Fault prediction method, device, equipment and storage medium

Abstract

The application discloses a fault prediction method and device, equipment and a storage medium. Wherein, the method comprises: determining the mutation point position corresponding to the mutated SMART attribute value in at least one self-monitoring, analysis and reporting technology (SMART) attribute value of the hard disk to be detected; determining the median of the mutation point position corresponding to the mutated SMART attribute value; based on the median of the mutation point position corresponding to the mutated SMART attribute value, at least one type of characteristic value is constructed; the at least one type of characteristic value is used to represent the mutation degree of the mutated SMART attribute value; based on the constructed at least one type of characteristic value, combined with the first model, the fault prediction result of the hard disk to be detected is determined; the first model is obtained by training the training sample data including each characteristic value in at least one type of characteristic value.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computers, and in particular to a fault prediction method and device, equipment and a storage medium. BACKGROUND

[0002] At present, among all server component failures, hard disk failures account for more than 70%, and if hard disk failure prediction can be achieved, the pressure of data operation and maintenance will be greatly reduced, and data migration can be carried out calmly, thereby reducing operation costs.

[0003] In related technologies, the commonly used hard disk failure prediction method is mainly to clean the Self-Monitoring Analysis and Reporting Technology (SMART) data collected periodically, and then directly apply clustering, random forest or logistic regression methods for fault analysis, but since the characteristics of the SMART data are not considered, the prediction accuracy is low and the universality is not strong. SUMMARY

[0004] To solve the technical problems in related technologies, the embodiments of the present application provide a fault prediction method, device, equipment and storage medium.

[0005] The technical solution of the embodiments of the present application is as follows:

[0006] The embodiments of the present application provide a fault prediction method, which comprises:

[0007] determining a mutation point position corresponding to a SMART attribute value that has mutated in at least one SMART attribute value of a hard disk to be detected;

[0008] determining a median value of the mutation point position corresponding to the SMART attribute value that has mutated;

[0009] constructing at least one type of feature value based on the median value of the mutation point position corresponding to the SMART attribute value that has mutated; the at least one type of feature value is used to represent the mutation degree of the SMART attribute value that has mutated;

[0010] determining a fault prediction result of the hard disk to be detected based on the at least one type of feature value constructed and in combination with a first model; the first model is obtained by training each feature value included in at least one type of feature value as training sample data.

[0011] In the above solution, the determination of the mutation point position corresponding to the SMART attribute value that has mutated in at least one SMART attribute value of the hard disk to be detected comprises:

[0012] determining, based on the historical sample data, a hard disk that has historically failed and a value of each SMART attribute of the hard disk that has historically failed;

[0013] detecting, based on the value of each SMART attribute of the hard disk that has historically failed, a mutation point corresponding to a SMART attribute value that has mutated in at least one SMART attribute value of the hard disk to be detected, to determine the mutation point corresponding to the SMART attribute value that has mutated;

[0014] determining, based on a length of a time sequence of the SMART attribute value of the hard disk that has historically failed, a number of the hard disks that have historically failed, and a number of the SMART attribute values of the hard disks that have historically failed, a position of the mutation point corresponding to the SMART attribute value that has mutated.

[0015] In the above scheme, when the median of the position of the mutation point corresponding to the SMART attribute value that has mutated is used to construct at least one type of feature value, the method comprises:

[0016] determining a main frequency of a Ricker wavelet based on a least common multiple of the median of the position of the mutation point corresponding to the SMART attribute value that has mutated;

[0017] determining a corresponding Ricker wavelet based on the main frequency of the Ricker wavelet;

[0018] determining a number of Ricker wavelet peaks of the time sequence of the SMART attribute value that has mutated based on a convolution result of the time sequence of the SMART attribute value that has mutated and the Ricker wavelet.

[0019] In the above scheme, when the median of the position of the mutation point corresponding to the SMART attribute value that has mutated is used to construct at least one type of feature value, the method comprises:

[0020] grouping the time sequence of the SMART attribute value that has mutated based on a least common multiple of the median of the position of the mutation point corresponding to the SMART attribute value that has mutated, to obtain at least two data groups; a length of each data group is less than or equal to the least common multiple;

[0021] determining squares of data in each data group for each data group; and summing the determined squares of the data to obtain a sum of squares of the corresponding data group;

[0022] determining a proportion of the sum of squares of the corresponding data group based on the sum of squares of the corresponding data group and a sum of squares of all data groups.

[0023] In the scheme, when the median of the mutation point position corresponding to the mutated SMART attribute value is used to construct at least one type of feature value, the method comprises:

[0024] The first interval is proportionally grouped according to the ratio of the least common multiple of the median of the mutation point position corresponding to the mutated SMART attribute value to the length of the time sequence of the mutated SMART attribute value, to obtain at least two subintervals.

[0025] According to each subinterval, data is intercepted from the time sequence of the mutated SMART attribute value to obtain a time subsequence of the mutated SMART attribute value corresponding to each subinterval.

[0026] The data in the time subsequence of the mutated SMART attribute value is subjected to mean value processing to obtain a mean value of the corresponding data group.

[0027] The mean value of the data group is subjected to variance processing to obtain a variance of the corresponding data group.

[0028] In the scheme, when the median of the mutation point position corresponding to the mutated SMART attribute value is used to construct at least one type of feature value, the method comprises:

[0029] The length of the time sequence of the mutated SMART attribute value is divided by the least common multiple of the median of the mutation point position corresponding to the mutated SMART attribute value.

[0030] The second interval is proportionally divided according to the division result to obtain at least two division results.

[0031] Based on the at least two division results and the time sequence of the mutated SMART attribute value, a data variance fluctuation flag is determined.

[0032] In the scheme, when the median of the mutation point position corresponding to the mutated SMART attribute value is used to construct at least one type of feature value, the method comprises:

[0033] The length of the time sequence of the mutated SMART attribute value is divided by the least common multiple of the median of the mutation point position corresponding to the mutated SMART attribute value.

[0034] The third interval is proportionally divided according to the division result to obtain at least two division results.

[0035] Based on the at least two division results and the time sequence of the mutated SMART attribute value, a data symmetry flag is determined.

[0036] In the scheme, when the median value of the mutation point position corresponding to the SMART attribute value based on the mutation is used to construct at least one type of feature value, the method comprises the following steps:

[0037] Based on the least square estimation strategy, the linear regression of the SMART attribute value time series with mutation is performed to obtain the corresponding linear regression result.

[0038] Based on the linear regression result, the overall linear trend feature of the data is determined.

[0039] In the scheme, the method further comprises the following steps:

[0040] The length of the SMART attribute value time series with mutation is determined.

[0041] The length of the determined SMART attribute value time series with mutation is optimized to obtain the optimal length of the SMART attribute value time series with mutation.

[0042] In the scheme, the length of the determined SMART attribute value time series with mutation is optimized to obtain the optimal length of the SMART attribute value time series with mutation, comprising the following steps:

[0043] The length of the SMART attribute value time series with mutation is determined. The length of the SMART attribute value time series with mutation is determined.

[0044] The maximum value of the score value of the fault prediction result of the hard disk to be detected is determined as the optimal length of the SMART attribute value time series with mutation.

[0045] In the scheme, the method further comprises the following steps: the at least one type of feature value is normalized to obtain the normalized at least one type of feature value.

[0046] The at least one type of feature value is normalized to obtain the normalized at least one type of feature value.

[0047] The normalized at least one type of feature value is input into the first model as input data to obtain the fault prediction result of the hard disk to be detected output by the first model; wherein the first model is a model for classification.

[0048] The application embodiment also provides a fault prediction device, the device comprises:

[0049] The first determining unit is configured to determine a mutation point position corresponding to a SMART attribute value that has mutated in at least one SMART attribute value of the hard disk to be detected.

[0050] The second determining unit is configured to determine a median value of the mutation point position corresponding to the SMART attribute value that has mutated.

[0051] The constructing unit is configured to construct at least one type of feature value based on the median value of the mutation point position corresponding to the SMART attribute value that has mutated; the at least one type of feature value is used to represent a mutation degree of the SMART attribute value that has mutated.

[0052] The third determining unit is configured to determine a failure prediction result of the hard disk to be detected based on the at least one type of feature value constructed and in combination with a first model; the first model is obtained by training each feature value included in at least one type of feature value as training sample data.

[0053] Embodiments of the present application further provide a failure prediction device, which comprises:

[0054] The processor is configured to determine a mutation point position corresponding to a SMART attribute value that has mutated in at least one SMART attribute value of the hard disk to be detected.

[0055] The median value of the mutation point position corresponding to the SMART attribute value that has mutated is determined.

[0056] At least one type of feature value is constructed based on the median value of the mutation point position corresponding to the SMART attribute value that has mutated; the at least one type of feature value is used to represent a mutation degree of the SMART attribute value that has mutated.

[0057] A failure prediction result of the hard disk to be detected is determined based on the at least one type of feature value constructed and in combination with a first model; the first model is obtained by training each feature value included in at least one type of feature value as training sample data.

[0058] Embodiments of the present application further provide another failure prediction device, which comprises a processor and a memory configured to store a computer program capable of running on the processor.

[0059] When the processor runs the computer program, the processor is configured to perform the steps of the failure prediction method provided by the embodiments of the present application.

[0060] The embodiment of the present application further provides a storage medium, which has a computer program stored thereon, and the computer program is executed by a processor to implement the steps of the failure prediction method provided by the embodiment of the present application.

[0061] The failure prediction method, device, equipment and storage medium provided by the embodiment of the present application determine a mutation point position corresponding to a SMART attribute value that has mutated in at least one SMART attribute value of a hard disk to be detected; determine a median value of the mutation point position corresponding to the SMART attribute value that has mutated; construct at least one type of feature value based on the median value of the mutation point position corresponding to the SMART attribute value that has mutated; the at least one type of feature value is used to represent a mutation degree of the SMART attribute value that has mutated; determine a failure prediction result of the hard disk to be detected based on the at least one type of feature value constructed and in combination with a first model; the first model is obtained by training each feature value included in at least one type of feature value as training sample data. By fully considering the characteristics of the mutation point of the SMART attribute value, the at least one type of feature value is reconstructed, and then the failure of the hard disk to be detected is predicted based on the at least one type of feature value constructed and in combination with the first model, so as to obtain the failure prediction result of the hard disk to be detected. In this way, the characteristics of the failed hard disk can be more accurately reflected, so as to effectively improve the accuracy and universality of the hard disk failure prediction. BRIEF DESCRIPTION OF DRAWINGS

[0062] Figure 1 A flowchart of a failure prediction method provided by the embodiment of the present application is shown in the figure;

[0063] Figure 2 A flowchart of another failure prediction method provided by the embodiment of the present application is shown in the figure;

[0064] Figure 3 A composition structure diagram of a failure prediction device provided by the embodiment of the present application is shown in the figure;

[0065] Figure 4 A hardware composition structure diagram of a failure prediction equipment provided by the embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical scheme and advantages of the embodiment of the present application more clear, the present application will be further described in detail below with reference to the drawings, and the described embodiments should not be regarded as limiting the present application. All other embodiments obtained by a person skilled in the art without making creative labor are within the scope of protection of the present application.

[0067] In the following description, "some embodiments" are referred to, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and the technical solutions described in the embodiments of the present application can be combined with each other without conflict.

[0068] The technical solutions of the present application will be described in further detail below in combination with the drawings and embodiments.

[0069] The present application provides a fault prediction method, which is applied to a server, Figure 1 A flowchart of a fault prediction method provided by the present application is shown in Figure 1 The fault prediction method comprises:

[0070] Step 101, determining a mutation point position corresponding to a mutated SMART attribute value in at least one SMART attribute value of a hard disk to be detected.

[0071] Step 102, determining a median value of the mutation point position corresponding to the mutated SMART attribute value.

[0072] In practical applications, the SMART of the hard disk specifies the standard that the hard disk manufacturer should follow, and is a data security technology commonly used in the hard disk at present. In the case that the hard disk is in a working state, the monitoring system of the server analyzes the state of the motor, circuit, disk and head of the hard disk, which is beneficial to predicting the hard disk failure and realizing relatively safe data protection.

[0073] In practical applications, for different brands and models of hard disks, the number and meaning of the SMART attribute values of the hard disk will be different, and are not completely consistent, but in general, the SMART attribute values shown in Table 1 below must be included in hard disks of various brands and models.

[0074]

[0075] Table 1

[0076] Here, the SMART attribute values of the hard disk to be detected include the original values of the SMART attribute values and the normalized values of the SMART attribute values, but in practical applications, since the normalization methods of different SMART attribute values by different hard disk manufacturers are not uniform, therefore, in order to simplify the implementation process of the fault prediction method of the present application, the original values of the SMART attribute values are used for analysis in the present application.

[0077] It should be noted that the brand and model of the hard disk to be detected are not limited in the present application, that is, the present application can be applied to fault prediction of hard disks of various brands and models.

[0078] In actual application, the server can first detect the mutation point corresponding to the SMART attribute value that has mutated in the at least one SMART attribute value of the hard disk to be detected based on the SMART attribute values of the hard disks that have historically failed by using the mutation point detection method, and then determine the position of the mutation point corresponding to the SMART attribute value that has mutated.

[0079] Based on this, in some embodiments, the determination of the position of the mutation point corresponding to the SMART attribute value that has mutated in the at least one SMART attribute value of the hard disk to be detected can be implemented by the following manner:

[0080] determining the hard disks that have historically failed based on the historical sample data, and the SMART attribute values of the hard disks that have historically failed;

[0081] detecting the mutation point corresponding to the SMART attribute value that has mutated in the at least one SMART attribute value of the hard disk to be detected based on the SMART attribute values of the hard disks that have historically failed, to determine the mutation point corresponding to the SMART attribute value that has mutated;

[0082] determining the position of the mutation point corresponding to the SMART attribute value that has mutated based on the length of the SMART attribute value time sequence of the hard disks that have historically failed, the number of the hard disks that have historically failed, and the number of the SMART attribute values of the hard disks that have historically failed.

[0083] In actual application, by analyzing the SMART attribute values of the hard disks that have historically failed, it is found that some SMART attribute values of the hard disk suddenly rise or fall within a certain period of time before the hard disk fails, indicating that the performance of the hard disk has changed significantly, which is a precursor to the failure of the hard disk. In addition, for a hard disk in good condition, the same analysis of the mutation of the SMART attribute values is also performed, and it is found that some SMART attribute values will also have mutations, but the mutation degree is different from that of the failed hard disk. Based on this, the mutation points of the SMART attribute values are analyzed in the embodiments of the present application, at least one type of feature value is extracted, and then the failure of the hard disk is classified and predicted.

[0084] Here, the SMART attribute values of the hard disks that have historically failed are matched with the at least one SMART attribute value of the hard disk to be detected, and the mutation point of the SMART attribute value with high matching degree is determined as the mutation point corresponding to the SMART attribute value that has mutated.

[0085] In actual application, the SMART attribute values of the hard disk are divided into training set data and test set data. For the training set data, the elements of the time series are sequentially added as sub-sequences in the SMART attribute value time series according to the position order, then the weighted sum result of all the sub-sequences is determined after sorting each sub-sequence, and finally the mutation point position corresponding to the mutated SMART attribute value is determined based on the weighted sum result of all the sub-sequences. The specific code implementation process is as follows:

[0086]

[0087] Here, the effective value (Pvalue) of the mutation point corresponding to the mutated SMART attribute value can be determined first, for example, the Pvalue can be determined by the following formula: Pvalue = exp (-6 * Y^2 / (L^3 + L^2)); then the determined effective value Pvalue is compared with the set threshold, and then it is determined whether the determined mutation point is effective according to the comparison result. For example, assuming that the set threshold is 0.05, if the Pvalue is less than 0.05, it is determined that the determined mutation point is an effective mutation point; if the Pvalue is greater than 0.05, it is determined that the determined mutation point is an ineffective mutation point.

[0088] It should be noted that the above process is repeated for each SMART attribute value of the hard disk that has historically failed, and then the positions of the mutation points of all the SMART attribute values of all the hard disks that have historically failed in the training set can be obtained. Next, the median of the array composed of the mutation point positions corresponding to all the mutated SMART attribute values is taken, that is, Y j = Median (Y ij ), if the current SMART attribute value does not have a mutation point, the SMART attribute value without the mutation point is deleted and does not enter the subsequent feature value construction process.

[0089] Step 103, constructing at least one type of feature value based on the median of the mutation point positions corresponding to the mutated SMART attribute values.

[0090] In the embodiments of the present application, the median of the mutation point positions corresponding to the mutated SMART attribute values can be used as an important parameter to construct at least one type of feature value. The at least one type of feature value can include but is not limited to: the number of Ricker wave peak values, the square sum proportion of data packets, the mean and variance of data packets, data variance fluctuation flag, data symmetry flag, data overall linear trend feature, etc. Here, the at least one type of feature value is used to represent the mutation degree of the mutated SMART attribute value.

[0091] In practical applications, since the SMART attribute values of the faulty hard disk and the good hard disk can both have mutation, in order to realize the fault prediction of the hard disk to be detected, the method of reconstructing the characteristic value is adopted in the embodiments of the present application to further distinguish the mutation degree of the SMART attribute value having mutation.

[0092] Here, according to the mutation degree and the duration of the mutation point, it can be determined whether the mutation of the SMART attribute value is of the faulty hard disk or of the good hard disk.

[0093] It should be noted that, in order to ensure the effectiveness of the SMART attribute value mutation analysis, in practical applications, the length of the SMART attribute value time sequence used to construct the characteristic value needs to meet the following condition: the length of the SMART attribute value time sequence used to construct the characteristic value must be greater than the median of the mutation point position corresponding to the SMART attribute value having mutation.

[0094] In some embodiments, based on the median of the mutation point position corresponding to the SMART attribute value having mutation, when constructing at least one type of characteristic value, the method comprises:

[0095] determining the main frequency of the Ricker wavelet based on the least common multiple of the median of the mutation point position corresponding to the SMART attribute value having mutation;

[0096] determining the corresponding Ricker wavelet based on the main frequency of the Ricker wavelet;

[0097] determining the Ricker wavelet peak value number of each SMART attribute value time sequence having mutation based on the convolution result of each SMART attribute value time sequence having mutation and the Ricker wavelet.

[0098] Here, the Ricker wavelet peak value number of each SMART attribute value time sequence having mutation is constructed as a type of characteristic value by referring to the commonly used Ricker wavelet in seismic wave analysis, so that the influence degree after the SMART attribute value mutation can be better analyzed.

[0099] Specifically, the minimum common multiple of the median of the mutation point positions corresponding to the mutated SMART attribute value is taken, assuming that the minimum common multiple is y, and the main frequency of the Ricker wavelet is f0, then the main frequency of the Ricker wavelet is determined based on the minimum common multiple, which can be obtained by the following formula: f0 = 1 / (n*y), wherein n is a positive integer, n*y<=L, L represents the total length of the SMART attribute value time sequence that has mutated; next, based on the main frequency of the Ricker wavelet, the corresponding Ricker wavelet is determined, and the expression of the Ricker wavelet is: R(t) = [1-2(πf0t) 2 ]exp(-(πf0t) 2 ), wherein R(t) represents the waveform of the Ricker wavelet, and t represents time; next, the SMART attribute value time sequence that has mutated is convolved with the Ricker wavelet to obtain the corresponding convolution result, and then the number of Ricker wavelet peaks is analyzed for the convolution result. In the process of determining the number of Ricker wavelet peaks, if it is determined through examination that it is a peak (maximum value) of the Ricker wavelet, the number of Ricker wavelet peaks is increased by 1, so that the number of Ricker wavelet peaks corresponding to each SMART attribute value time sequence that has mutated can be obtained as a characteristic value of the SMART attribute value time sequence that has mutated, and the number of Ricker wavelet peaks obtained each time a positive integer n is taken as a characteristic value, and there are n characteristic values.

[0100] In some embodiments, when the median of the mutation point positions corresponding to the mutated SMART attribute value is used to construct at least one type of characteristic value, the method comprises:

[0101] Grouping the SMART attribute value time sequence that has mutated based on the minimum common multiple of the median of the mutation point positions corresponding to the mutated SMART attribute value, to obtain at least two data groups; the length of each data group is less than or equal to the minimum common multiple;

[0102] For each data group, the square of each data in the corresponding data group is determined; and the squares of the determined data are summed to obtain the sum of squares of the corresponding data group;

[0103] Based on the sum of squares of the corresponding data group and the sum of squares of all data groups, the proportion of the sum of squares of the corresponding data group is determined.

[0104] Specifically, the minimum common multiple of the median of the mutation point positions corresponding to the mutated SMART attribute value is taken, assuming that the minimum common multiple is y, the mutated SMART attribute value time sequence is grouped, at least two data groups are obtained, the length of each data group is less than or equal to y, that is, the number of data groups is the total length of the mutated SMART attribute value time sequence divided by y, and when the remainder is not 0, the remainder corresponds to a data group. Assuming that n data groups are divided, for each data group, the squares of the data in the corresponding data group are calculated, the squares of the data are summed to obtain the sum of squares of the corresponding data group, and then the ratio of the sum of squares of each data group to the sum of squares of all data groups is calculated. For example, the mutated SMART attribute value time sequence is divided into n data groups, n characteristic values are formed, the ratio of the sum of squares of the first data group to the sum of squares of all data groups is the first characteristic value, and so on. The ratio of the sum of squares of the nth data group to the sum of squares of all data groups is the nth characteristic value.

[0105] In some embodiments, when the at least one type of characteristic value is constructed based on the median of the mutation point positions corresponding to the mutated SMART attribute value, the method comprises:

[0106] The first interval is proportionally grouped based on the ratio of the minimum common multiple of the median of the mutation point positions corresponding to the mutated SMART attribute value to the length of the mutated SMART attribute value time sequence, at least two subintervals are obtained;

[0107] Data corresponding to each subinterval is obtained by intercepting data from the mutated SMART attribute value time sequence according to each subinterval, and a mutated SMART attribute value time subsequence corresponding to each subinterval is obtained;

[0108] The data in the mutated SMART attribute value time subsequence is subjected to mean value processing to obtain the mean value of the corresponding data group;

[0109] The mean value of the data group is subjected to variance processing to obtain the variance of the corresponding data group.

[0110] Specifically, the minimum common multiple of the median of the mutation point positions corresponding to the mutated SMART attribute value is taken, assuming that the minimum common multiple is y, the length of the mutated SMART attribute value time sequence is L, and the first interval is [0, 1]. [0, 1] is divided into at least two subintervals according to y / L, for example, y / L is 0.2, and the at least two subintervals are divided into [0, 0.2], [0.2, 0.4], [0.4, 0.6], [0.6, 0.8], [0.8, 1], and the intervals are constructed according to the ascending traversal, that is:

[0111] [0,0.2], [0,0.4], [0,0.6], [0,0.8], [0,1]

[0112] [0.2,0.4], [0.2,0.6], [0.2,0.8], [0.2,1]

[0113] [0.4,0.6], [0.4,0.8], [0.4,1]

[0114] [0.6,0.8], [0.6,1]

[0115] [0.8,1]

[0116] Data is extracted from the SMART attribute value time series that has undergone a mutation according to each of the above subintervals to obtain the SMART attribute value time subsequences that have undergone a mutation corresponding to each subinterval. For example, the subsequence from the sequence [0, 0.2] to the position corresponding to 0.2 of the total length is taken. Then, the data in the SMART attribute value time subsequence that has undergone a mutation is averaged to obtain the mean of the corresponding data group, which serves as a eigenvalue. The mean of the data group is then subjected to variance processing to obtain the variance of the corresponding data group, which serves as another eigenvalue. Assuming that the number of subintervals is n, the number of eigenvalues ​​constructed in this way is 2n.

[0117] In some embodiments, when constructing at least one type of feature value based on the median value of the mutation point position corresponding to the SMART attribute value that has mutated, the method includes:

[0118] Determine the length of the time series of the SMART attribute value that has mutated, and the integer divisibility result of the least common multiple of the median of the mutation point position corresponding to the SMART attribute value that has mutated;

[0119] Based on the integer division result, the second interval is divided into equal proportions to obtain at least two equal division results;

[0120] Based on the at least two equally divided results and the SMART attribute value time series that undergoes a sudden change, a data variance fluctuation mark is determined.

[0121] Specifically, the minimum common multiple of the median of the mutation point positions corresponding to the mutated SMART attribute value is taken, assuming that the minimum common multiple is y, the length of the time series of the mutated SMART attribute value is L, assuming that the second interval is [0, 0.5], [0, 0.5] is equally divided according to L divided by y, and the result of the equal division is the value of r. For example, assuming that L / y is 0.1, the value of r is 0.1, 0.2, 0.3, 0.4, and 0.5. The data variance fluctuation flag is std(S), and the data variance fluctuation flag can be determined according to the formula std(S)>r*(max(S)-min(S)). The type of the data variance fluctuation flag is bool, wherein S is the time series of the mutated SMART attribute value, max(S) is the maximum value of the time series of the mutated SMART attribute value, and min(S) is the minimum value of the time series of the mutated SMART attribute value. In general, the number of data variance fluctuation flags is the same as the number of values of r.

[0122] In some embodiments, when the median of the mutation point positions corresponding to the mutated SMART attribute value is used to construct at least one type of feature value, the method comprises:

[0123] determining the length of the time series of the mutated SMART attribute value, and the integer division result of the minimum common multiple of the median of the mutation point positions corresponding to the mutated SMART attribute value;

[0124] based on the integer division result, equally dividing the third interval in proportion to obtain at least two equal division results;

[0125] based on the at least two equal division results and the time series of the mutated SMART attribute value, determining a data symmetry flag.

[0126] Specifically, the minimum common multiple of the median of the mutation point positions corresponding to the mutated SMART attribute value is taken, assuming that the minimum common multiple is y, the length of the time series of the mutated SMART attribute value is L, and the third interval is [0, W]. [0, W] is equally divided according to L divided by y, and the result of the equal division is the value of r. For example, assuming that L / y is 0.1 and W is 0.3, the value of r is 0.1, 0.2, and 0.3. The calculation method of the data symmetry flag is |mean(S)-median(S)|<r*(max(S)-min(S)), and the result is a bool type, where S is the time series of the mutated SMART attribute value, max(S) is the maximum value of the time series of the mutated SMART attribute value, min(S) is the minimum value of the time series of the mutated SMART attribute value, mean(S) is the mean of the time series of the mutated SMART attribute value, and median(S) is the median of the time series of the mutated SMART attribute value. If the above formula is true, the data symmetry requirement is met, and the data symmetry flag can be determined. In general, the number of data symmetry flags is the same as the number of values of r. The value of W depends on the brand and model of the hard disk to be detected, and corresponds to the maximum value of the difference between the mean and the median.

[0127] In some embodiments, when constructing at least one type of feature value based on the median of the mutation point positions corresponding to the mutated SMART attribute value, the method comprises:

[0128] Based on the least squares estimation strategy, the time series of the mutated SMART attribute value is linearly regressed to obtain the corresponding linear regression result;

[0129] Based on the linear regression result, the data overall linear trend feature is determined.

[0130] Specifically, the data overall linear trend feature can include intercept (intercept), slope (slope), rvalue (right value), and stderr (standard output). Assuming that the time series of the mutated SMART attribute value is represented by Y, and the timestamp of Y is taken as the independent variable X, the timestamp X of the time series of the mutated SMART attribute value is linearly regressed to obtain Y' = BX + A + ε, where ε represents an error. Using the least squares estimation strategy, B = (X T X) -1 X T Y', B represents the slope slope, A is the intercept intercept; rvalue = ∑(Y'-E(Y)) 2 / ∑(Y-Y') 2where E(Y) represents the result of taking the mean of Y; stderr = ∑(Y-Y') 2 (n-2), where n represents the length of the time series of the SMART attribute values that have mutated.

[0131] In step 104, based on the constructed at least one type of feature value, a first model is combined to determine a failure prediction result of the hard disk to be detected.

[0132] Here, the first model is obtained by training training sample data including each feature value in the at least one type of feature value.

[0133] In some embodiments, the determination of the failure prediction result of the hard disk to be detected based on the constructed at least one type of feature value and the first model can be achieved in the following manner:

[0134] The constructed at least one type of feature value is input as input data to the first model to obtain the failure prediction result of the hard disk to be detected output by the first model; and the first model is a model for classification.

[0135] In actual applications, in order to improve the calculation speed of the first model, the at least one type of feature value can be normalized to obtain normalized at least one type of feature value, and then the normalized at least one type of feature value is combined with the first model to predict the failure of the hard disk to be detected.

[0136] Based on this, in some embodiments, the method further includes normalizing the at least one type of feature value to obtain normalized at least one type of feature value.

[0137] Correspondingly, the determination of the failure prediction result of the hard disk to be detected based on the constructed at least one type of feature value and the first model can be achieved in the following manner:

[0138] The normalized at least one type of feature value is input as input data to the first model to obtain the failure prediction result of the hard disk to be detected output by the first model; and the first model is a model for classification.

[0139] Here, the normalization of the at least one type of feature value can be logarithmic normalization of the at least one type of feature value. Assuming that the normalized at least one type of feature value is y, y can be determined by the following formula: y = log10(x+1), where x represents the at least one type of feature value determined above.

[0140] In practical applications, the first model may be a random forest model, which is used to perform predictive analysis on the constructed at least one type of feature value to obtain a result of whether the hard disk to be detected fails within a prediction period.

[0141] The following is a set of prediction results corresponding to different sequence lengths. The code implementation process of fault prediction using the random forest model is as follows:

[0142]

[0143] In practical applications, since the length of the time series of SMART attribute values ​​that undergo mutations has a certain impact on the prediction results, in order to ensure the accuracy of the fault prediction results, the length of the time series of SMART attribute values ​​that undergo mutations needs to be optimized in the embodiments of the present application.

[0144] Based on this, in some embodiments, the method further includes:

[0145] Determine the length of the time series of SMART attribute values ​​where a sudden change occurs;

[0146] The determined length of the SMART attribute value time series that has undergone a mutation is optimized to obtain an optimal value of the length of the SMART attribute value time series that has undergone a mutation.

[0147] In actual application, the length of the determined SMART attribute value time series that has undergone a mutation is optimized to obtain the optimal length of the SMART attribute value time series that has undergone a mutation, including:

[0148] Performing interpolation processing on the length of the determined time series of the SMART attribute value that has undergone a sudden change to determine a score value of a fault prediction result of the hard disk to be detected;

[0149] The length of the SMART attribute value time series corresponding to the maximum value of the score value of the fault prediction result of the hard disk to be detected is determined as the optimal length of the SMART attribute value time series that has undergone a mutation.

[0150] Here, the length of the determined time sequence of the mutated SMART attribute value is interpolated, which can be cubic spline interpolation of the length of the time sequence of the mutated SMART attribute value. Specifically, according to the test set result, the score value F_score of the multiple prediction results and the SMART attribute value time sequence length L value are respectively constituted into an array, and then the cubic spline interpolation method is used to determine the score value F_score of the prediction result, and the expression is: F_score=a*L^3+b*L^2+c*L+d, where a, b, c, and d represent the coefficients of the cubic spline interpolation formula, and the L value corresponding to the maximum F_score value is selected as the optimal value of the length of the time sequence of the mutated SMART attribute value.

[0151] In practical applications, because the number of actual fault hard disks accounts for a small proportion of the total number of hard disks, when the number of positive and negative samples is unbalanced in the fault prediction process, it is easy to cause low prediction accuracy, therefore, the SMART attribute value samples of the fault hard disk can be expanded, D is the number of expanded fault disk samples, which is used to improve the accuracy of hard disk fault prediction. In addition, the advance prediction time of the hard disk fault state is less than or equal to D time units. For example, when D is 20, a fault hard disk sequence can be expanded to 20 equal-length sequences, and the prediction time needs to be less than or equal to 20 time units. When the sampling period is a day, the time unit of the prediction is a day.

[0152] The embodiment of the application further provides another fault prediction method, which is applied to a server, Figure 2 The flowchart of another fault prediction method provided by the embodiment of the application is shown in Figure 2 The fault prediction method comprises the following steps:

[0153] Step 201, determining a mutation point position corresponding to a mutated SMART attribute value in at least one SMART attribute value of a hard disk to be detected;

[0154] Step 202, determining a median value of the mutation point position corresponding to the mutated SMART attribute value;

[0155] Step 203, constructing at least one type of feature value based on the median value of the mutation point position corresponding to the mutated SMART attribute value;

[0156] Here, the at least one type of feature value is used to represent the mutation degree of the mutated SMART attribute value.

[0157] In the embodiments of the present application, the median of the mutation point position corresponding to the mutated SMART attribute value can be used as an important parameter to construct at least one type of feature value, which can include but is not limited to: the number of Ricker wavelet peaks, the square sum proportion of data packets, the mean and variance of data packets, data variance fluctuation flag, data symmetry flag, data overall linear trend feature, etc.

[0158] In practical applications, in order to ensure the effectiveness of the SMART attribute value mutation analysis, the length of the SMART attribute value time series used to construct the feature value in practical applications must satisfy the following condition: the length of the SMART attribute value time series used to construct the feature value must be greater than the median of the mutation point position corresponding to the mutated SMART attribute value.

[0159] Step 204, normalizing the at least one type of feature value to obtain normalized at least one type of feature value;

[0160] Step 205, inputting the normalized at least one type of feature value as input data into the first model to obtain the failure prediction result of the hard disk to be detected output by the first model.

[0161] Here, the first model is obtained by training each feature value included in the at least one type of feature value as training sample data.

[0162] Here, the first model is a model for classification, and specifically can be a random forest model, which is used to perform prediction analysis on the normalized at least one type of feature value to obtain the result of whether the hard disk to be detected fails within a prediction period.

[0163] It should be noted that the specific processing process of failure prediction has been described in detail above, and can be understood by referring to the failure prediction method described above, which will not be described here.

[0164] The scheme for fault prediction provided in the embodiments of the present application determines a mutation point position corresponding to a SMART attribute value that has mutated in at least one SMART attribute value of a hard disk to be detected; determines a median of the mutation point position corresponding to the SMART attribute value that has mutated; constructs at least one type of feature value based on the median of the mutation point position corresponding to the SMART attribute value that has mutated; the at least one type of feature value is used to represent a mutation degree of the SMART attribute value that has mutated; and a fault prediction result of the hard disk to be detected is determined based on the at least one type of feature value constructed and in combination with a first model; the first model is obtained by training each feature value included in at least one type of feature value as training sample data.

[0165] By adopting the scheme of the embodiments of the present application, the mutation point characteristics of the SMART attribute value are fully considered, at least one type of feature value is reconstructed, and then the fault of the hard disk to be detected is predicted based on the at least one type of feature value constructed and in combination with the first model, so as to obtain the fault prediction result of the hard disk to be detected, thus the characteristics of the fault disk can be more accurately reflected, and the accuracy and universality of the hard disk fault prediction can be effectively improved.

[0166] In order to implement the fault prediction method of the embodiments of the present application, the embodiments of the present application also provide a fault prediction device, Figure 3 The composition structure of the fault prediction device provided in the embodiments of the present application is shown in a schematic diagram as Figure 3 The fault prediction device comprises:

[0167] A first determination unit 31 is configured to determine a mutation point position corresponding to a SMART attribute value that has mutated in at least one SMART attribute value of a hard disk to be detected.

[0168] A second determination unit 32 is configured to determine a median of the mutation point position corresponding to the SMART attribute value that has mutated.

[0169] A construction unit 33 is configured to construct at least one type of feature value based on the median of the mutation point position corresponding to the SMART attribute value that has mutated; the at least one type of feature value is used to represent a mutation degree of the SMART attribute value that has mutated.

[0170] A third determination unit 34 is configured to determine a fault prediction result of the hard disk to be detected based on the at least one type of feature value constructed and in combination with a first model; the first model is obtained by training each feature value included in at least one type of feature value as training sample data.

[0171] In some embodiments, the first determining unit 31 is specifically configured to:

[0172] determine, based on the historical sample data, a hard disk that has historically failed and a value of each SMART attribute of the hard disk that has historically failed;

[0173] detect, based on the value of each SMART attribute of the hard disk that has historically failed, a mutation point corresponding to a mutated SMART attribute value of at least one SMART attribute value of the hard disk to be detected, to determine the mutation point corresponding to the mutated SMART attribute value;

[0174] determine, based on a length of a time sequence of the SMART attribute value of the hard disk that has historically failed, a number of the hard disks that have historically failed, and a number of the SMART attribute values of the hard disks that have historically failed, a position of the mutation point corresponding to the mutated SMART attribute value.

[0175] In some embodiments, when the constructing unit 33 constructs the at least one type of feature value based on the median of the position of the mutation point corresponding to the mutated SMART attribute value, the failure prediction apparatus further comprises:

[0176] a fourth determining unit configured to determine a main frequency of a Ricker wavelet based on a least common multiple of the median of the position of the mutation point corresponding to the mutated SMART attribute value;

[0177] determine, based on the main frequency of the Ricker wavelet, a corresponding Ricker wavelet;

[0178] determine, based on a convolution result of each time sequence of the mutated SMART attribute value and the Ricker wavelet, a number of Ricker wavelet peaks of each time sequence of the mutated SMART attribute value.

[0179] In some embodiments, when the constructing unit 33 constructs the at least one type of feature value based on the median of the position of the mutation point corresponding to the mutated SMART attribute value, the failure prediction apparatus further comprises:

[0180] a fifth determining unit configured to group, based on a least common multiple of the median of the position of the mutation point corresponding to the mutated SMART attribute value, the time sequence of the mutated SMART attribute value to obtain at least two data groups; a length of each data group is less than or equal to the least common multiple;

[0181] determine, for each data group, a square of each data in the corresponding data group; and sum the determined squares of the data to obtain a square sum of the corresponding data group;

[0182] Determine a square sum proportion of the corresponding data group based on the square sum of the corresponding data group and the square sum of all data groups.

[0183] In some embodiments, when the constructing unit 33 constructs the at least one type of feature value based on the median of the mutation point position corresponding to the mutated SMART attribute value, the fault prediction apparatus further comprises:

[0184] A sixth determining unit is configured to perform equal-proportion grouping on the first interval based on a ratio of a least common multiple of the median of the mutation point position corresponding to the mutated SMART attribute value to a length of the time sequence of the mutated SMART attribute value, to obtain at least two sub-intervals;

[0185] According to the sub-interval, data is intercepted from the time sequence of the mutated SMART attribute value to obtain a time sub-sequence of the mutated SMART attribute value corresponding to each sub-interval;

[0186] The data in the time sub-sequence of the mutated SMART attribute value is subjected to mean processing to obtain a mean value of the corresponding data group;

[0187] The mean value of the data group is subjected to variance processing to obtain a variance of the corresponding data group.

[0188] In some embodiments, when the constructing unit 33 constructs the at least one type of feature value based on the median of the mutation point position corresponding to the mutated SMART attribute value, the fault prediction apparatus further comprises:

[0189] A seventh determining unit is configured to determine a result of integer division of a length of the time sequence of the mutated SMART attribute value by a least common multiple of the median of the mutation point position corresponding to the mutated SMART attribute value;

[0190] Based on the result of integer division, a second interval is subjected to equal-proportion division to obtain at least two equal-division results;

[0191] Based on the at least two equal-division results and the time sequence of the mutated SMART attribute value, a data variance fluctuation flag is determined.

[0192] In some embodiments, when the constructing unit 33 constructs the at least one type of feature value based on the median of the mutation point position corresponding to the mutated SMART attribute value, the fault prediction apparatus further comprises:

[0193] An eighth determining unit is configured to determine a result of integer division of a length of the time sequence of the mutated SMART attribute value by a least common multiple of the median of the mutation point position corresponding to the mutated SMART attribute value;

[0194] based on the integer division result, performing equal proportion division on the third interval to obtain at least two equal division results;

[0195] based on the at least two equal division results and the time sequence of the mutated SMART attribute value, determining a data symmetry mark.

[0196] In some embodiments, when the construction unit 33 constructs at least one type of feature value based on the median of the mutation point position corresponding to the mutated SMART attribute value, the fault prediction apparatus further comprises:

[0197] a ninth determination unit configured to perform linear regression processing on the time sequence of the mutated SMART attribute value based on a least square estimation strategy to obtain a corresponding linear regression result;

[0198] based on the linear regression result, determining a data overall linear trend feature.

[0199] In some embodiments, the fault prediction apparatus further comprises:

[0200] a tenth determination unit configured to determine the length of the time sequence of the mutated SMART attribute value;

[0201] an optimization unit configured to perform optimization processing on the determined length of the time sequence of the mutated SMART attribute value to obtain an optimal value of the length of the time sequence of the mutated SMART attribute value.

[0202] In actual application, the optimization unit is specifically configured to:

[0203] perform interpolation processing on the determined length of the time sequence of the mutated SMART attribute value to determine a score value of the fault prediction result of the to-be-detected hard disk;

[0204] determine the length of the time sequence of the SMART attribute value corresponding to the maximum value of the score value of the fault prediction result of the to-be-detected hard disk as the optimal value of the length of the time sequence of the mutated SMART attribute value.

[0205] In some embodiments, the fault prediction apparatus further comprises:

[0206] a normalization unit configured to perform normalization processing on the at least one type of feature value to obtain normalized at least one type of feature value;

[0207] Correspondingly, the third determination unit 34 is specifically configured to:

[0208] Input the normalized at least one type of feature value as input data into the first model to obtain a failure prediction result of the hard disk to be detected output by the first model; wherein the first model is a model for classification.

[0209] Here, in actual application, the first determining unit 31, the second determining unit 32, the constructing unit 33 and the third determining unit 34 can be realized by a processor in the failure prediction device in combination with a communication interface.

[0210] It should be noted that the failure prediction device provided in the above embodiments is only exemplified by the division of the above program modules when performing failure prediction, and in actual application, the above processing can be completed by different program modules according to needs, that is, the internal structure of the device is divided into different program modules to complete all or part of the above-described processing. In addition, the failure prediction device and the failure prediction method provided in the above embodiments belong to the same concept, and the specific implementation process is detailed in the method embodiments, which will not be repeated here.

[0211] Based on the composition structure of the above program modules, and in order to realize the failure prediction method of the embodiments of the present application, the embodiments of the present application also provide a failure prediction device, Figure 4 The hardware composition structure diagram of the failure prediction device provided in the embodiments of the present application is shown in Figure 4 As shown in the figure, the failure prediction device 40 comprises:

[0212] The communication interface 41 is used to obtain at least one SMART attribute value of the hard disk to be detected.

[0213] The processor 42 is connected with the communication interface 41, and is used to execute the failure prediction method provided in one or more technical solutions when running a computer program. The computer program is stored on the memory 43.

[0214] Specifically, the processor 42 is used to determine the mutation point position corresponding to the SMART attribute value which has occurred mutation in the at least one SMART attribute value of the hard disk to be detected.

[0215] Determine the median of the mutation point position corresponding to the SMART attribute value which has occurred mutation.

[0216] Based on the median of the mutation point position corresponding to the SMART attribute value which has occurred mutation, at least one type of feature value is constructed; the at least one type of feature value is used to represent the mutation degree of the SMART attribute value which has occurred mutation.

[0217] determine, based on the at least one type of feature value and in combination with a first model, a failure prediction result of the hard disk to be detected; the first model is obtained by training each feature value included in the at least one type of feature value as training sample data.

[0218] In some embodiments, the processor 42 is specifically configured to:

[0219] determine, based on historical sample data, a hard disk that has historically failed and each SMART attribute value of the hard disk that has historically failed;

[0220] detect, based on each SMART attribute value of the hard disk that has historically failed, a mutation point corresponding to a mutated SMART attribute value of at least one SMART attribute value of the hard disk to be detected, to determine the mutation point corresponding to the mutated SMART attribute value;

[0221] determine, based on a length of a SMART attribute value time sequence of the hard disk that has historically failed, a number of hard disks that have historically failed, and a number of SMART attribute values of the hard disks that have historically failed, a mutation point position corresponding to the mutated SMART attribute value.

[0222] In some embodiments, the processor 42 is further configured to:

[0223] when constructing the at least one type of feature value based on the median of the mutation point position corresponding to the mutated SMART attribute value, determine a main frequency of a Ricker wavelet based on a least common multiple of the median of the mutation point position corresponding to the mutated SMART attribute value;

[0224] determine, based on the main frequency of the Ricker wavelet, a corresponding Ricker wavelet;

[0225] determine, based on a convolution result of each SMART attribute value time sequence that has mutated and the Ricker wavelet, a Ricker wavelet peak value number of the SMART attribute value time sequence that has mutated.

[0226] In some embodiments, the processor 42 is further configured to:

[0227] when constructing the at least one type of feature value based on the median of the mutation point position corresponding to the mutated SMART attribute value, group the SMART attribute value time sequence that has mutated based on a least common multiple of the median of the mutation point position corresponding to the mutated SMART attribute value, to obtain at least two data groups; a length of each data group is less than or equal to the least common multiple;

[0228] For each data packet, determine squares of data in the corresponding data packet; sum the determined squares of data to obtain a sum of squares of the corresponding data packet;

[0229] Based on the sum of squares of the corresponding data packet and the sum of squares of all data packets, determine a proportion of the sum of squares of the corresponding data packet.

[0230] In some embodiments, the processor 42 is further configured to:

[0231] In constructing the at least one type of feature value based on the median of the mutation point position corresponding to the SMART attribute value that has mutated, the first interval is proportionally grouped based on a ratio of a least common multiple of the median of the mutation point position corresponding to the SMART attribute value that has mutated to a length of the SMART attribute value time sequence that has mutated, to obtain at least two sub-intervals.

[0232] According to each sub-interval, data is intercepted from the SMART attribute value time sequence that has mutated to obtain a SMART attribute value time sub-sequence that has mutated corresponding to each sub-interval.

[0233] The data in the SMART attribute value time sub-sequence that has mutated is subjected to mean processing to obtain a mean value of the corresponding data packet.

[0234] The mean value of the data packet is subjected to variance processing to obtain a variance of the corresponding data packet.

[0235] In some embodiments, the processor 42 is further configured to:

[0236] In constructing the at least one type of feature value based on the median of the mutation point position corresponding to the SMART attribute value that has mutated, the length of the SMART attribute value time sequence that has mutated is determined to be an integer division result of a least common multiple of the median of the mutation point position corresponding to the SMART attribute value that has mutated.

[0237] Based on the integer division result, the second interval is proportionally divided to obtain at least two equal division results.

[0238] Based on the at least two equal division results and the SMART attribute value time sequence that has mutated, a data variance fluctuation flag is determined.

[0239] In some embodiments, the processor 42 is further configured to:

[0240] In constructing the at least one type of feature value based on the median of the mutation point position corresponding to the mutated SMART attribute value, the length of the time sequence of the mutated SMART attribute value is determined by the integer division result of the least common multiple of the median of the mutation point position corresponding to the mutated SMART attribute value.

[0241] Based on the integer division result, the third interval is equally divided to obtain at least two equal division results.

[0242] Based on the at least two equal division results and the time sequence of the mutated SMART attribute value, the data symmetry flag is determined.

[0243] In some embodiments, the processor 42 is further configured to:

[0244] In constructing the at least one type of feature value based on the median of the mutation point position corresponding to the mutated SMART attribute value, the time sequence of the mutated SMART attribute value is linearly regressed based on a least square estimation strategy to obtain a corresponding linear regression result.

[0245] Based on the linear regression result, the data overall linear trend feature is determined.

[0246] In some embodiments, the processor 42 is further configured to:

[0247] The length of the time sequence of the mutated SMART attribute value is determined.

[0248] The determined length of the time sequence of the mutated SMART attribute value is optimized to obtain an optimal length of the time sequence of the mutated SMART attribute value.

[0249] In actual application, the processor 42 is specifically configured to:

[0250] The length of the time sequence of the mutated SMART attribute value is determined.

[0251] The maximum value of the score value of the hard disk to be detected is determined as the optimal length of the time sequence of the mutated SMART attribute value.

[0252] In some embodiments, the processor 42 is further configured to:

[0253] The at least one type of feature value is normalized to obtain at least one type of normalized feature value.

[0254] Correspondingly, the processor 42 is specifically configured to:

[0255] Input the normalized at least one type of feature value as input data to the first model to obtain a failure prediction result of the hard disk to be detected output by the first model; wherein the first model is a model for classification.

[0256] It should be noted that the specific processing process of the communication interface 41 and the processor 42 will be described in the method embodiments, and will not be described here.

[0257] Of course, in actual application, various components in the failure prediction device 40 are coupled together through the bus system 44. It can be understood that the bus system 44 is used to realize the connection and communication between the components. The bus system 44 includes not only a data bus, but also a power bus, a control bus and a state signal bus. However, in order to clearly illustrate, all kinds of buses are marked as the bus system 44 in the Figure 4 .

[0258] The memory 43 in the embodiments of the present application is used to store various types of data to support the operation of the failure prediction device 40. Examples of these data include: any computer programs used to operate on the failure prediction device 40.

[0259] The failure prediction method disclosed in the above embodiments of the present application can be applied in the processor 42 or implemented by the processor 42. The processor 42 can be an integrated circuit chip with processing capability. In the implementation process, each step of the above method can be completed by the integrated logic circuit of hardware in the processor 42 or the instruction in the form of software. The processor 42 mentioned above can be a general processor, a digital signal processor (DSP), or other programmable logic device, discrete gate or transistor logic device, discrete hardware component, etc. The processor 42 can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present application. The general processor can be a microprocessor or any conventional processor, etc. In combination with the steps of the method disclosed in the embodiments of the present application, the hardware decoding processor can be directly embodied to execute the above-mentioned failure prediction method, or the hardware and software modules in the decoding processor are combined to execute the above-mentioned failure prediction method. The software module can be located in a storage medium, and the storage medium is located in the memory 43. The processor 42 reads the information in the memory 43 and combines the hardware to complete the steps of the above-mentioned failure prediction method.

[0260] In the exemplary embodiments, the failure prediction device 40 can be implemented by one or more of an Application Specific Integrated Circuit (ASIC), a DSP, a Programmable Logic Device (PLD), a Complex Programmable Logic Device (CPLD), a Field-Programmable Gate Array (FPGA), a general purpose processor, a controller, a Micro Controller Unit (MCU), a Microprocessor, or other electronic elements for executing the steps of the aforementioned failure prediction method.

[0261] It can be understood that the memory 43 of the embodiments of the present application can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. The non-volatile memory can be a Read Only Memory (ROM), a Programmable Read-Only Memory (PROM), an Erasable Programmable Read-Only Memory (EPROM), an Electrically Erasable Programmable Read-Only Memory (EEPROM), a Ferromagnetic Random Access Memory (FRAM), a Flash Memory, a magnetic surface memory, an optical disc, or a Compact Disc Read-Only Memory (CD-ROM). The magnetic surface memory can be a disk memory or a tape memory.

[0262] The volatile memory can be a random access memory (RAM) used as an external cache. By way of example, and not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), SyncLink dynamic random access memory (SLDRAM), and direct rambus random access memory (DRRAM). The memory described herein is intended to include, without being limited to, these and any other suitable types of memory.

[0263] In an example embodiment, the present embodiment also provides a storage medium, i.e., a computer storage medium, specifically a computer readable storage medium, such as a memory 43 storing a computer program, which can be executed by the processor 42 in the fault prediction device 40 to complete the steps of the aforementioned fault prediction method. The computer readable storage medium can be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disc or CD-ROM, etc.; or various devices including one or any combination of the above memories.

[0264] In the present embodiment, the terms "first", "second", etc. are only used to distinguish similar objects, and do not represent a specific order or sequence of the objects. It can be understood that the "first", "second", etc. can be interchanged in a specific order or sequence as allowed, so that the present embodiment described herein can be implemented in an order other than that illustrated or described herein.

[0265] The above description is only specific embodiments of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed by the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A fault prediction method, characterized in that: The method comprises: Determine a mutation point position corresponding to a mutation in at least one Self-Monitoring, Analysis, and Reporting Technology (SMART) attribute value of the hard disk to be tested; Determining the median value of the mutation point position corresponding to the SMART attribute value that has mutated; Constructing at least one type of characteristic value based on the median of the mutation point position corresponding to the mutated SMART attribute value; the at least one type of characteristic value is used to characterize the degree of mutation of the mutated SMART attribute value; Based on the constructed at least one type of characteristic value, combined with the first model, the fault prediction result of the hard disk to be detected is determined; the first model is obtained by training the training sample data using each characteristic value included in the at least one type of characteristic value as training sample data.

2. The method according to claim 1, characterized in that The determining of a mutation point position corresponding to a SMART attribute value that has undergone a mutation in at least one SMART attribute value of the hard disk to be detected includes: Determine hard disks that have experienced historical failures and the SMART attribute values ​​of the hard disks that have experienced historical failures based on historical sample data; Based on the SMART attribute values ​​of the hard disk that has historically failed, detecting a mutation point corresponding to a SMART attribute value that has mutated in at least one SMART attribute value of the hard disk to be detected, to determine the mutation point corresponding to the SMART attribute value that has mutated; Based on the length of the SMART attribute value time series of the hard disks that have failed in the past, the number of hard disks that have failed in the past, and the number of SMART attribute values ​​of the hard disks that have failed in the past, the mutation point position corresponding to the SMART attribute value that has mutated is determined.

3. The method according to claim 1, characterized in that When constructing at least one type of characteristic value based on the median value of the mutation point position corresponding to the SMART attribute value that has mutated, the method includes: Determining the main frequency of the Ricker wavelet based on the least common multiple of the median values ​​of the mutation point positions corresponding to the SMART attribute values ​​that have undergone mutation; Determining a corresponding Ricker wavelet based on a dominant frequency of the Ricker wavelet; Based on the convolution result of each SMART attribute value time series that has undergone mutation and the Ricker wavelet, the number of Ricker wavelet peaks of each SMART attribute value time series that has undergone mutation is determined.

4. The method according to claim 1, wherein When constructing at least one type of characteristic value based on the median value of the mutation point position corresponding to the SMART attribute value that has mutated, the method includes: Grouping the time series of SMART attribute values ​​that have mutated based on the least common multiple of the medians of the mutation point positions corresponding to the SMART attribute values ​​that have mutated to obtain at least two data groups; the length of each data group is less than or equal to the least common multiple; For each data group, determining the square of each data in the corresponding data group; summing the determined squares of each data to obtain a square sum of the corresponding data group; Based on the sum of squares of the corresponding data group and the sum of squares of all data groups, a proportion of the sum of squares of the corresponding data group is determined.

5. The method according to claim 1, wherein When constructing at least one type of characteristic value based on the median value of the mutation point position corresponding to the SMART attribute value that has mutated, the method includes: The first interval is divided into equal groups based on the ratio of the least common multiple of the median values ​​of the mutation point positions corresponding to the SMART attribute values ​​that have mutated to the length of the time series of the SMART attribute values ​​that have mutated, to obtain at least two sub-intervals; Data is intercepted from the time series of the SMART attribute value that has undergone mutation according to each sub-interval, so as to obtain the time sub-series of the SMART attribute value that has undergone mutation corresponding to each sub-interval; Performing mean processing on the data in the SMART attribute value time subsequence that has undergone the mutation to obtain the mean of the corresponding data group; Perform variance processing on the mean of the data group to obtain the variance of the corresponding data group.

6. The method according to claim 1, characterized in that When constructing at least one type of characteristic value based on the median value of the mutation point position corresponding to the SMART attribute value that has mutated, the method includes: Determine the length of the time series of the SMART attribute value that has mutated, and the integer divisibility result of the least common multiple of the median of the mutation point position corresponding to the SMART attribute value that has mutated; Based on the integer division result, the second interval is divided into equal proportions to obtain at least two equal division results; Based on the at least two equally divided results and the SMART attribute value time series that undergoes a sudden change, a data variance fluctuation mark is determined.

7. The method according to claim 1, characterized in that When constructing at least one type of characteristic value based on the median value of the mutation point position corresponding to the SMART attribute value that has mutated, the method includes: Determine the length of the time series of the SMART attribute value that has mutated, and the integer divisibility result of the least common multiple of the median of the mutation point position corresponding to the SMART attribute value that has mutated; Based on the integer division result, the third interval is divided into equal proportions to obtain at least two equal division results; A data symmetry flag is determined based on the at least two equally divided results and the SMART attribute value time series that undergoes a mutation.

8. The method according to claim 1, characterized in that When constructing at least one type of characteristic value based on the median value of the mutation point position corresponding to the SMART attribute value that has mutated, the method includes: Based on the least squares estimation strategy, linear regression processing is performed on the time series of SMART attribute values ​​that have undergone mutations to obtain the corresponding linear regression results; Based on the linear regression results, the overall linear trend characteristics of the data are determined.

9. The method according to claim 1, characterized in that The method further comprises: Determine the length of the time series of SMART attribute values ​​where a sudden change occurs; The determined length of the SMART attribute value time series that has undergone a mutation is optimized to obtain an optimal value of the length of the SMART attribute value time series that has undergone a mutation.

10. The method according to claim 9, characterized in that The optimizing process for the length of the determined SMART attribute value time series that has undergone a mutation to obtain the optimal length of the SMART attribute value time series that has undergone a mutation includes: Performing interpolation processing on the length of the determined time series of the SMART attribute value that has undergone a sudden change to determine a score value of a fault prediction result of the hard disk to be detected; The length of the SMART attribute value time series corresponding to the maximum value of the score value of the failure prediction result of the hard disk to be detected is determined as the optimal length of the SMART attribute value time series that has undergone a mutation.

11. The method according to claim 1, wherein The method further includes: normalizing the at least one type of eigenvalue to obtain a normalized eigenvalue of the at least one type; The determining of a fault prediction result of the hard disk to be detected based on the constructed at least one type of characteristic value and in combination with the first model includes: The normalized feature value of at least one type is input as input data to the first model to obtain the fault prediction result of the hard disk to be detected output by the first model; wherein the first model is a model for classification.

12. A fault prediction device, characterized in that: The device comprises: A first determining unit is configured to determine a mutation point position corresponding to a mutation in at least one Self-Monitoring, Analysis, and Reporting Technology (SMART) attribute value of the hard disk to be detected; A second determining unit is used to determine the median value of the mutation point position corresponding to the SMART attribute value that has mutated; a construction unit, configured to construct at least one type of characteristic value based on the median value of the mutation point position corresponding to the mutated SMART attribute value; the at least one type of characteristic value is used to characterize the degree of mutation of the mutated SMART attribute value; The third determination unit is used to determine the fault prediction result of the hard disk to be detected based on the constructed at least one type of characteristic value and in combination with the first model; the first model is obtained by training the training sample data using each characteristic value included in the at least one type of characteristic value as training sample data.

13. A fault prediction device, characterized in that: The device comprises: a processor, configured to determine a position of a mutation point corresponding to a mutation in at least one Self-Monitoring, Analysis, and Reporting Technology (SMART) attribute value of the hard disk to be detected; Determining the median value of the mutation point position corresponding to the SMART attribute value that has mutated; Constructing at least one type of characteristic value based on the median of the mutation point position corresponding to the mutated SMART attribute value; the at least one type of characteristic value is used to characterize the degree of mutation of the mutated SMART attribute value; Based on the constructed at least one type of characteristic value, combined with the first model, the fault prediction result of the hard disk to be detected is determined; the first model is obtained by training the training sample data using each characteristic value included in the at least one type of characteristic value as training sample data.

14. A fault prediction device, characterized in that: The apparatus comprises: a processor and a memory for storing a computer program capable of running on the processor; Wherein, when the processor is used to run the computer program, it executes the steps of the fault prediction method according to any one of claims 1 to 11.

15. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the fault prediction method according to any one of claims 1 to 11 are implemented.

Images