CN112527572A - Disk failure prediction method and device, computer readable storage medium and server - Google Patents
Disk failure prediction method and device, computer readable storage medium and server Download PDFInfo
- Publication number
- CN112527572A CN112527572A CN201910886471.XA CN201910886471A CN112527572A CN 112527572 A CN112527572 A CN 112527572A CN 201910886471 A CN201910886471 A CN 201910886471A CN 112527572 A CN112527572 A CN 112527572A
- Authority
- CN
- China
- Prior art keywords
- disk
- prediction
- data
- training
- parameter
- 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
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/22—Detection or location of defective computer hardware by testing during standby operation or during idle time, e.g. start-up testing
- G06F11/2205—Detection or location of defective computer hardware by testing during standby operation or during idle time, e.g. start-up testing using arrangements specific to the hardware being tested
- G06F11/2221—Detection or location of defective computer hardware by testing during standby operation or during idle time, e.g. start-up testing using arrangements specific to the hardware being tested to test input/output devices or peripheral units
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/22—Detection or location of defective computer hardware by testing during standby operation or during idle time, e.g. start-up testing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/22—Detection or location of defective computer hardware by testing during standby operation or during idle time, e.g. start-up testing
- G06F11/2263—Detection or location of defective computer hardware by testing during standby operation or during idle time, e.g. start-up testing using neural networks
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Hardware Design (AREA)
- Quality & Reliability (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Debugging And Monitoring (AREA)
Abstract
The invention discloses a disk failure prediction method, a disk failure prediction device, a computer readable storage medium and a server, wherein the method comprises the steps of obtaining disk operation state parameters to be analyzed, and analyzing the disk operation state parameters to obtain analyzed disk parameters; and constructing a prediction characteristic according to the analysis disk parameter, and performing prediction analysis on the prediction characteristic based on a neural network to obtain a prediction result. According to the method, the prediction characteristics are established through the disk running state parameters, the prediction results are obtained through prediction analysis of the prediction characteristics based on the neural network, the intelligent prediction of disk failure time can be realized, and the problems of time and labor waste and low efficiency existing in the current manual disk failure detection mode are solved.
Description
Technical Field
The invention relates to the technical field of computer storage systems, in particular to a disk failure prediction method, a disk failure prediction device, a computer readable storage medium and a server.
Background
With the rapid development of internet technology, the data storage demand is increasing, and internet users all over the world perform various interactions every day. To store these important user data, huge data centers throughout the world need to be built. Inside the data center, data are all stored on a storage medium, and the safety of data storage is a precondition for ensuring the safety of user data. Due to the limitation of economic factors, a large-scale data center still adopts a traditional storage system mainly comprising a disk, but frequent failures of the disk cause errors and losses of files and data in the disk, so that the service is abnormal, even basic service cannot be provided, and the user experience is greatly influenced. Therefore, predicting the failure time of a disk in order to replace the disk that is about to fail in time is particularly important.
At present, the judgment of the disk fault is mainly realized by operation and maintenance personnel in a manual detection mode so as to ensure the stable operation of a data center. However, for a large data center with millions of disks, the manual detection of disk failures requires a large amount of manpower and financial investment for the enterprise to perform field maintenance, and this is a huge workload for operation and maintenance personnel. In addition, intelligent operation and maintenance in a manual detection mode can only find and process disks which are disconnected or have performance which does not meet the requirements of the field. At this time, if the disk is damaged, the normal product function use of the user is affected if the disk is damaged, and the loss of user data is directly caused if the disk is damaged, which brings immeasurable loss to the operation service provider.
Disclosure of Invention
The embodiment of the invention provides a disk failure prediction method and device, a computer readable storage medium and a server, and solves the problems of time and labor waste and low efficiency in the current manual disk failure detection mode.
In a first aspect, a first embodiment of the present invention provides a disk failure prediction method, where the method includes obtaining a disk operating state parameter to be analyzed, and analyzing the disk operating state parameter to obtain an analyzed disk parameter;
and constructing a prediction characteristic according to the analysis disk parameter, and performing prediction analysis on the prediction characteristic based on a neural network to obtain a prediction result.
Optionally, before the obtaining of the disk running state parameter to be analyzed, the method further includes:
acquiring a disk SMART parameter and disk running I/O information from a server to be tested according to a set period, and uploading the disk SMART parameter and the I/O information to a designated server;
the acquiring of the disk running state parameters to be analyzed comprises acquiring the disk SMART parameters to be analyzed and the I/O information from the specified server.
Optionally, the analyzing the disk operating state parameter to obtain an analyzed disk parameter includes:
and analyzing the SMART parameters of the disk to obtain basic information of the disk, parameter data of the disk and information of the offline disk.
Optionally, before constructing the prediction feature according to the analysis disk parameter, the method further includes:
obtaining an original training sample set according to the disk parameter data and the basic information;
sorting the offline disc parameter data according to sampling time according to the offline disc information;
intercepting sorted offline disc parameter data based on a time sliding window;
and adding the intercepted offline parameter data into the original training sample set to obtain an expanded training sample set.
Optionally, after obtaining the extended training sample set, the method further includes:
determining training statistical characteristics of the disk based on the extended training sample set;
classifying the training statistical characteristics of the magnetic disk according to the offline disk information and marking the magnetic disk to obtain a training characteristic set;
and processing the training feature set to obtain a training data set, and training the neural network based on the training data set.
Optionally, the constructing a prediction feature according to the analysis disk parameter includes:
calculating the predicted statistical characteristics of the disk according to the SMART parameters of the disk and the I/O information of the disk operation;
and performing feature processing on the predicted statistical features to obtain the predicted features.
Optionally, the method further includes:
constructing an incremental training feature set according to the analyzed disk parameters;
incrementally training the neural network based on the incremental training feature set to update the neural network.
Optionally, the prediction method further includes:
and displaying the prediction result through a visual interface.
In a second aspect, a second embodiment of the present invention provides a disk failure prediction apparatus, including:
the magnetic disk data analysis module is used for acquiring magnetic disk running state parameters to be analyzed and analyzing the magnetic disk running state parameters to obtain analyzed magnetic disk parameters;
and the disk failure prediction module is used for constructing prediction characteristics according to the analyzed disk parameters and carrying out prediction analysis on the prediction characteristics based on a neural network to obtain a prediction result.
In a third aspect, a third embodiment of the present invention provides a computer-readable storage medium, on which an information delivery implementation program is stored, and the program, when executed by a processor, implements the steps of the disk failure prediction method as in the first embodiment.
In a fourth aspect, a fourth embodiment of the present invention provides a server, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the computer program is executed by the processor, the steps of the disk failure prediction method in the first embodiment are implemented.
According to the embodiment of the invention, the prediction characteristics are constructed through the disk running state parameters, the prediction results are obtained by performing prediction analysis on the prediction characteristics based on the neural network, the intelligent prediction of disk failure time can be realized, the problems of time and labor waste and low efficiency existing in the current manual disk failure detection mode are solved, and the positive technical effect is achieved.
Drawings
FIG. 1 is a schematic flow chart of an embodiment of the present invention;
FIG. 2 is a flowchart illustrating a training feature set acquisition process according to an embodiment of the present invention;
FIG. 3 is a flow chart of training a neural network according to an embodiment of the present invention;
FIG. 4 is a diagram illustrating bad disk data partitioning according to an embodiment of the present invention;
FIG. 5 is a diagram illustrating a neural network according to an embodiment of the present invention;
FIG. 6 is a schematic diagram of a disk failure time prediction apparatus according to an embodiment of the present invention;
FIG. 7 is a block diagram of a disk data collection module according to an embodiment of the present invention;
FIG. 8 is a schematic diagram of a model building and training module according to an embodiment of the present invention;
FIG. 9 is a block diagram of a disk failure prediction module according to an embodiment of the present invention;
FIG. 10 is a schematic diagram of an interface display module according to an embodiment of the invention;
FIG. 11 is a flow chart of an example of the present invention.
Detailed Description
Exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be embodied in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.
A first embodiment of the present invention provides a disk failure prediction method, as shown in fig. 1, the method including,
step S2: acquiring a disk running state parameter to be analyzed, and analyzing the disk running state parameter to obtain an analyzed disk parameter;
step S4: and constructing a prediction characteristic according to the analysis disk parameter, and performing prediction analysis on the prediction characteristic based on a neural network to obtain a prediction result.
According to the method, the prediction characteristics are constructed through the disk running state parameters, the prediction results are obtained through prediction analysis on the prediction characteristics based on the neural network, and intelligent prediction of disk failure time can be achieved; on the other hand, the occurrence time of the fault is predicted in advance, the occurrence of a part of faults is prevented, and the reliability of the data center can be greatly improved.
Optionally, in another implementation manner of this embodiment, before the obtaining of the disk running state parameter to be analyzed, the method further includes:
step S1: collecting a disk SMART parameter and disk running I/O information from a server to be tested according to a set period, and uploading the disk SMART parameter and the I/O information to a specified server.
The acquiring of the disk running state parameters to be analyzed comprises acquiring the disk SMART parameters to be analyzed and the I/O information from the specified server.
Specifically, in step S1, a disk SMART data acquisition script is deployed on each server to be tested in the previous data center, the data acquisition script acquires the SMART parameters of the disk from the server at regular time according to a set acquisition period, for example, the acquisition is triggered at regular time every hour, and the data is uploaded to a remote specified server, in addition to acquiring the SMART parameters of the disk, I/O information of the running of the disk is acquired in this embodiment, after the data acquisition is completed, the acquisition script can also automatically store the data in a local path, and store the data whose storage time exceeds a set threshold. The data uploading script can be triggered regularly every day, and the disk running state data acquired in the previous day is uploaded to the remote server.
And correspondingly acquiring the running state parameters of the disk to be analyzed, namely acquiring the SMART parameters of the disk and the running I/O information of the disk from a remote server.
Optionally, in another implementation manner of this embodiment, in step S2, the analyzing the disk operating state parameter to obtain an analyzed disk parameter includes:
and analyzing the SMART parameters of the disk to obtain basic information of the disk, parameter data of the disk and information of the offline disk.
Specifically, if the SMART parameter uploaded to the disk on the remote server in step S1 is raw data, in step S2, the running state parameter of the disk needs to be analyzed into a data format capable of performing statistical operations, and according to the requirement, the SMART parameter of the raw disk is analyzed into disk basic information, disk parameter data, and offline information in this embodiment.
Wherein, the basic information of the disk comprises basic parameters of the disk, including: basic information such as local points of the magnetic disks, hosts, host IP, serial numbers, models, types of the magnetic disks, acquisition time and the like;
the disk parameter data comprises SMART parameter values, serial number information and acquisition time of the disk;
the offline disc information comprises basic information of offline disc data, and the data analysis part is used for processing the offline disc to be damaged to train a model, so that the data of a fault disc does not need to be collected manually, and the automatic operation of the system is realized.
Optionally, in another implementation manner of this embodiment, step S3: training the neural network based on the disk operating state parameters, as shown in fig. 2, further includes:
step S31, obtaining an original training sample set according to the disk parameter data and the basic information;
step S32, sorting the parameter data of the wire dropping disks according to the information of the wire dropping disks and sampling time;
and step S33, intercepting the sorted offline parameter data based on the time sliding window and adding the intercepted offline parameter data into the original training sample set to obtain an expanded training sample set.
Specifically, the above steps are mainly performed to construct a training feature set, in this embodiment, in step S31, an original training sample set may be obtained by specifically processing a parameter data file of a disk according to a model training period and the information of the offline disk, and specifically, training set features may be divided into three types, which are, respectively, a data set 1 that predicts that a disk fails within 10 days, a data set 2 that predicts that a disk fails within one month, and a data set 3 that predicts a disk with a good state.
In this embodiment, all good disk data within 10 days of the current acquisition time are taken as raw data for predicting good disk state, and the mean, variance and range of each id of each disk are calculated as a training data set.
For the bad disc data, as shown in fig. 4, step S32 may sort each id data point of each bad disc in sequence according to sampling time, so as to obtain a time-line-based bad disc sequence.
As shown in fig. 4, in this example, the implementation of step S33 takes a time window with a size of 10 days as an example, and the data in the time window is intercepted to obtain the original data segment of the training sample. In order to obtain original data for predicting that a failure occurs in a disk within 10 days, a time window starting position is placed on the offline time of the disk, the data is intercepted and added into a training sample set 1, then the time window slides forwards for 2 days, the sliding is performed for 5 times, a data segment intercepted by moving the time window every time is counted into the training sample set 1, and through the sliding of the time sliding window, a sample number which is 5 times that of an original bad disk is constructed; similarly, in order to obtain the original data for predicting the failure of the disk within one month, firstly, a time window is placed at a position 10 days away from the disk dropping, the data segment is intercepted and placed in a sample set 2, then the time window is moved, the distance of each movement is 4 days, the movement is carried out for 5 times, the data segment intercepted by the 5 sliding time window is added into the sample set 2, and through the sliding of the time sliding window, the number of samples which is 5 times of the number of the original bad disks is constructed. Therefore, sample set 1 will be sample data for predicting a failure in the disk within 10 days, and sample set 2 will be sample data for predicting a failure in the disk within one month.
Therefore, the sorted data of the wire dropping discs are intercepted based on the time sliding window and added into the original training sample set to obtain an expanded training sample set, the data extraction mode in the embodiment is equivalent to the construction of the data volume which is 5 times of the original bad disc number, and the problem of unbalanced positive and negative sample numbers is relieved to a great extent.
Assuming that a data center has 10000 disks, and the annual failure rate of the disks in an actual field use environment is about three thousandths, the number of failed disks is 30. If a sliding window is not adopted, the number of samples of the original bad plate in 10 days is 30, the number of samples of the original bad plate in one month is also 30, the number of samples of the good plate is 9970, and the distribution of the three types of samples and the unbalance of the samples are obtained. If the model is trained by the data, the model prediction is biased to a healthy disk, and the prediction result is greatly influenced. Even if 10 data centers are collected and damaged, the number of fault samples in 10 days and one month of the disk is increased to 300, and the number of samples is still very small. After the sliding window is adopted, the number of fault samples in 10 days and one month of the disk is increased to 1500. And the sliding time length and the sliding times are modifiable, if the sliding distance is set to be small, the sliding times are increased, and the constructed sample data of the bad disk is more. Therefore, the method provided by the invention can be used for remarkably improving the number of fault samples in 10 days and one month of the training set disk.
Optionally, in another implementation manner of this embodiment, before constructing the predicted feature according to the parsed disk parameters, the method further includes, in step S3: training the neural network based on the disk operating state parameters, as shown in fig. 3, further includes the following steps:
step S34: determining training statistical characteristics of the disk based on the extended training sample set;
step S35: classifying the training statistical characteristics of the magnetic disk according to the offline disk information and marking the magnetic disk to obtain a training characteristic set;
step S36: and processing the training feature set to obtain a training data set, and training the neural network based on the training data set.
Specifically, step S34 can be implemented by calculating three statistical characteristics of the mean, variance and range of each ID of each disk using the disk basic information and the disk parameter data into which the SMART parameter of the original disk is parsed in the foregoing embodiment.
In this embodiment, statistical characteristics are further calculated by using the I/O information of the disk operation in the foregoing embodiment, and to perform disk failure time prediction, the training data need not only consider the SMART parameter of the disk, but also need to consider factors such as the I/O load of the disk. However, the acquisition script cannot directly acquire the Read/write speed of the disk, and only the current Read sector number (Sectors _ Read) and the current write sector number (Sectors _ write) of the disk can be acquired. Therefore, the current time period writing speed of the disk in this embodiment is calculated by the following formula:
the size of each sector of the disk is 512k, and the unit of time is milliseconds. Therefore, the above formula needs to be converted into units, and the calculation formula after conversion is as
Similarly, the current read-in data speed can be calculated as follows:
and calculating three statistical characteristics of the mean value, the variance and the range of the I/O of the disk according to the writing data speed and the reading data speed.
After obtaining the training statistical characteristics of the disks, step S35 may be implemented by classifying and labeling the characteristics of each disk according to the offline disk information file to obtain a training characteristic set.
Further, step S36 can be implemented by performing feature processing on the obtained training feature set to form a neural network training data set, then setting the neural network parameters, and inputting the training data set and the labels into the neural network for training.
On the basis of the foregoing embodiment, in the implementation process, the data of the last 10 days of the good disk can be intercepted and added to the sample set 3. The mean, variance and range of each data in each sample set were calculated separately. The method includes the steps that feature data are constructed from original data and are labeled, the feature data are further processed in the embodiment, feature dimension reduction is carried out through Principal Component Analysis (PCA), SMOTE up-sampling is carried out, and sample data of all classes are guaranteed to be balanced. The constructed features cannot be directly input into the neural network for training, the features need to be processed, the features can be classified according to the labels, and vectorization of the features and vectorization of the labels are performed so as to be input into the neural network for training.
Optionally, in another implementation manner of this embodiment, the constructing a prediction feature according to the parsed disk parameters includes:
calculating the predicted statistical characteristics of the disk according to the SMART parameters of the disk and the I/O information of the disk operation;
and performing feature processing on the predicted statistical features to obtain the predicted features.
Specifically, the method is similar to step S3, for example, the predicted statistical characteristics of the disk are calculated according to the collected SMART parameters of the disk and the I/O information of the disk operation, and the three statistical characteristics of the mean, the variance and the range of the collected data for ten days of each ID of each disk are calculated according to the basic information of the disk and the data of the disk parameters. Three statistical features of mean, variance and range of I/O of the disk are calculated according to the writing data speed and the reading data speed of the disk, which is different from the step S3 in that the features do not need to be classified and labeled (marked).
And further, processing the obtained predicted statistical characteristics, converting the predicted statistical characteristics into predicted characteristics of a data format which can be input into the neural network, then performing result prediction through the deep neural network, and storing the predicted results in a database.
In this embodiment, the neural network model is as shown in fig. 5, and the neural network model structure mainly includes an input layer, a hidden layer, and an output layer. The input layer mainly uses the input of the features, and the neuron number of the input layer is automatically adjusted according to the dimension of the input features. In order to better learn the different classes of features, the model is provided with 4 hidden layers in this embodiment. softmax regression is used to optimize the classification results as an additional processing layer to the output layer, transforming the neural network output into a probability distribution. In order to prevent overfitting of the neural network, in this embodiment, the weight of the neural network is initialized to a random value, a dropout regularization method is adopted in the mathematical calculation of each layer, and in addition, the sum of a cross entropy cost function and the L2 regularization of the weight can also be adopted as the cost function in this embodiment. When the model is trained, the weight value is updated by using a batch gradient descent method, and the exponential decay learning rate is used, so that the convergence speed is accelerated, and the situation that the model falls into local optimum can be prevented.
On the basis of the first embodiment, in a second embodiment of the present invention, the method may further include:
step S5: and constructing an incremental training feature set according to the analysis disk parameters, and performing incremental training on the neural network based on the incremental training feature set to update the neural network.
Specifically, the incremental training scheme is used for updating the neural network model periodically, for example, if a training period is set for performing update training on the neural network after the predicted statistical features are processed, the neural network model is updated periodically, and the accuracy of the neural network prediction model is ensured.
Optionally, the prediction method further includes:
step S6: and displaying the prediction result through a visual interface.
The scheme may specifically include that the parts of fault statistics, device list, disk list, user management, etc. are displayed separately.
By the technical scheme, the health conditions of all the disks of the data center can be monitored, the time of the disk about to fail can be predicted, the prediction result is displayed by combining a visual interface, an alarm is given to the disk about to fail within ten days, and field operation and maintenance personnel are reminded to replace the disk about to fail in time. Therefore, the workload of field operation and maintenance personnel is greatly liberated, the operation and maintenance cost of the data center is reduced, and the operation stability is improved.
A third embodiment of the present invention provides a disk failure prediction apparatus, as shown in fig. 6, the apparatus including:
the magnetic disk data analysis module is used for acquiring magnetic disk running state parameters to be analyzed and analyzing the magnetic disk running state parameters to obtain analyzed magnetic disk parameters;
and the disk failure prediction module is used for constructing prediction characteristics according to the analyzed disk parameters and carrying out prediction analysis on the prediction characteristics based on a neural network to obtain a prediction result.
The device of this embodiment can be very big liberation the work load of on-the-spot fortune dimension personnel, reduced data center's fortune dimension cost, improved operation stability.
Optionally, in another implementation manner of this embodiment, the apparatus further includes:
the disk data acquisition module is used for acquiring a disk SMART parameter and disk running I/O information from a server to be tested according to a set period and uploading the disk SMART parameter and the I/O information to a specified server;
the disk data analysis module is further configured to obtain the disk SMART parameter to be analyzed and the I/O information from the specified server.
Optionally, the disk data analysis module is further configured to analyze the disk SMART parameter to obtain disk basic information, disk parameter data, and offline disk information.
Specifically, in this embodiment, the disk data acquisition module: the data acquisition script of the module can be compiled by adopting a shell language and is deployed on a server to be tested, so that the function of automatically acquiring the SMART data of the magnetic disk is realized, the acquisition script is started at regular time by a timing function on the premise of not influencing the field service, the SMART data of the magnetic disk is acquired, and the acquired data is packaged and uploaded to a remote appointed server.
In this embodiment, as shown in fig. 7, the data collection script may be set to trigger once at a time of one hour, and the collection script may obtain the value of the disk attribute ID by reading the SMART system built in the disk. After data acquisition is finished, the acquisition script can automatically store the data under a local path, the storage time of the data exceeds three months, and the system can automatically delete the data to prevent the data from occupying too large storage space. The data uploading script is triggered once a day at regular time, and the running state data of the disk acquired in the previous day is uploaded to a remote server for prediction and training.
In the disk data analysis module, the original SMART data collected by the disk data collection module is a txt file, and the characteristics cannot be directly constructed, so that the disk data analysis module analyzes the collected data in this embodiment. According to the requirement, the original data is analyzed into three csv files, namely a disk basic information file, a disk parameter data file and a offline disk information file. The basic parameters of the disk stored in the disk basic information file comprise basic information such as a local point to which the disk belongs, a host IP, a serial number, a model number, a disk type, acquisition time and the like; the magnetic disc parameter data file stores SMART parameter values, serial number information and acquisition time of the magnetic disc; the offline disc information file is basic information for storing offline disc data, and in this embodiment, the disk data analysis module treats the offline disc as a damaged disc, so that it is unnecessary to manually collect data of a failed disc, and automatic operation of the system is realized.
Optionally, the model building and training module is configured to obtain an original training sample set according to parameter data and basic information of the disk before the disk failure prediction module builds the prediction feature according to the analysis disk parameter;
sorting the parameter data of the wire dropping disc according to sampling time according to the information of the wire dropping disc;
and intercepting the sorted offline parameter data based on a time sliding window and adding the intercepted offline parameter data into the original training sample set to obtain an expanded training sample set.
Specifically, the parameter data file of the disk can be processed according to the model training period and the offline disk information to obtain an original training sample set. As shown in fig. 8, in the training feature construction section, the training features are classified into three types in the present embodiment, and a data set 1 that predicts the occurrence of a failure in 10 days of the disk, a data set 2 that predicts the occurrence of a failure in one month of the disk, and a data set 3 of a good-condition disk have been constructed. The method takes all good disk data within 10 days of current acquisition time as original training data for predicting good disk state, and calculates the mean value, variance and extreme difference of each id of each disk as a training data set.
For bad disk data, the data set partitioning and processing flow is shown in fig. 4. And sequencing each id data point of each bad disk according to sampling time, as shown in fig. 4, setting a time window with the size of 10 days, and intercepting data in the time window to obtain an original data segment of a training sample. In order to obtain original data for predicting that a failure occurs in a disk within 10 days, a time window starting position is placed on the offline time of the disk, the data is intercepted and added into a training sample set 1, then the time window slides forwards for 2 days, the sliding is performed for 5 times, a data segment intercepted by moving the time window every time is counted into the training sample set 1, and through the sliding of the time sliding window, a sample number which is 5 times that of an original bad disk is constructed; similarly, in order to obtain the original data for predicting the failure of the disk within one month, firstly, a time window is placed at a position 10 days away from the disk dropping, the data segment is intercepted and placed in a sample set 2, then the time window is moved, the distance of each movement is 4 days, the movement is carried out for 5 times, the data segment intercepted by the 5 sliding time window is added into the sample set 2, and through the sliding of the time sliding window, the number of samples which is 5 times of the number of the original bad disks is constructed. Therefore, sample set 1 will be sample data for predicting a failure in the disk within 10 days, and sample set 2 will be sample data for predicting a failure in the disk within one month.
Therefore, the sorted data of the wire dropping discs are intercepted based on the time sliding window and added into the original training sample set to obtain an expanded training sample set, the data extraction mode of the embodiment is equivalent to the construction of the data volume which is 5 times of the original bad disc number, and the problem of unbalanced positive and negative sample numbers is relieved to a great extent.
Optionally, in another implementation manner of this embodiment, the apparatus further includes:
the model building and training module is used for determining the training statistical characteristics of the disk based on the extended training sample set;
classifying the training statistical characteristics of the magnetic disk according to the offline disk information and marking the magnetic disk to obtain a training characteristic set;
and processing the training feature set to obtain a training data set, and training the neural network based on the training data set.
Specifically, the main process of the model construction and training module is as follows: as shown in fig. 8, first, three file data obtained by analyzing by using a disk data analysis module are used, three statistical characteristics of a mean value, a variance, and a range of each ID of each disk are respectively calculated according to a disk parameter data file and a disk basic information file, and the characteristics of each disk are classified and labeled according to a offline disk information file; secondly, performing characteristic processing to form neural network training data; and finally, setting parameters of the neural network, and inputting the training data and the labels into the neural network for training.
I/O load calculation
For predicting the disk failure time, the training data not only needs to consider the SMART parameters of the disk, but also needs to consider factors such as the I/O load of the disk and the like. However, the acquisition script cannot directly acquire the Read/write speed of the disk, and only the current Read sector number (Sectors _ Read) and the current write sector number (Sectors _ write) of the disk can be acquired. Therefore, the current time period writing speed of the disk can be calculated by the following formula:
the size of each sector of the disk is 512k, and the unit of time is milliseconds. Therefore, the above equation needs to be converted in units, and the calculation formula after conversion is:
similarly, the current read-in data speed can be calculated as follows:
from this, the average I/O read/write rate over this time interval can be calculated.
On the basis of the foregoing embodiment, the data of the last 10 days of the good disc is intercepted and added to the sample set 3. The mean, variance and range of each data in each sample set were calculated separately. After the feature data is constructed by the original data and the label is attached, the feature data can be further processed, feature dimensionality reduction is carried out through Principal Component Analysis (PCA), SMOTE up-sampling is carried out, and sample data of all classes are guaranteed to be balanced. The constructed features cannot be directly input into the neural network for training, the features need to be processed, the features can be classified according to the labels, and vectorization of the features and vectorization of the labels are performed so as to be input into the neural network for training.
Optionally, the disk failure prediction module is further configured to:
calculating the predicted statistical characteristics of the disk according to the SMART parameters of the disk and the I/O information of the disk operation;
and performing feature processing on the predicted statistical features to obtain the predicted features.
As shown in fig. 9, in this embodiment, the disk failure prediction module may be written in python language, and includes three parts: the device comprises a prediction feature construction part, a prediction feature processing part and a deep neural network prediction part. The process is that a magnetic disk data analysis module is started at regular time to analyze original data into three csv files, namely a magnetic disk basic information file, a magnetic disk parameter data file and a offline disk information file. And the prediction characteristic construction part loads a disk basic information file and a parameter data file, and calculates the mean value, the variance and the extreme difference of data acquired by each id of each disk within ten days similarly to the model construction and training module process. Unlike the model building and training module flow, there is no need to classify and label features. The prediction characteristic processing part loads the original characteristics of the disk and carries out characteristic processing to convert the original characteristics into a data format which can be input into a neural network. And finally, loading the neural network model by the deep neural network prediction part to perform result prediction, and storing the prediction result in a database.
On the basis of the third embodiment, in a fourth embodiment of the present invention, the model building and training module is further configured to build an incremental training feature set according to the analytic disk parameter;
incrementally training the neural network based on the incremental training feature set to update the neural network.
In order to make the neural network perfect and update continuously, the embodiment is also provided with a neural network increment training part, when the network is trained again, the system starts the increment training part on the basis of the original model, so that the neural network also learns and updates synchronously along with the increase of the data volume, and compared with the mode of retraining again, the increment training greatly reduces the training time.
Optionally, in another embodiment of this embodiment, the apparatus further comprises,
and the interface display module is used for displaying the prediction result through a visual interface.
The interface display module is used for visualizing the prediction result and intelligently managing the disk, the prediction result of the disk failure time is always displayed in a visual interface mode through the embodiment, the visual interface greatly improves the user experience brought by the disk failure time prediction system, and the user can conveniently monitor the health state of the disk on site in real time.
After the interface display script is started, a user name and a password are required to log in, the interface display module firstly reads a stored prediction result from the database, and counts faults within ten days, faults within one month, the number of well-operated disks and the number of undetected disks.
The interface consists of a plurality of management modules: 1) a main interface module: and counting the total number of the equipment, the total number of the disks, the number of the fault equipment in ten days, the number of the fault equipment in one month and the number of the equipment which runs well in the current data center. And counting faults within ten days, faults within one month, the number of well-operated disks and the number of undetected disks, and displaying the statistical value and the occupation ratio of the disks by using a histogram and a sector diagram.
2) A device list module: the section counts the disk information mounted on each server device and displays the health status.
3) Disk list part: the part will count the information of all disks under the data center and display the health status of the disks in a list form.
4) A user management section: the interface display module is internally provided with a user management system which is responsible for adding, deleting and authorizing users at all levels.
Specifically, the interface display module in this embodiment: as shown in fig. 10, the module is an interface for a user to interact with the system, and includes four parts, namely, failure statistics, an equipment list, a disk list, and user management, and the functions of the four parts are described below.
A fault statistics module: and starting an interface display module, wherein the module can automatically acquire data from a database for storing the prediction result and display the prediction result on an interface. The system display module displays the current health state and the future health state of the disk in a period of time through various forms such as reports, sector graphs and bar charts, the time of future failure of the disk to be failed is displayed by different colors, red represents that the disk fails within 10 days, yellow represents that the disk fails within 10 days to one month, green represents that the running state is good, and gray represents that the current data center of the disk does not meet the prediction conditions or the number of acquisition days does not meet 10 days.
A device list module: the section counts the disk information mounted on each server device and displays the health status. In addition, the fuzzy query and screening function is added to the part, and a user can screen required data according to requirements.
A disk list module: the part will count the information of all disks under the data center and display the health status of the disks in a list form. The part also comprises fuzzy query and screening functions, and in addition, the disk information export function is supported, and the export can be selected as a csv/. png/. excel file so as to meet the use requirements under different use scenes.
A user management module: the interface display module is internally provided with a user management system which is responsible for adding, deleting and authorizing users at all levels, and only authorized legal users can log in the system after inputting correct passwords, so that the health state information of the operation of the disk of the data center can be checked. The information submitted by the user includes, but is not limited to, name, mailbox, phone, etc. Through the user management system, the safety of the data center is improved.
A fifth embodiment of the present invention provides a computer-readable storage medium, and the flow of disk failure prediction stored and executed in the computer storage medium of this embodiment is the same as that of the first or second embodiment. In terms of engineering implementation, the embodiment may be implemented by software plus a necessary general hardware platform, and certainly may also be implemented by hardware, but the former is a better implementation mode in many cases. With this understanding, the flow of disk failure prediction of the present invention can be embodied in the form of a computer software product stored in a computer storage medium (e.g., ROM/RAM, magnetic disk, optical disk) and including instructions for causing a device (e.g., wireless router, server, etc.) to execute the flow of disk failure prediction of the first or second embodiment of the present invention.
A sixth embodiment of the present invention provides a server, including a memory, a processor, and a computer program stored on the memory and operable on the processor, where the computer program, when executed by the processor, implements the steps of the disk failure prediction method of the first or second embodiment.
The following describes the scheme of the present invention with reference to a specific example, as shown in fig. 11, a disk failure prediction process includes a disk data acquisition module and a disk failure prediction module;
the disk data acquisition module firstly deploys an acquisition script on a server of a data center, wherein the acquisition script consists of two parts: a data acquisition part and a data uploading part;
the data acquisition part script is set to be triggered at regular intervals of one hour, the acquisition script can obtain the value of the disk attribute ID and the I/O information of the disk operation by reading a built-in SMART system in the disk, and the acquired SMART information comprises the following specific steps:
ID 001: bottom layer data Read Error Rate (Raw Read Error Rate)
ID 003: spindle Spin-Up Time Spin Up Time
ID 005: remapped sector Count Reallocated Sectors Count/Retired Block Count,
ID 007: seek Error Rate Seek Error Rate
ID 193: magnetic head loading/unloading counting Load/Unload Cycle Count
ID 197: current Pending mapping Sector Count
ID 198: offline Uncorrectable Sector Count Offline unscramble Sector Count
ID 199: ultra ATA CRC Error Rate
The I/O information of the disk operation comprises:
ID 503: current Read sector count Sectors _ Read
ID 507: current write sector count sector _ Written
ID 509: I/O cumulative read-write time I/O _ Spend _ ms
ID 512: current _ time _ ms of the Current time
The acquired original data of the disk is stored locally, then a data uploading part script is triggered every other day at regular time, and the running state data of the disk acquired the previous day is uploaded to a remote server.
The flow chart now has two branches, as shown in FIG. 11: the model building and training module and the disk failure prediction module are used for predicting the failure of the disk; the model in this example is constructed and trained in advance according to the scheme of the model construction and training module, and is placed in the system. In this example, the user can regularly start the training model to perform migration learning every 3 months when the module is used, so that the system model can continuously perform self-learning and self-updating according to the pre-created environment and the fault disk to adapt to a new use scene, and the disk fault prediction module is regularly started once a day.
The disk failure prediction module is performed first:
and (4) regularly triggering and starting a disk SMART data analysis module every other day for the construction of the prediction characteristics, analyzing the original data of the previous day into three csv files, namely a disk basic information file, a disk parameter data file and a offline disk information file, and storing the csv files locally.
And calculating the I/O data value of the disk according to the disk parameter data file. The mean, variance and range of each acquisition ID for each disk within the nearest 10 skylight opening were then calculated.
Specifically, the mean, variance, range of value and raw _ value of ID 001; variance, range of value and raw _ value of ID 003; the mean, variance, range of value and raw _ value of ID 005; the mean, variance, range of value and raw _ value of ID 007; the mean, variance, range of value and raw _ value of ID 193; mean, variance, range of raw _ value of ID 197; raw _ value mean, variance, range of ID 198; mean, variance, range of raw _ value of ID 199; mean, variance, range of raw _ value values of ID 510; the mean, variance and range of I/O read-write speed; thus, the dimension of the original feature data is 46 dimensions.
The predictive feature processing module performs feature processing on the raw features into a data format that can be input into a neural network input terminal.
The neural network prediction module loads the trained model, inputs the characteristic data into the network to obtain a prediction result, and stores the result in the database.
And the user interface display module reads the latest saved table in the database and displays the running state information of the disk on the interface.
As shown in fig. 11, the model building and training module starts the transfer learning periodically every 3 months, and first, the SMART data parsing module is performed to parse the original data into three csv files and add the three csv files to the local file.
The training feature construction part reads three local csv files, firstly calculates I/O load according to the scheme, secondly carries out data set division and feature construction on the analysis data, and finally calculates the mean value, variance and range of each sample in the data set and labels, wherein the training feature dimension is 46 dimensions.
The training characteristic processing part constructs characteristic data from the original data and labels are attached to the characteristic data, the characteristic data needs to be processed, characteristic dimension reduction is carried out through Principal Component Analysis (PCA), and the dimension after dimension reduction is 42 dimensions. And then SMOTE upsampling is carried out to ensure the sample data of each category to be balanced.
At this time, whether incremental learning is performed needs to be judged:
if no model exists in the system, if the selection is not yes, firstly constructing model parameters which comprise an input layer, four hidden layers, an output layer and the like, and in addition, other hyper-parameters needing to be configured with the neural network comprise: selecting tanh as an activation function, selecting the sum of L2 regular and cross entropy cost functions as a loss function, selecting an exponential decay learning rate as a learning rate, adopting a dropout regularization optimization method and a batch gradient descent method as a learning algorithm, setting iteration times, training a model by using a training sample set, and storing the model.
It should be noted that, in this document, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.
The above-mentioned serial numbers of the embodiments of the present invention are merely for description and do not represent the merits of the embodiments.
Through the above description of the embodiments, those skilled in the art will clearly understand that the method of the above embodiments can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware, but in many cases, the former is a better implementation manner. Based on such understanding, the technical solutions of the present invention may be embodied in the form of a software product, which is stored in a storage medium (such as ROM/RAM, magnetic disk, optical disk) and includes instructions for enabling a terminal (such as a mobile phone, a computer, a server, an air conditioner, or a network device) to execute the method according to the embodiments of the present invention.
While the present invention has been described with reference to the embodiments shown in the drawings, the present invention is not limited to the embodiments, which are illustrative and not restrictive, and it will be apparent to those skilled in the art that various changes and modifications can be made therein without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (11)
1. A disk failure prediction method, characterized in that the method comprises,
acquiring a disk running state parameter to be analyzed, and analyzing the disk running state parameter to obtain an analyzed disk parameter;
and constructing a prediction characteristic according to the analysis disk parameter, and analyzing the prediction characteristic based on a neural network to obtain a prediction result.
2. The prediction method according to claim 1, wherein before the obtaining the disk operating state parameter to be resolved, the method further comprises:
acquiring a disk SMART parameter and disk running I/O information from a server to be tested according to a set period, and uploading the disk SMART parameter and the I/O information to a designated server;
the acquiring of the disk running state parameters to be analyzed comprises acquiring the disk SMART parameters to be analyzed and the I/O information from the specified server.
3. The prediction method of claim 2, wherein the analyzing the disk operating state parameter to obtain an analyzed disk parameter comprises:
and analyzing the SMART parameters of the disk to obtain basic information of the disk, parameter data of the disk and information of the offline disk.
4. The prediction method of claim 3, wherein before constructing the predicted features based on the parsed disk parameters, the method further comprises:
obtaining an original training sample set according to the disk parameter data and the basic information;
sorting the offline disc parameter data according to sampling time according to the offline disc information;
intercepting sorted offline disc parameter data based on a time sliding window;
and adding the intercepted offline parameter data into the original training sample set to obtain an expanded training sample set.
5. The prediction method of claim 4, wherein after obtaining the extended training sample set, the method further comprises:
determining training statistical characteristics of the disk based on the extended training sample set;
classifying the training statistical characteristics of the magnetic disk according to the offline disk information and marking the magnetic disk to obtain a training characteristic set;
and processing the training feature set to obtain a training data set, and training the neural network based on the training data set.
6. The prediction method according to any one of claims 2 to 5, wherein the constructing a prediction feature from the parsed disk parameters comprises:
calculating the predicted statistical characteristics of the disk according to the SMART parameters of the disk and the I/O information of the disk operation;
and performing feature processing on the predicted statistical features to obtain the predicted features.
7. The prediction method according to any one of claims 1 to 5, characterized in that the method further comprises:
constructing an incremental training feature set according to the analyzed disk parameters;
incrementally training the neural network based on the incremental training feature set to update the neural network.
8. The prediction method according to any one of claims 1 to 5, characterized in that the prediction method further comprises:
and displaying the prediction result through a visual interface.
9. A disk failure prediction apparatus, the apparatus comprising:
the magnetic disk data analysis module is used for acquiring magnetic disk running state parameters to be analyzed and analyzing the magnetic disk running state parameters to obtain analyzed magnetic disk parameters;
and the disk failure prediction module is used for constructing prediction characteristics according to the analyzed disk parameters and carrying out prediction analysis on the prediction characteristics based on a neural network to obtain a prediction result.
10. A computer-readable storage medium, on which an information transfer-implementing program is stored, which, when executed by a processor, implements the steps of the disk failure prediction method according to any one of claims 1 to 8.
11. A server comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the computer program when executed by the processor implementing the steps of the disk failure prediction method according to any one of claims 1 to 8.
Priority Applications (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910886471.XA CN112527572A (en) | 2019-09-19 | 2019-09-19 | Disk failure prediction method and device, computer readable storage medium and server |
| PCT/CN2020/112876 WO2021052168A1 (en) | 2019-09-19 | 2020-09-01 | Disk fault prediction method and apparatus, computer-readable storage medium, and server |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910886471.XA CN112527572A (en) | 2019-09-19 | 2019-09-19 | Disk failure prediction method and device, computer readable storage medium and server |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112527572A true CN112527572A (en) | 2021-03-19 |
Family
ID=74883506
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910886471.XA Pending CN112527572A (en) | 2019-09-19 | 2019-09-19 | Disk failure prediction method and device, computer readable storage medium and server |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN112527572A (en) |
| WO (1) | WO2021052168A1 (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113590406A (en) * | 2021-08-16 | 2021-11-02 | 湖南博匠信息科技有限公司 | Method and system for detecting solid state disk fault based on electrical variable |
| CN113971003A (en) * | 2021-10-17 | 2022-01-25 | 中国船舶重工集团公司第七一六研究所 | Online sampling device and method for SMART data of magnetic disk |
| CN115827356A (en) * | 2023-02-23 | 2023-03-21 | 北京大道云行科技有限公司 | Server disk fault detection method and system, storage medium and electronic equipment |
| WO2024164713A1 (en) * | 2023-02-06 | 2024-08-15 | 天翼云科技有限公司 | Method and apparatus for determining state of health of memory, and electronic device and storage medium |
Family Cites Families (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102129397A (en) * | 2010-12-29 | 2011-07-20 | 深圳市永达电子股份有限公司 | Method and system for predicating self-adaptive disk array failure |
| CN104050071B (en) * | 2014-06-30 | 2017-02-08 | 苏州科达科技股份有限公司 | Storage device failure early warning method and system |
| CN104268040B (en) * | 2014-09-16 | 2016-04-06 | 深圳市腾讯计算机系统有限公司 | A kind of disk performance detection method and device |
| CN105068901A (en) * | 2015-07-27 | 2015-11-18 | 浪潮电子信息产业股份有限公司 | Method for detecting magnetic disc |
| CN108681496A (en) * | 2018-05-09 | 2018-10-19 | 北京奇艺世纪科技有限公司 | Prediction technique, device and the electronic equipment of disk failure |
| CN108959004B (en) * | 2018-06-28 | 2022-02-18 | 郑州云海信息技术有限公司 | Disk failure prediction method, device, equipment and computer readable storage medium |
| CN109491850A (en) * | 2018-11-21 | 2019-03-19 | 北京北信源软件股份有限公司 | A kind of disk failure prediction technique and device |
-
2019
- 2019-09-19 CN CN201910886471.XA patent/CN112527572A/en active Pending
-
2020
- 2020-09-01 WO PCT/CN2020/112876 patent/WO2021052168A1/en active Application Filing
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113590406A (en) * | 2021-08-16 | 2021-11-02 | 湖南博匠信息科技有限公司 | Method and system for detecting solid state disk fault based on electrical variable |
| CN113971003A (en) * | 2021-10-17 | 2022-01-25 | 中国船舶重工集团公司第七一六研究所 | Online sampling device and method for SMART data of magnetic disk |
| CN113971003B (en) * | 2021-10-17 | 2024-08-02 | 中国船舶集团有限公司第七一六研究所 | Online sampling device and method for disk SMART data |
| WO2024164713A1 (en) * | 2023-02-06 | 2024-08-15 | 天翼云科技有限公司 | Method and apparatus for determining state of health of memory, and electronic device and storage medium |
| CN115827356A (en) * | 2023-02-23 | 2023-03-21 | 北京大道云行科技有限公司 | Server disk fault detection method and system, storage medium and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| WO2021052168A1 (en) | 2021-03-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112527572A (en) | Disk failure prediction method and device, computer readable storage medium and server | |
| CN109739739B (en) | Disk failure prediction method, device and storage medium | |
| US11651375B2 (en) | Below-the-line thresholds tuning with machine learning | |
| KR20200074087A (en) | Chinese medicine production process knowledge system | |
| CN110837866A (en) | XGboost-based electric power secondary equipment defect degree evaluation method | |
| CN111507376A (en) | Single index abnormality detection method based on fusion of multiple unsupervised methods | |
| CN108304567B (en) | Method and system for identifying working condition mode and classifying data of high-voltage transformer | |
| CN113869721A (en) | Substation equipment health state classification method and apparatus | |
| CN117094184B (en) | Modeling method, system and medium of risk prediction model based on intranet platform | |
| CN115033591B (en) | Intelligent detection method, system, storage medium and computer equipment for electric charge data abnormality | |
| CN111986027A (en) | Abnormal transaction processing method and device based on artificial intelligence | |
| CN116861331A (en) | Expert model decision-fused data identification method and system | |
| CN113891342B (en) | Base station inspection method and device, electronic equipment and storage medium | |
| CN114818353A (en) | Train control vehicle-mounted equipment fault prediction method based on fault characteristic relation map | |
| CN115358481A (en) | Early warning and identification method, system and device for enterprise ex-situ migration | |
| CN113177644A (en) | Automatic modeling system based on word embedding and depth time sequence model | |
| CN113935413A (en) | Distribution network wave recording file waveform identification method based on convolutional neural network | |
| CN117291575A (en) | Equipment maintenance method, equipment maintenance device, computer equipment and storage medium | |
| CN113586554B (en) | Hydraulic motor detection system and method | |
| CN115511152A (en) | Prediction method, system, device and storage medium for building equipment fault maintenance | |
| CN115081514A (en) | Industrial equipment fault identification method under data imbalance condition | |
| CN113886592A (en) | Quality detection method for operation and maintenance data of power information communication system | |
| CN117131947B (en) | Overhead transmission line fault prediction method, device, equipment and storage medium | |
| CN116611022B (en) | Intelligent campus education big data fusion method and platform | |
| CN113435655B (en) | Sector dynamic management decision method, server and system |
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 |