country Status (1)

Country Link
CN (1) CN116302870A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117520104A (en) * 2024-01-08 2024-02-06 中国民航大学 System for predicting abnormal state of hard disk

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117520104A (en) * 2024-01-08 2024-02-06 中国民航大学 System for predicting abnormal state of hard disk
CN117520104B (en) * 2024-01-08 2024-03-29 中国民航大学 System for predicting abnormal state of hard disk

Similar Documents

Publication Publication Date Title
US10679751B2 (en) Cell abnormality diagnosis system using DNN learning, and diagnosis management method of the same
EP3292492B1 (en) Predicting likelihoods of conditions being satisfied using recurrent neural networks
TWI766618B (en) Key point detection method, electronic device and computer readable storage medium
CN107870845A (en) Towards the management method and system of micro services framework applications
EP3274888A1 (en) Analyzing health events using recurrent neural networks
US20070260563A1 (en) Method to continuously diagnose and model changes of real-valued streaming variables
KR20190113924A (en) Methods and devices for building scoring models and evaluating user credit
CN109948680B (en) Classification method and system for medical record data
WO2021056914A1 (en) Automatic modeling method and apparatus for object detection model
CN113010389A (en) Training method, fault prediction method, related device and equipment
WO2020172607A1 (en) Systems and methods for using deep learning to generate acuity scores for critically ill or injured patients
CN116302870A (en) Mechanical hard disk health assessment method, system and storage medium based on evolutionary diagram
CN115526641A (en) Flexible board product production quality tracing method, system, device and storage medium
CN110520702A (en) Monitor the heat health of electronic equipment
CN114528942A (en) Construction method of data sample library of engineering machinery, failure prediction method and engineering machinery
CN109698026B (en) Component identification in fault handling of medical devices
CN109934352B (en) Automatic evolution method of intelligent model
CN116166967A (en) Data processing method, equipment and storage medium based on meta learning and residual error network
CN109003680B (en) Epileptic data statistical method and device
Zamanzadeh et al. Autopopulus: a novel framework for autoencoder imputation on large clinical datasets
CN112395167A (en) Operation fault prediction method and device and electronic equipment
US11710068B2 (en) Labeling a dataset
CN117612714A (en) Sepsis characteristic variable measurement method, sepsis characteristic variable measurement device, terminal equipment and medium
CN114449701B (en) Light type regulation and control method and device based on medical record information and electronic equipment
CN112562849B (en) Clinical automatic diagnosis method and system based on hierarchical structure and co-occurrence structure

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination