A solid state drive fault prediction method and system based on artificial intelligence

Abstract

The present invention relates to the field of solid-state drive fault prediction, and in particular to a solid-state drive fault prediction method and system based on artificial intelligence. The method comprises the following steps: fine-grained division of the solid-state drive into storage units, and serialization of the storage units to construct a storage unit sequence for each partition; obtaining a solid-state drive operation log; performing multi-time point state evolution on the storage unit sequence for each partition based on the solid-state drive operation log, and performing deep dynamic evolution learning to generate a state evolution law for each storage unit; performing periodic access load peak calculation on the solid-state drive operation log, thereby generating a simulated access data packet for each storage unit; performing dynamic pressure access simulation based on the state evolution law of each storage unit and the simulated access data packet of each storage unit to obtain response state evaluation data for each storage unit. The present invention realizes efficient and accurate solid-state drive fault prediction.

Description

Technical Field

[0001] The present invention relates to the technical field of solid state drive failure prediction, and in particular to a solid state drive failure prediction method and system based on artificial intelligence. Background Art

[0002] With the rapid development of information technology, solid-state drives (SSDs), as a new type of storage medium, have gradually replaced traditional mechanical hard disks (HDDs) in various electronic devices. SSDs are widely used in personal computers, servers, data centers, mobile storage devices and other fields due to their advantages such as high read and write speeds, low power consumption and shock resistance, and have become a key component in modern data storage systems. However, although SSDs have a long service life, their performance may gradually decline or even fail during long-term operation, affecting the stability and security of data storage.

[0003] SSD failures are usually caused by a variety of internal and external factors, including but not limited to too many writes to the storage unit, unit aging, electrical failures, temperature fluctuations, hardware damage, etc. These factors often lead to data loss, reduced storage speed, increased response delays, and other problems. In severe cases, they may even cause the hard drive to fail completely. Due to the complex internal structure of SSDs, their failures are often not discovered in time through traditional physical inspections and manual diagnosis. Therefore, relying on traditional detection methods for fault prediction can no longer meet the efficiency and accuracy requirements of modern storage devices for fault detection.

[0004] Currently, research on SSD failure prediction mainly relies on real-time monitoring and data analysis of the hard disk health status. Common methods include empirical analysis based on hard disk health indicators (such as SMART data) and identifying potential problems through regular stress testing. However, traditional prediction methods usually rely on manual observation and rule setting, which not only has certain limitations, but also has difficulty in dealing with complex failure modes and emergencies. With the diversification of hard disk failure modes, the accuracy and response speed of traditional methods often cannot meet actual needs. Summary of the invention

[0005] In order to solve the above technical problems, the present invention proposes a solid state drive fault prediction method and system based on artificial intelligence to solve at least one of the above technical problems.

[0006] To achieve the above object, the present invention provides a solid state drive fault prediction method based on artificial intelligence, comprising the following steps:

[0007] Step S1: fine-grained partitioning of the solid state drive into storage units, and serialization of the storage units to construct a storage unit sequence for each partition;

[0008] Step S2: obtaining the SSD operation log; performing multi-time point state evolution on the storage unit sequence of each partition based on the SSD operation log, and performing deep dynamic evolution learning to generate the state evolution law of each storage unit;

[0009] Step S3: Calculate the periodic access load peak value of the solid state drive operation log, and perform data access stress test analysis to generate a simulated access data packet for each storage unit;

[0010] Step S4: performing dynamic pressure access simulation based on the state evolution law of each storage unit and the simulated access data packet of each storage unit to obtain response state evaluation data of each storage unit;

[0011] Step S5: mining the state trend evolution of the response state evaluation data of each storage unit, and locating the abnormal mutation partition to obtain the abnormal mutation sector;

[0012] Step S6: predict the fault probability of the abnormal mutation sector and perform a fault risk assessment to obtain a fault prediction assessment result.

[0013] The present invention helps to understand the hard disk storage structure by fine-grained division and serialization of storage units, providing basic data for subsequent analysis. Constructing a storage unit sequence for each partition can help monitor the state changes of the hard disk storage unit and provide data support for subsequent fault prediction. Learning the state evolution of storage units based on the hard disk operation log helps to understand the state change trend of storage units at different time points. Generating the state evolution law of storage units can help predict the future state of storage units and discover potential problems in advance. Calculating the access load peak of storage units and performing data access stress test analysis can help evaluate the performance and stability of the hard disk. Generating simulated access data packets can help simulate real workloads and provide support for subsequent stress tests. Providing data basis, dynamic pressure access simulation can evaluate the response capability of storage units under different loads and discover potential performance problems. Obtaining storage unit response status evaluation data helps understand the performance of hard disks under different loads and provide a basis for fault prediction. Trend evolution mining of response status evaluation data can help observe the law of storage unit state changes and discover anomalies in advance. Locating abnormal mutation sectors can help accurately locate potential failure points and provide guidance for fault prediction and maintenance. Predicting the failure probability of abnormal mutation sectors and conducting failure risk assessment can help prevent hard disk failures in advance and reduce losses. Obtaining fault prediction evaluation results can help formulate targeted maintenance plans and improve the reliability and stability of hard disks.

[0014] Preferably, step S1 comprises the following steps:

[0015] Step S11: performing physical address mapping partition identification on the solid state hard disk, thereby obtaining the sector partition data of the solid state hard disk;

[0016] Step S12: performing fine-grained division of the solid state hard disk sector partition data into storage units, and extracting the storage units of each partition;

[0017] Step S13: performing unit sequence coding on the storage unit of each partition to obtain the sequence coding of each storage unit;

[0018] Step S14: Serialize the storage units based on the sequential coding of each storage unit to construct a storage unit sequence for each partition.

[0019] The present invention helps to understand the storage structure and partition situation of the hard disk by performing physical address mapping partition identification on the hard disk, and obtaining the hard disk sector partition data can provide basic data support for subsequent storage unit analysis and fault prediction. Fine-grained division of storage units can help to better understand the storage structure and partition situation of the hard disk, and extracting the storage units of each partition can help to deeply analyze the performance and status of each storage unit, providing data support for subsequent fault prediction. Unit sequence encoding of the storage units of each partition helps to identify and distinguish different storage units, and obtaining the sequence encoding of each storage unit can help monitor the state change of the storage unit, providing a data basis for subsequent fault prediction. Storage unit serialization based on the sequence encoding of the storage unit can help record the state change history of the storage unit, and constructing the storage unit sequence of each partition helps to analyze the state evolution law of the storage unit, providing data support for fault prediction and maintenance.

[0020] Preferably, the specific steps of step S2 are:

[0021] Step S21: Obtaining the solid state drive operation log;

[0022] Step S22: performing sliding filtering and smoothing on the solid state drive operation log to generate a filtering and smoothing operation log;

[0023] Step S23: performing storage unit feature matching on the filtered smoothing operation log based on the storage unit sequence of each partition, thereby obtaining an operation log segment of each storage unit;

[0024] Step S24: mining the state characteristics of each storage unit on the operation log fragment of each storage unit to generate the state characteristics of each storage unit;

[0025] Step S25: performing multi-time point state evolution on the state characteristics of each storage unit, thereby obtaining multiple time point state evolution characteristics;

[0026] Step S26: Perform deep dynamic evolution learning on the state evolution features at multiple time points to generate the state evolution law of each storage unit.

[0027] The present invention can provide a detailed record of the operation status of the hard disk by acquiring the operation log of the solid-state hard disk, providing a data basis for subsequent analysis. The information contained in the operation log can help understand the usage and operation characteristics of the hard disk and provide a basis for fault prediction. Sliding filtering and smoothing the operation log of the solid-state hard disk can remove noise and extract trend changes in data. Generating a filtered and smoothed operation log is helpful to more accurately analyze the operation status of the hard disk and provide clear data for subsequent feature extraction and fault prediction. Feature matching of the filtered and smoothed operation log based on the storage unit sequence can associate the operation log and the storage unit, and obtaining the operation log fragment of each storage unit is helpful to deeply understand the operation status and characteristics of each storage unit and the operation of the storage unit. Mining state features of log fragments can extract key state feature information. Generating state features of each storage unit helps analyze the performance and health status of the storage unit and provide data support for fault prediction. Multi-point state evolution of the state features of the storage unit can observe the state change trend of the storage unit at different time points. Obtaining state evolution features at multiple time points helps analyze the state evolution law of the storage unit and provide a more comprehensive data basis for fault prediction. Deep dynamic evolution learning of state evolution features at multiple time points can discover the dynamic change law of the storage unit state. Generating the state evolution law of each storage unit helps accurately predict the future state changes of the storage unit, discover potential problems in advance and perform maintenance.

[0028] Preferably, the specific steps of step S24 are:

[0029] Calculate the read and write speed of each storage unit for each operation log segment to generate the read and write speed of each storage unit;

[0030] Perform read and write delay distribution analysis based on the read and write speed of each storage unit to generate storage unit read and write delay distribution data;

[0031] Based on the storage unit read and write delay distribution data, the data throughput time series change analysis is performed on the operation log fragment of each storage unit to obtain the throughput time series change data of each storage unit;

[0032] Perform data access behavior detection on the operation log fragments of each storage unit and extract the data access behavior characteristics of each storage unit;

[0033] Access error identification is performed on the data access behavior characteristics of each storage unit to obtain storage unit access error data;

[0034] Classifying the error type of storage unit access error data to obtain the access error type;

[0035] Calculate the access error frequency according to the access error type and generate the access error classification frequency;

[0036] The storage unit status feature mining is performed on the throughput time series change data and access error classification frequency of each storage unit to generate the status feature of each storage unit.

[0037] The present invention helps evaluate the performance of storage units by calculating the read and write speed of each storage unit. Generating the read and write speed data of each storage unit can provide a basis for subsequent performance analysis and fault prediction. Delay distribution analysis based on the read and write speed can reveal the fluctuation of the read and write performance of the storage unit. Generating the read and write delay distribution data of the storage unit helps understand the stability of the performance of the storage unit and provides a reference for fault prediction. Analyzing the throughput time series variation of the storage unit can reveal the variation trend of the storage unit performance over time. Obtaining the throughput time series variation data of each storage unit helps monitor performance fluctuations and provides a basis for fault prediction and maintenance. Detecting the data access behavior of the storage unit and extracting features can understand the usage mode of the storage unit and extract the data access behavior of the storage unit. The behavior characteristics of the access error can help to discover abnormal behavior or potential problems and provide support for fault prediction. Identifying and classifying the access errors of the storage unit can help to discover data access anomalies or hard disk problems. Obtaining the access error data of the storage unit can help to warn of possible failures in advance and improve the stability and reliability of the hard disk. Classifying the access error types and calculating the frequency can help understand the occurrence of different types of errors. Generating access error classification frequency data can help to evaluate the stability and risk level of the hard disk. Combining the throughput timing change data and the access error classification frequency for state feature mining can comprehensively evaluate the state of the storage unit. Generating the state characteristics of each storage unit can help to comprehensively analyze the performance, stability and risk of the hard disk and provide decision support for fault prediction and maintenance.

[0038] Preferably, step S3 specifically comprises the following steps:

[0039] Step S31: Calculate the data request waiting queue length for the operation log fragment of each storage unit to obtain the data request waiting queue length of each storage unit;

[0040] Step S32: Calculate the periodic access load peak value according to the length of the data request waiting queue of each storage unit to obtain the access load peak value of each storage unit;

[0041] Step S33: performing data access stress test analysis based on the access load peak value of each storage unit to generate an access stress evaluation value of the storage unit;

[0042] Step S34: Perform data access simulation according to the access pressure evaluation value of the storage unit, thereby generating a simulated access data packet for each storage unit.

[0043] The present invention helps to understand the queuing situation of data access by calculating the length of the data request waiting queue of each storage unit. Obtaining the length of the data request waiting queue of each storage unit can help evaluate the response speed and performance bottleneck of the storage unit. Calculating the access load peak according to the length of the data request waiting queue helps to determine the maximum load capacity of the storage unit. Obtaining the access load peak of each storage unit can provide guidance for system load balancing and performance optimization. Performing data access stress test analysis based on the access load peak can evaluate the performance of the storage unit under high load conditions. Generating an access stress evaluation value of the storage unit helps to identify the performance bottleneck of the system and formulate a performance improvement strategy. Performing data access simulation according to the access pressure evaluation value can simulate the performance of the storage unit under different loads. Generating a simulated access data packet for each storage unit helps to evaluate the stability, capacity requirements and performance expectations of the system, and provides support for fault prediction and performance optimization.

[0044] Preferably, the specific steps of step S4 are:

[0045] Step S41: Based on the state evolution law of each storage unit, dynamic state mapping modeling is performed on the storage unit sequence of each partition to construct multiple storage unit digital twin models;

[0046] Step S42: using the simulated access data packet of each storage unit to perform dynamic pressure access simulation on multiple storage unit digital twin models, and collecting storage unit access simulation response data;

[0047] Step S43: decomposing the storage unit access simulation response data into multiple time periods, and extracting the access simulation response data of multiple time periods;

[0048] Step S44: performing storage unit response status evaluation on the access simulation response data of multiple time periods to obtain response status evaluation data of each storage unit.

[0049] The present invention helps to understand the behavior of storage units in different states by performing dynamic state mapping modeling based on the state evolution law of storage units. Constructing multiple digital twin models of storage units can provide a basis for real-time monitoring and prediction of the state evolution of storage units. Using simulated access data packets to perform dynamic pressure access simulation on the digital twin model helps to evaluate the response of the storage unit under actual load. Collecting storage unit simulation response data can help verify the accuracy and reliability of the digital twin model. Multi-period decomposition of access simulation response data can reveal the response characteristics of storage units in different time periods. Extracting data from multiple time periods helps to analyze the performance and response laws of storage units under different loads. Evaluating the storage unit response state of access simulation response data in multiple time periods helps to comprehensively evaluate the performance and stability of the storage unit. Obtaining the response state evaluation data of each storage unit can help identify potential problems, predict failures and optimize performance.

[0050] Preferably, the specific steps of step S44 are:

[0051] Calculating the unit access response speed for the access simulation response data of multiple time periods to obtain the unit access response speeds of multiple time periods;

[0052] Perform access delay analysis on access response speeds of multiple time period units and extract access delay data for each time period;

[0053] Perform delay variation analysis on the access delay data of each time period to obtain the unit access delay characteristics;

[0054] Calculate the data access accuracy of access simulation response data in multiple time periods to obtain the access accuracy of each time period;

[0055] The access accuracy rate of each time period is identified by the time period accuracy distribution to generate multi-time period accuracy distribution data;

[0056] The storage unit response status is evaluated based on the unit access delay characteristics and multi-period accuracy distribution data to obtain response status evaluation data of each storage unit.

[0057] The present invention helps to evaluate the performance of the storage unit in different time periods by calculating the unit access response speed of each time period, and obtaining the unit access response speeds of multiple time periods can help identify performance fluctuations and bottlenecks. Performing access delay analysis on the unit access response speeds of multiple time periods can reveal the access delay of the storage unit, and extracting the access delay data of each time period helps to evaluate the stability and change trend of the access speed. The characteristics and change laws of the storage unit access delay can be understood through delay change analysis. Obtaining the unit access delay characteristics can help to discover delay anomalies and optimize access performance. Calculating the data access accuracy of each time period helps to evaluate the accuracy and reliability of the data access of the storage unit. Obtaining the access accuracy of each time period can provide an evaluation basis for data integrity and reliability. By identifying the time period accuracy distribution, the distribution of the data access accuracy of the storage unit in different time periods can be understood. Generating multi-time period accuracy distribution data can help to discover fluctuations and trends in data access quality. Combining the unit access delay characteristics and multi-time period accuracy distribution data to perform storage unit response status evaluation helps to comprehensively evaluate the performance and reliability of the storage unit. Obtaining the response status evaluation data of each storage unit can help predict faults, optimize performance and improve data access efficiency.

[0058] Preferably, the specific steps of step S5 are:

[0059] Step S51: mining the state trend evolution of the response state evaluation data of each storage unit to generate storage unit response state trend evolution data;

[0060] Step S52: performing time series evolution fitting on the storage unit response state trend evolution data to construct a state trend evolution curve;

[0061] Step S53: performing state trend mutation identification on the state trend evolution curve, and extracting the state trend mutation point of the storage unit;

[0062] Step S54: based on the state trend mutation point of the storage unit, locate the abnormal mutation partition of the solid state hard disk to obtain the abnormal mutation sector.

[0063] The present invention helps to understand the trend and law of storage unit state changes by performing state trend evolution mining on storage unit response state evaluation data. Generating storage unit response state trend evolution data can provide a basis for anomaly detection and fault prediction. Performing time series evolution fitting on storage unit response state trend evolution data can establish a state trend evolution curve to better describe the state change trend. Constructing the state trend evolution curve helps to visualize state changes and facilitates analysis and identification of abnormal changes. By performing state trend mutation identification on the state trend evolution curve, mutation points in storage unit state changes can be captured. Extracting the state trend mutation points of storage units helps to identify abnormal situations and critical moments when problems occur. Abnormal mutation partition positioning of solid-state hard disks based on the state trend mutation points of storage units can accurately locate the specific area where the abnormality occurs. Obtaining the abnormal mutation sector can help quickly locate the fault area, take corresponding measures and improve the maintenance efficiency of the solid-state hard disk.

[0064] Preferably, the specific steps of step S6 are:

[0065] Step S61: predicting the fault probability of the abnormal mutation sector, thereby obtaining abnormal sector fault prediction data;

[0066] Step S62: performing dynamic propagation path evolution analysis on the abnormal sector fault prediction data to obtain fault propagation path evolution data;

[0067] Step S63: performing partition propagation path distribution analysis on the solid state drive according to the fault propagation path evolution data to obtain the solid state drive fault propagation path distribution data;

[0068] Step S64: performing a fault risk assessment on the fault propagation path distribution data of the solid state drive to obtain a fault prediction assessment result.

[0069] The present invention can help predict the possibility of abnormal sector failure by predicting the failure probability for abnormal mutation sectors. Obtaining abnormal sector failure prediction data can help take measures in advance to reduce potential data loss and losses caused by hard disk failure. Dynamic propagation path evolution analysis of abnormal sector failure prediction data can reveal the propagation path and evolution process of the fault in the solid-state hard disk. Obtaining fault propagation path evolution data can help understand the diffusion law of the fault and improve the efficiency of responding to and handling the fault. Partition propagation path distribution analysis based on fault propagation path evolution data can understand the distribution of fault propagation paths in the solid-state hard disk. Obtaining solid-state hard disk fault propagation path distribution data can help determine fault hotspot areas and key monitoring areas, improve fault detection and management efficiency, perform fault risk assessment on solid-state hard disk fault propagation path distribution data to assess the failure risk level of the solid-state hard disk, and obtain fault prediction assessment results can help formulate reasonable maintenance strategies, reduce the failure rate of solid-state hard disks, and improve system stability and reliability.

[0070] In this specification, a solid state drive fault prediction system based on artificial intelligence is provided, which is used to execute the solid state drive fault prediction method based on artificial intelligence as described above, including:

[0071] The storage unit partitioning module is used to perform fine-grained partitioning of the storage units of the solid-state drive, serialize the storage units, and construct a storage unit sequence for each partition;

[0072] The dynamic evolution learning module is used to obtain the SSD operation log; based on the SSD operation log, the storage unit sequence of each partition is subjected to multi-time point state evolution, and deep dynamic evolution learning is performed to generate the state evolution law of each storage unit;

[0073] The load peak module is used to calculate the periodic access load peak of the SSD operation log and perform data access stress test analysis to generate simulated access data packets for each storage unit;

[0074] A dynamic access simulation module is used to perform dynamic pressure access simulation based on the state evolution law of each storage unit and the simulated access data packet of each storage unit to obtain response state evaluation data of each storage unit;

[0075] The state trend evolution mining module is used to mine the state trend evolution of the response state evaluation data of each storage unit, and locate the abnormal mutation partition to obtain the abnormal mutation sector;

[0076] The fault prediction module is used to predict the fault probability of abnormal mutation sectors and perform fault risk assessment to obtain fault prediction assessment results.

[0077] The present invention helps to monitor the hard disk state more finely and improve the accuracy of anomaly detection by dividing the storage unit in fine granularity. The serialized storage unit can provide ordered data for subsequent analysis, which is convenient for learning and predicting the state evolution law. Obtaining the operation log for state evolution learning helps to monitor the hard disk state in real time and discover potential faults. By learning the state evolution law of the storage unit, the hard disk state change can be predicted in advance and the risk of failure can be reduced. Calculating the load peak and stress test analysis helps to understand the working condition and load of the hard disk. Generating simulated access data packets can help simulate real data access scenarios and evaluate the performance and stability of the hard disk. Dynamic stress access simulation based on the state evolution law and the simulated access data packet can simulate the hard disk response under the real workload. Obtaining the response state evaluation data of the storage unit helps to monitor the hard disk state in real time and discover potential faults in time. Mining the state trend evolution of the response state evaluation data can reveal the law and trend of the storage unit state change. Abnormal mutation partition positioning helps to accurately locate abnormal situations, early warning and processing of hard disk abnormalities. Fault probability prediction and risk assessment help to predict hard disk failures in advance and reduce the risk of accidental data loss and hard disk damage. The fault prediction evaluation result can guide the formulation of maintenance strategies and improve the reliability and stability of solid-state hard disks. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] Figure 1 A schematic diagram of the steps of a solid state drive fault prediction method based on artificial intelligence of the present invention;

[0079] Figure 2 Detailed implementation flow chart of step S1;

[0080] Figure 3 Detailed implementation flow chart of step S2;

[0081] Figure 4 Detailed implementation flow chart of step S3. DETAILED DESCRIPTION

[0082] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0083] The present application example provides a solid state drive fault prediction method and system based on artificial intelligence. The execution subject of the solid state drive fault prediction method and system based on artificial intelligence includes but is not limited to: mechanical equipment, data processing platform, cloud server node, network upload device, etc. equipped with the system can be regarded as the general computing node of the present application, and the data processing platform includes but is not limited to: at least one of an audio image management system, an information management system, and a cloud data management system.

[0084] See also Figures 1 to 4The present invention provides a solid state drive fault prediction method based on artificial intelligence, and the solid state drive fault prediction method based on artificial intelligence includes the following steps:

[0085] Step S1: fine-grained partitioning of the solid state drive into storage units, and serialization of the storage units to construct a storage unit sequence for each partition;

[0086] Step S2: obtaining the SSD operation log; performing multi-time point state evolution on the storage unit sequence of each partition based on the SSD operation log, and performing deep dynamic evolution learning to generate the state evolution law of each storage unit;

[0087] Step S3: Calculate the periodic access load peak value of the solid state drive operation log, and perform data access stress test analysis to generate a simulated access data packet for each storage unit;

[0088] Step S4: performing dynamic pressure access simulation based on the state evolution law of each storage unit and the simulated access data packet of each storage unit to obtain response state evaluation data of each storage unit;

[0089] Step S5: mining the state trend evolution of the response state evaluation data of each storage unit, and locating the abnormal mutation partition to obtain the abnormal mutation sector;

[0090] Step S6: predict the fault probability of the abnormal mutation sector and perform a fault risk assessment to obtain a fault prediction assessment result.

[0091] The present invention helps to understand the hard disk storage structure by fine-grained division and serialization of storage units, providing basic data for subsequent analysis. Constructing a storage unit sequence for each partition can help monitor the state changes of the hard disk storage unit and provide data support for subsequent fault prediction. Learning the state evolution of storage units based on the hard disk operation log helps to understand the state change trend of storage units at different time points. Generating the state evolution law of storage units can help predict the future state of storage units and discover potential problems in advance. Calculating the access load peak of storage units and performing data access stress test analysis can help evaluate the performance and stability of the hard disk. Generating simulated access data packets can help simulate real workloads and provide support for subsequent stress tests. Providing data basis, dynamic pressure access simulation can evaluate the response capability of storage units under different loads and discover potential performance problems. Obtaining storage unit response status evaluation data helps understand the performance of hard disks under different loads and provide a basis for fault prediction. Trend evolution mining of response status evaluation data can help observe the law of storage unit state changes and discover anomalies in advance. Locating abnormal mutation sectors can help accurately locate potential failure points and provide guidance for fault prediction and maintenance. Predicting the failure probability of abnormal mutation sectors and conducting failure risk assessment can help prevent hard disk failures in advance and reduce losses. Obtaining fault prediction evaluation results can help formulate targeted maintenance plans and improve the reliability and stability of hard disks.

[0092] In the embodiment of the present invention, refer to Figure 1 , is a schematic flow chart of the steps of a solid state drive fault prediction method based on artificial intelligence of the present invention. In this example, the steps of the solid state drive fault prediction method based on artificial intelligence include:

[0093] Step S1: fine-grained partitioning of the solid state drive into storage units, and serialization of the storage units to construct a storage unit sequence for each partition;

[0094] In this embodiment, the solid-state drive (SSD) is composed of multiple storage units (usually referred to as pages or blocks), which are used to store data. The size and organization of each storage unit are usually determined by the hardware design. The common page size is 4KB or 8KB, and the block size may be 128KB to 512KB. According to the usage and performance requirements of the solid-state drive, determine how to divide the storage units, for example: by logical partition: logical partition based on the file system, by physical partition: physical partition based on the hardware level, use the hardware interface (such as NVMe or SATA) to obtain the physical structure information of the hard disk, determine the address and size of each storage unit, based on page division: each page is regarded as an independent storage unit, for example, a 4KB page can be regarded as a storage unit, based on block division: when a higher level of management is required, Aggregate multiple pages in each block into a logical storage unit, create a mapping table, record the physical address, logical address and status (such as used, available, bad block, etc.) of each storage unit, and convert the divided storage units into a serialized data structure. The purpose of serialization is to facilitate subsequent data processing and analysis. Select a suitable data structure (such as an array, a linked list) to store the order of storage units in each partition, traverse the mapping table, and fill the logical address and status information of the storage unit into the selected data structure in sequence to form a complete storage unit sequence. If there are multiple logical partitions in the solid-state drive, such as primary partitions, extended partitions, etc., the storage units of each partition need to be serialized separately, and the above serialization process is repeated for each logical partition to generate a storage unit sequence for each partition. This can be done by traversing the logical address range of the mapping table.

[0095] Step S2: obtaining the SSD operation log; performing multi-time point state evolution on the storage unit sequence of each partition based on the SSD operation log, and performing deep dynamic evolution learning to generate the state evolution law of each storage unit;

[0096] In this embodiment, the source of the operation log is determined. Generally, the operation log of the solid-state hard disk can be obtained in the following ways: for example, using the SMART (Self-Monitoring, Analysis and Reporting Technology) command to obtain the log from the hard disk, using the tools provided by the operating system (such as Windows Event Viewer or Linux syslog) to obtain the operation information related to the hard disk, and extracting the operation log of the solid-state hard disk through the programming interface or command line tool. The log should contain the following key information: timestamp: record the occurrence time of each event, event type: such as read and write request, error event, performance index, etc., storage unit information: the address and status of the storage unit involved, for each partition's storage unit sequence, build an analysis framework to analyze the evolution of multi-time points, and select a suitable model to analyze the state change of each storage unit. Common models include: time series Column models: such as ARIMA (autoregressive integrated moving average model) or LSTM (long short-term memory neural network), Markov process: used to model the transition probability of storage unit status, time period division: divide the log data into multiple time periods by time to analyze the state changes in each time period, extract the state information of the storage unit in each time period from the log (such as the number of reads, writes, errors, etc.) and associate it with the storage unit sequence, calculate the state changes of each storage unit between different time periods, record the improvement, decline or maintenance of the state, and select a suitable deep learning model for dynamic evolution learning, for example: LSTM network: LSTM is suitable for processing time series data and can capture the dynamic changes of storage unit status over time, GRU (gated recurrent unit): another deep learning model suitable for time series with a relatively simple structure.

[0097] Step S3: Calculate the periodic access load peak value of the solid state drive operation log, and perform data access stress test analysis to generate a simulated access data packet for each storage unit;

[0098] In this embodiment, the log data is cleaned to remove invalid or duplicate records, ensure the consistency of data format, and fill in missing values ​​(if necessary). The definition of the row cycle is determined, which can usually be divided into hours, days or weeks. The selection of the row cycle depends on the analysis objectives and the characteristics of the data. According to the defined row cycle, the number of access requests for each storage unit is counted. The following steps can be used: group the log data according to the row cycle, calculate the number of read and write requests for each storage unit in each cycle, for each row cycle, calculate the access load, which is usually expressed as the number of access requests, in the statistical results, identify the load peak in each row cycle, record the time when the peak occurs and its corresponding storage unit, for example: use an aggregate function (such as maximum value) to find the maximum number of access requests in each cycle, and record the corresponding storage unit, Based on the row cycle load peak, determine the parameters of the stress test, such as the number of concurrent accesses, request type (read or write), request size, etc., and generate a simulated access data packet based on the identified load peak. The specific steps include: Design the data packet format: Determine the data structure of the simulated request, including the target storage unit, request type, request size, timestamp and other attributes, use a random number generator or a specific distribution model (such as Poisson distribution) to generate simulated requests, ensure that the request conforms to the actual access pattern, and send the generated simulated access data packet to the solid-state drive for stress testing. The test process may include: Initiating multiple requests at the same time to simulate high load conditions and observe the responsiveness of the hard disk. During the test, monitor key performance indicators (such as response time, IOPS (input and output operations per second), latency, etc.), and record these data for analysis.

[0099] Step S4: performing dynamic pressure access simulation based on the state evolution law of each storage unit and the simulated access data packet of each storage unit to obtain response state evaluation data of each storage unit;

[0100] In this embodiment, the state evolution law of each storage unit is obtained, including historical access patterns, state transition probabilities, and performance indicators. This information will be used to guide the simulation process. The generated simulated access data packet is extracted to ensure that the data packet contains information such as the address of the target storage unit, the request type (read or write), the request size, and the timestamp. A suitable stress testing tool or framework is selected to perform dynamic access simulation. Commonly used tools include: using programming languages ​​such as Python and Java to write scripts to simulate access requests of the storage system. Existing testing tools such as fio and IOzone can perform load testing and performance evaluation. The parameters of the simulation are determined, including: concurrency: the number of requests initiated simultaneously, simulation time: the total duration of the simulation run, request ratio: the ratio of read requests to write requests, and the dynamic stress access simulation is performed using the selected tool or script. The specific steps are as follows: ensure that the solid-state drive is in a testable state, prepare monitoring tools to obtain real-time performance data, dynamically generate and send requests based on the content of the simulated access data packet, and the generation of requests can follow the following logic: predict the type and size of the next request based on the historical state and evolution law of the storage unit, and predict the type and size of the next request based on the timestamp. Schedule the initiation time of the request to ensure that the simulation process is as realistic as possible. Use multi-threading or asynchronous call technology to initiate concurrent requests to simulate high-load access scenarios. During the simulation, monitor the performance indicators of each storage unit in real time, including: response time: the processing time of each request, IOPS (input and output operations per second): record the total number of requests during the simulation, delay: the average delay time of the request, error rate: monitor failed requests during the simulation, record the error type and occurrence time, after the simulation is completed, collect all performance monitoring data and associate it with the request of each storage unit. The specific steps include: extract response time: record the response time of each request and calculate the average response time, statistical performance indicators: based on the monitoring data, statistically calculate the IOPS, delay and error rate and other performance indicators of each storage unit, based on the collected performance data, generate response status evaluation data for each storage unit, including: storage unit address: identify the specific location of the storage unit, average response time: the average response time of each storage unit in the simulation, IOPS: the number of input and output operations per second under the simulation pressure, error rate: the proportion of failed requests for each storage unit during the simulation.

[0101] Step S5: mining the state trend evolution of the response state evaluation data of each storage unit, and locating the abnormal mutation partition to obtain the abnormal mutation sector;

[0102] In this embodiment, the response status evaluation data of each storage unit is obtained, including indicators such as average response time, IOPS, and error rate. These data will be used for subsequent trend analysis and anomaly detection. The collected response status data is cleaned to remove noise and outliers to ensure the integrity and consistency of the data. If necessary, standardization or normalization can be performed to facilitate better comparison and analysis. A suitable trend analysis method is selected. Common methods include: moving average method: smoothing data to reduce short-term fluctuations and highlight long-term trends. Linear regression analysis: identifying the direction and strength of trends by fitting data. Time series analysis: such as ARIMA model, which can capture the seasonality and trend of data. Trend analysis is performed on the response status evaluation data. The performance indicator changes of each storage unit in different time periods are calculated. For example, the trend of the average response time is calculated using the moving average method. The performance trend line of each storage unit is obtained by linear regression, and the slope is calculated. The trend analysis results are visualized. A line chart or other forms of charts are used to display the state changes of each storage unit for intuitive observation. The identification criteria for abnormal mutations are determined. Common criteria include: changes Rate threshold: set a threshold. When the rate of change of the response state exceeds this threshold, it can be identified as an anomaly. Inflection point detection: use the first-order and second-order derivatives to identify the inflection points of trend changes, detect the state changes of acceleration or deceleration, calculate the rate of change of the state change of each storage unit, identify the mutation points that exceed the threshold, identify the abnormal mutation points of each storage unit according to the set standards, and record the corresponding timestamps and values. Partition the storage units according to physical locations or logical locations to facilitate the analysis of anomaly detection. They can be grouped according to the physical addresses or partitions of the storage units. According to the detected abnormal mutation points, identify which partitions may be affected. The specific steps include: associating the abnormal mutation points with the partition information of the storage units, identifying the areas where abnormal mutations occur, counting the number and severity of abnormal mutations in each partition, determining the partition with the greatest impact, and recording the identified abnormal mutation sector information into structured data, including the following: Sector number: identifies the specific sector of the abnormal mutation, abnormal type: describes the characteristics of the abnormality (such as mutation amplitude, time, etc.), impact assessment: conduct a preliminary assessment of the possible impact on the sector.

[0103] Step S6: predict the fault probability of the abnormal mutation sector and perform a fault risk assessment to obtain a fault prediction assessment result.

[0104] In this embodiment, the information of the identified abnormal mutation sectors is obtained from step S5, including the status change, response time, IOPS and error rate of each sector, and the historical fault data related to the abnormal mutation sectors is collected, including the previous fault records, fault types, occurrence time and repair results, etc. These data will be used for model training and verification. According to the data characteristics, a suitable fault probability prediction model is selected. Commonly used models include: statistical models: such as logistic regression, linear regression or survival analysis, which are suitable for simple fault prediction; machine learning models: such as random forest, support vector machine (SVM) or neural network, which are suitable for processing complex nonlinear relationships, and determining the key features that affect the fault probability, including: performance indicators: such as response time, IOP S, error rate, etc., state change characteristics: such as the latest state change amplitude, change frequency, etc., historical fault characteristics: such as the number of historical faults, fault intervals, etc., data cleaning: remove missing values ​​or outliers, normalization: normalize the features to ensure that different features are trained on the same scale, divide the data into training set and test set (for example, 70% for training and 30% for testing), train the selected fault probability prediction model on the training set, and perform hyperparameter tuning to improve model performance, evaluate the prediction performance of the model on the test set, and use indicators such as accuracy, recall, F1-score and ROC-AUC Etc., to ensure the effectiveness of the model, apply the trained model to the feature data of the abnormal mutation sector, calculate the failure probability of each sector, and each storage unit should get a corresponding failure probability value. Organize the prediction results, including the failure probability and related features of each abnormal mutation sector, to form a failure probability prediction data set, and determine the key indicators for evaluating failure risks, such as: Failure probability: output from the prediction model, Impact range: the number of storage units or data loss that may be affected by the failure, Recovery time: an estimate of the time required to repair the failure, Based on the collected indicators, calculate the risk score of each abnormal mutation sector. Common methods include: Weighted scoring method: give weights according to the importance of each indicator, and calculate the comprehensive risk Scoring, Decision Tree Model: Use the decision tree algorithm to automatically generate risk scores based on input features. According to the calculated risk scores, classify all abnormal mutation sectors (such as high risk, medium risk, and low risk) and sort them so that high-risk sectors are processed first. The fault prediction evaluation results are saved in a structured format to the database or report, which contains the following information: Storage unit address: identifies the storage unit with abnormal mutations, Failure probability: The failure probability of each storage unit, Risk score: The calculated risk score and classification, organize the analysis results, and generate a fault prediction evaluation report, which contains detailed information of the abnormal mutation sectors, failure probability, risk assessment results, and recommended measures, so that relevant personnel can make decisions and handle them.

[0105] In this embodiment, refer to Figure 2, is a flowchart of detailed implementation steps of step S1. In this embodiment, the detailed implementation steps of step S1 include:

[0106] Step S11: performing physical address mapping partition identification on the solid state hard disk, thereby obtaining the sector partition data of the solid state hard disk;

[0107] Step S12: performing fine-grained division of the solid state hard disk sector partition data into storage units, and extracting the storage units of each partition;

[0108] Step S13: performing unit sequence coding on the storage unit of each partition to obtain the sequence coding of each storage unit;

[0109] Step S14: Serialize the storage units based on the sequential coding of each storage unit to construct a storage unit sequence for each partition.

[0110] In this embodiment, a hard disk management tool (such as fdisk, gdisk or a disk management tool provided by the operating system) is used to obtain the physical address information and partition table of the solid state disk, parse the data in the partition table, identify the start and end sector addresses of each partition, which is usually contained in the master boot record (MBR) or GUID partition table (GPT) of the hard disk, store the identified physical address information of each partition (such as the start sector, end sector, partition size, etc.) in a structured data format (such as CSV, JSON or a database), determine the size of the storage unit for fine-grained division (usually the sector size, such as 512 bytes or 4096 bytes), this step can be obtained through the hard disk specification, for example, for a partition with a start sector of 100 and an end sector of 200, it can be divided into multiple storage units, and the address range of each unit is: storage unit 0: [100, 101], storage unit 1: [101, 102], create a structured data format for storage units, record the address information of each storage unit, the partition it belongs to, etc., determine how to sequentially encode the storage units, for example: use a simple integer encoding (starting from 0 and increasing), use a unique identifier (such as UUID) for encoding, traverse the storage units of each partition, encode each unit according to the selected rules, store each storage unit and its corresponding sequential encoding in a data structure (such as a dictionary, a list, or a database table), determine the serialization format, for example, use JSON, XML, or a custom text format, convert the storage units of each partition and their sequential encoding into serialized data in the selected format, for example, use Python's JSON library for serialization, and save the serialized data to a file system or database for subsequent access and processing.

[0111] In this embodiment, refer to Figure 3, is a flowchart of detailed implementation steps of step S2. In this embodiment, the detailed implementation steps of step S2 include:

[0112] Step S21: Obtaining the solid state drive operation log;

[0113] Step S22: performing sliding filtering and smoothing on the solid state drive operation log to generate a filtering and smoothing operation log;

[0114] Step S23: performing storage unit feature matching on the filtered smoothing operation log based on the storage unit sequence of each partition, thereby obtaining an operation log segment of each storage unit;

[0115] Step S24: mining the state characteristics of each storage unit on the operation log fragment of each storage unit to generate the state characteristics of each storage unit;

[0116] Step S25: performing multi-time point state evolution on the state characteristics of each storage unit, thereby obtaining multiple time point state evolution characteristics;

[0117] Step S26: Perform deep dynamic evolution learning on the state evolution features at multiple time points to generate the state evolution law of each storage unit.

[0118] In this embodiment, the operation log type of the solid-state hard disk is confirmed. Common log sources include the firmware log of the hard disk, the operating system log (such as Windows Event Viewer), and the hard disk monitoring tool (such as SMART data). Select a suitable tool, such as smartctl, which is a command line tool for obtaining the SMART information of the hard disk. Execute relevant commands to extract the detailed operation log of the hard disk, save the extracted log output as a text file or import it into a database for subsequent processing, clean the extracted log, remove irrelevant information, ensure the uniformity of data format (such as timestamp, event type, value, etc.), and prepare for subsequent analysis. Select a suitable sliding filtering algorithm according to the characteristics of the data. The simple moving average is suitable for removing high-frequency noise, while the weighted moving average can give a higher weight to the most recent data, thereby better reflecting the current state. Perform a sliding filtering operation on the smoothed operation log. By setting a window (such as 5 data points), the data in the window is filtered. Calculate the row average and replace the data value of the center point to reduce the impact of noise. Store the filtered and smoothed operation log as a new file or database record for subsequent feature matching and analysis. Refer to the storage unit sequence extracted in the previous step to ensure that the address and feature information of each storage unit are ready for subsequent matching. Perform feature matching on the filtered and smoothed operation log to identify log fragments related to the storage unit sequence. This can be achieved by setting matching criteria (such as thresholds and similarity calculations). For example, based on the read and write characteristics of the storage unit, the corresponding log records can be matched to extract the operation log fragments related to the specific storage unit. , record the operation log fragments corresponding to each storage unit in a structured data format for subsequent state feature mining and analysis, define the state features to be mined, such as: read and write speed: evaluate the performance of the storage unit by calculating the data transmission rate, delay: record the response time of each read and write operation, error rate: calculate the error frequency of each storage unit based on the error events recorded in the operation log, perform detailed analysis on the operation log fragments of each storage unit, extract the defined state features, and use data aggregation technology, such as counting the read and write speed and delay by time period to generate a detailed performance report, and store the extracted state features in the result. Structured data, such as database tables or data frames, for subsequent analysis and use. Collect the state characteristics of each storage unit in chronological order to ensure the integrity and accuracy of the data, including read and write speeds, delays, and error rates at different time points. Analyze the changes in state characteristics at multiple time points and extract evolutionary characteristics, such as: Growth rate: The rate of change of state characteristics over time, which helps identify the trend of performance degradation. Volatility: The standard deviation of state characteristics, which reflects the stability of performance. According to the nature of the data, select a suitable deep learning model, such as: Long short-term memory network (LSTM): Suitable for processing time series data and can capture long-term dependencies.Convolutional Neural Network (CNN): Suitable for feature extraction and pattern recognition, it organizes the state evolution features of multiple time points into model input format, and divides them into training sets and test sets to facilitate model training and verification. It uses the selected deep learning model for training, and optimizes the model parameters through the back propagation algorithm to improve the ability to predict the state evolution law. It uses the trained model to predict the future state of the storage unit and generate the state evolution law. These laws can be used to monitor the health status of the storage unit and timely warn of potential failures.

[0119] In this embodiment, the specific steps of step S24 are:

[0120] Calculate the read and write speed of each storage unit for each operation log segment to generate the read and write speed of each storage unit;

[0121] Perform read and write delay distribution analysis based on the read and write speed of each storage unit to generate storage unit read and write delay distribution data;

[0122] Based on the storage unit read and write delay distribution data, the data throughput time series change analysis is performed on the operation log fragment of each storage unit to obtain the throughput time series change data of each storage unit;

[0123] Perform data access behavior detection on the operation log fragments of each storage unit and extract the data access behavior characteristics of each storage unit;

[0124] Access error identification is performed on the data access behavior characteristics of each storage unit to obtain storage unit access error data;

[0125] Classifying the error type of storage unit access error data to obtain the access error type;

[0126] Calculate the access error frequency according to the access error type and generate the access error classification frequency;

[0127] The storage unit status feature mining is performed on the throughput time series change data and access error classification frequency of each storage unit to generate the status feature of each storage unit.

[0128] In this embodiment, it is ensured that the operation log fragment of each storage unit has been extracted. The log should contain information such as the timestamp, data size, and operation type (read / write) of the read and write operations. For each log fragment of each storage unit, the read and write speed is calculated. The read and write speed can be calculated by the following formula: Read and write speed = Operation completion time / data size, traverse all read and write operations, record the time and size of each operation, calculate the speed of each operation, store the read and write speed information of each storage unit in structured data, such as a database table or data frame, for subsequent analysis, extract the timestamp of each read and write operation from the operation log, calculate the delay, that is, the difference between the start time and the end time of the operation, use statistical analysis methods (such as histograms or probability density functions) to analyze the read and write delay distribution of each storage unit, and calculate indicators such as the average, standard deviation, maximum and minimum values ​​of the delay. Integrate the read and write speeds of all read and write operations into time series data to ensure that they are arranged in chronological order. The throughput can be obtained by adding all read and write speeds within a certain time window to form time series data. For example, for the throughput per minute, the total read and write speed within the minute can be recorded at the end of each minute. The calculated throughput time series change data is stored in structured data for subsequent analysis. Extract relevant information of all read and write operations from the operation log, including operation type, access time, data block address, etc. According to the access time, frequency and data block address, detect each The access behavior characteristics of each storage unit, which may include access frequency, access mode (sequential or random), etc., extract possible error records from the operation log, such as read and write failures, timeouts, etc., identify access errors based on specific keywords or error codes in the log, and set a series of conditions to identify errors, such as operations that exceed the response time or failed read and write requests. Extract all identified error records from the access error data and classify them according to the nature of the error (such as hardware failure, software failure, timeout, etc.). A classification system can be built to facilitate the assignment of errors to different categories. The classification results of each error are stored in structured data for subsequent frequency calculation and analysis. According to the classification results of the error type, the frequency of occurrence of each type of error is counted. For example, the number of occurrences of each error type within a certain time window is calculated. The throughput timing change data of the storage unit is combined with the access error classification frequency for comprehensive analysis to extract deeper state features. By analyzing the throughput change and error frequency, the state features of each storage unit are extracted, such as: overall performance (comprehensive performance of throughput and latency), and the relationship between error rate and performance.

[0129] In this embodiment, refer to Figure 4 , is a flowchart of detailed implementation steps of step S3. In this embodiment, the detailed implementation steps of step S3 include:

[0130] Step S31: Calculate the data request waiting queue length for the operation log fragment of each storage unit to obtain the data request waiting queue length of each storage unit;

[0131] Step S32: Calculate the periodic access load peak value according to the length of the data request waiting queue of each storage unit to obtain the access load peak value of each storage unit;

[0132] Step S33: performing data access stress test analysis based on the access load peak value of each storage unit to generate an access stress evaluation value of the storage unit;

[0133] Step S34: Perform data access simulation according to the access pressure evaluation value of the storage unit, thereby generating a simulated access data packet for each storage unit.

[0134] In this embodiment, ensure that the operation log fragments of each storage unit have been extracted. These logs should contain information such as the timestamp of the request, the request type and the response time. According to the timestamp information, build the request waiting queue of each storage unit. The length of the request waiting queue can be defined as the number of requests in a waiting state within a specific time period. Traverse the logs of each storage unit to record the start time and end time of each request. For each time point, check whether there is a request currently waiting. If so, count it into the waiting queue length. Use the sliding window method to set a time period (for example, 1 second). Count the number of requests within this time period to obtain the queue length. Obtain the request waiting queue length data calculated for each storage unit in different time periods. Use the statistical analysis method to analyze the request waiting queue length of each storage unit in different time periods to determine the peak value of the access load. For example, the maximum waiting queue per minute can be calculated. The column length is used as the access load peak value of the time period. According to the access load peak value, the parameters of the stress test are set, such as the access frequency, duration and request type of the test. A simulation tool or a custom script is used to initiate a request to the storage unit to simulate the access situation under the peak load. The response time, error rate and throughput of the storage unit under this load are monitored. According to the results of the stress test, the access stress evaluation value is calculated, which can be a comprehensive evaluation of indicators such as the average value, maximum value and error rate of the response time. According to the stress evaluation value, the parameters of the simulated access are set, such as the number, type and interval time of requests, to reflect the actual usage. A program for simulating access is designed to generate corresponding request data packets, which should include information such as the access address, request type (read / write), data size, etc. Run the simulation program and record the detailed information of each request generated during the simulation, including the response time of the request, success or failure, etc.

[0135] In this embodiment, step S4 includes the following steps:

[0136] Step S41: Based on the state evolution law of each storage unit, dynamic state mapping modeling is performed on the storage unit sequence of each partition to construct multiple storage unit digital twin models;

[0137] Step S42: using the simulated access data packet of each storage unit to perform dynamic pressure access simulation on multiple storage unit digital twin models, and collecting storage unit access simulation response data;

[0138] Step S43: decomposing the storage unit access simulation response data into multiple time periods, and extracting the access simulation response data of multiple time periods;

[0139] Step S44: performing storage unit response status evaluation on the access simulation response data of multiple time periods to obtain response status evaluation data of each storage unit.

[0140] In this embodiment, simulated access data packets of each storage unit are obtained from the previous steps. These data packets should contain request information representing the actual access mode. The simulated access data packets of each storage unit are used to perform dynamic pressure access simulation on the constructed digital twin model. The simulation should include different load scenarios to evaluate the behavior of the storage unit under different pressures. During the simulation, the response time, load conditions and other relevant performance indicators of the storage unit are recorded in real time. The storage unit access response data collected during the simulation are sorted, including request processing time, success rate and error information, etc., and stored in structured data for subsequent analysis. Simulated access data packets of each storage unit are obtained from the previous steps. These data packets should contain request information representing the actual access mode. The simulated access data packets of each storage unit are used to perform dynamic pressure access simulation on the constructed digital twin model. The simulation should include different load scenarios to evaluate the behavior of the storage unit under different pressures. During the simulation, the response time, load conditions and other relevant performance indicators of the storage unit are recorded in real time. The collected storage unit access response data is sorted, including request processing time, success rate and error information, and stored in structured data for subsequent analysis. The response data is decomposed into multiple time periods (for example, by minute, hour or specific event) according to needs to analyze performance changes in different time periods. The response data in each time period is summarized and key performance indicators (KPIs) are extracted, such as average response time, maximum response time, number of errors, etc., to ensure that the data in each time period can reflect the performance status of the storage unit. The access simulation response data of multiple time periods are integrated for comprehensive evaluation and indicators for evaluating the response status of the storage unit are determined, such as: response time: average response time of access requests, success rate: proportion of successfully processed requests, error rate: proportion of failed requests, statistical analysis is performed on the integrated data, and the response status evaluation data of each storage unit is calculated. Visual tools (such as charts and dashboards) can be used to help display the evaluation results. The response status evaluation data of each storage unit is saved to the database or related documents for subsequent reference and decision support.

[0141] In this embodiment, step S44 includes the following steps:

[0142] Calculating the unit access response speed for the access simulation response data of multiple time periods to obtain the unit access response speeds of multiple time periods;

[0143] Perform access delay analysis on access response speeds of multiple time period units and extract access delay data for each time period;

[0144] Perform delay variation analysis on the access delay data of each time period to obtain the unit access delay characteristics;

[0145] Calculate the data access accuracy of access simulation response data in multiple time periods to obtain the access accuracy of each time period;

[0146] The access accuracy rate of each time period is identified by the time period accuracy distribution to generate multi-time period accuracy distribution data;

[0147] The storage unit response status is evaluated based on the unit access delay characteristics and multi-period accuracy distribution data to obtain response status evaluation data of each storage unit.

[0148] In this embodiment, the access response speed can be defined as the inverse of the processing time of each request, and the calculation formula is: response speed = 1 / response time. The access response data in each time period is traversed to calculate the response speed of each request, and all requests in each time period are averaged to obtain the unit access response speed of the time period. The access delay can be defined as the response time of the request, which usually means the time from the request initiation to the receipt of the response. The access response data of each time period is statistically analyzed, and the access delay data of each time period is extracted, including the average delay, maximum delay, minimum delay and standard deviation of the delay, etc. The access delay data of each time period is analyzed for changes to identify the rising and falling trends of the delay. Visual tools (such as charts) can be used to display the delay changes. The delay change rate is calculated, and the delay change rate = {(current delay − previous time period delay) / previous time period delay}. The delay characteristics (such as change rate, trend) of each time period are stored in structured data for subsequent status evaluation. The access accuracy rate can be defined as the ratio of the number of successful requests to the total number of requests. The access accuracy rate = Total number of requests / number of successful requests, count the request data in each time period, calculate the access accuracy and record the results of each time period, perform accuracy distribution analysis, use histograms or other visualization methods to identify the distribution of accuracy, calculate the frequency in each accuracy range to understand the overall performance of access accuracy, store the accuracy distribution data of multiple time periods in structured data for subsequent evaluation and report generation, integrate latency characteristics, accuracy data and other related performance indicators for comprehensive evaluation, and identify key indicators for evaluating the response status of storage units, such as: average response speed: the average response speed of all time periods, average delay: the average delay of all time periods, accuracy: the average access accuracy of all time periods.

[0149] In this embodiment, the specific steps of step S5 are:

[0150] Step S51: mining the state trend evolution of the response state evaluation data of each storage unit to generate storage unit response state trend evolution data;

[0151] Step S52: performing time series evolution fitting on the storage unit response state trend evolution data to construct a state trend evolution curve;

[0152] Step S53: performing state trend mutation identification on the state trend evolution curve, and extracting the state trend mutation point of the storage unit;

[0153] Step S54: based on the state trend mutation point of the storage unit, locate the abnormal mutation partition of the solid state hard disk to obtain the abnormal mutation sector.

[0154] In this embodiment, response status evaluation data of the storage unit is extracted from the previous analysis. The data may include multiple performance indicators, such as response speed: such as read and write speed, latency: average time for request processing, accuracy: ratio of request success to failure, ensure that the data is arranged in chronological order for subsequent analysis, apply the moving average method to each indicator to smooth short-term fluctuations and highlight long-term trends, select an appropriate window size (such as 5 time periods), calculate the average value at each time point in the window, use the linear regression method to model, use time as the independent variable and the performance indicator as the dependent variable, determine the best fit line by the least squares method, analyze the slope of the trend to determine whether the indicator is rising, falling or remaining stable, use a chart (such as a line chart) to show the trend of each performance indicator over time, so that the trend is more intuitive and easy to understand, and organize step S51 The state trend evolution data generated in the process is ensured to be complete and arranged in chronological order for time series fitting. According to the characteristics of the data, models such as linear regression, exponential fitting or polynomial regression can be selected. For rapidly changing data, exponential fitting may be more appropriate, while for steadily changing data, linear regression is more effective. The least squares method or other statistical methods are used to estimate the model parameters. The fitting effect is tested by calculating the residual (the difference between the actual value and the fitted value). The goodness of fit (such as R² value) is selected as the evaluation criterion to ensure the validity of the model. The state trend evolution curve data generated by the fitting is saved as structured data, which may include the fitting parameters. , predicted values ​​and residuals, to facilitate subsequent analysis and reference, set mutation recognition standards, for example: change rate threshold: set a threshold, when the rate of change of the curve in a short period of time exceeds this threshold, it is regarded as a mutation, inflection point detection: calculate the derivative of the curve, and find the point of sharp change. The second-order derivative can help identify changes in acceleration or deceleration. Use a threshold-based method to detect mutation points, or use a more complex algorithm (such as the CUSUM algorithm) to monitor state changes, which can improve the sensitivity and accuracy of mutation detection. Record the timestamp, value and change amplitude of the identified state trend mutation point in structured data for subsequent analysis and positioning. Extract the state trend mutation point data of the storage unit from step S53, confirm its accuracy and completeness, and use it for abnormal partition positioning. Combine the mutation point with the operation log of the storage unit, find the operation record (such as read and write request, error information, etc.) near the mutation point, understand the state of the storage unit when the mutation occurs, partition the storage unit according to the time and nature of the mutation point, and identify the sectors that may be affected. For example, if a sector frequently has errors or the response time increases significantly when a mutation occurs, then the sector may have problems.

[0155] In this embodiment, the specific steps of step S6 are:

[0156] Step S61: predicting the fault probability of the abnormal mutation sector, thereby obtaining abnormal sector fault prediction data;

[0157] Step S62: performing dynamic propagation path evolution analysis on the abnormal sector fault prediction data to obtain fault propagation path evolution data;

[0158] Step S63: performing partition propagation path distribution analysis on the solid state drive according to the fault propagation path evolution data to obtain the solid state drive fault propagation path distribution data;

[0159] Step S64: performing a fault risk assessment on the fault propagation path distribution data of the solid state drive to obtain a fault prediction assessment result.

[0160] In this embodiment, according to the information of the identified abnormal mutation sector, it is ensured that the status, historical fault records, and performance indicators (such as response time, read and write speed, etc.) of each sector are included, and a suitable fault prediction model is selected, such as: statistical model: such as logistic regression or survival analysis, which is used to estimate the probability of fault occurrence, machine learning model: such as decision tree, random forest or neural network, use historical data to train the model to predict faults, determine key features that affect the probability of faults, such as: read and write times, error rate, delay, temperature, etc., divide the data into a training set and a test set, usually 70% for training and 30% for testing, use the training set data to train the selected model, and optimize the hyperparameters to improve the prediction accuracy, evaluate the model performance on the test set, use indicators such as accuracy, recall rate, F1-score, etc. to verify the effectiveness of the model, apply the trained model to the data of the abnormal mutation sector, calculate the failure probability of each sector, generate abnormal sector fault prediction data, identify and define the concept of fault propagation path, and the fault may be propagated in different ways, such as: physical propagation: such as failure of adjacent sectors caused by electrical faults, logical propagation: such as failure of adjacent sectors caused by software faults, etc. Multiple storage unit anomalies caused by faults are detected. Graph theory or network analysis methods are used to construct a mathematical model of fault propagation. Nodes represent storage units and edges represent potential propagation paths. By simulating the fault propagation process, how the fault propagation path evolves under different conditions (such as load changes and environmental factors) is analyzed. The propagation path and its changes in each simulation are recorded to generate fault propagation path evolution data. According to the collected propagation path data, the fault propagation path in each partition is counted, for example: the number of fault propagations and the impact range in each partition are calculated. Heat maps or charts are used to show the distribution of fault propagation in different partitions to help identify high-risk areas and determine key indicators for fault risk assessment, such as: comprehensive failure probability: the failure probability calculated based on the prediction model, propagation potential: the fault impact range evaluated based on the propagation path data, the risk score of each storage unit is calculated by weighted average or other methods, various indicators are comprehensively considered, the risk score of each storage unit is analyzed, high-risk units are identified, and corresponding preventive measures are formulated. The fault prediction assessment results (including risk scores and high-risk unit lists) are stored in a structured format for subsequent reference and decision support.

[0161] In this embodiment, a solid state drive fault prediction system based on artificial intelligence is provided, which is used to execute the solid state drive fault prediction method based on artificial intelligence as described above, including:

[0162] The storage unit partitioning module is used to perform fine-grained partitioning of the storage units of the solid-state drive, serialize the storage units, and construct a storage unit sequence for each partition;

[0163] The dynamic evolution learning module is used to obtain the SSD operation log; based on the SSD operation log, the storage unit sequence of each partition is subjected to multi-time point state evolution, and deep dynamic evolution learning is performed to generate the state evolution law of each storage unit;

[0164] The load peak module is used to calculate the periodic access load peak of the SSD operation log and perform data access stress test analysis to generate simulated access data packets for each storage unit;

[0165] A dynamic access simulation module is used to perform dynamic pressure access simulation based on the state evolution law of each storage unit and the simulated access data packet of each storage unit to obtain response state evaluation data of each storage unit;

[0166] The state trend evolution mining module is used to mine the state trend evolution of the response state evaluation data of each storage unit, and locate the abnormal mutation partition to obtain the abnormal mutation sector;

[0167] The fault prediction module is used to predict the fault probability of abnormal mutation sectors and perform fault risk assessment to obtain fault prediction assessment results.

[0168] The present invention helps to monitor the hard disk state more finely and improve the accuracy of anomaly detection by dividing the storage unit in fine granularity. The serialized storage unit can provide ordered data for subsequent analysis, which is convenient for learning and predicting the state evolution law. Obtaining the operation log for state evolution learning helps to monitor the hard disk state in real time and discover potential faults. By learning the state evolution law of the storage unit, the hard disk state change can be predicted in advance and the risk of failure can be reduced. Calculating the load peak and stress test analysis helps to understand the working condition and load of the hard disk. Generating simulated access data packets can help simulate real data access scenarios and evaluate the performance and stability of the hard disk. Dynamic stress access simulation based on the state evolution law and the simulated access data packet can simulate the hard disk response under the real workload. Obtaining the response state evaluation data of the storage unit helps to monitor the hard disk state in real time and discover potential faults in time. Mining the state trend evolution of the response state evaluation data can reveal the law and trend of the storage unit state change. Abnormal mutation partition positioning helps to accurately locate abnormal situations, early warning and processing of hard disk abnormalities. Fault probability prediction and risk assessment help to predict hard disk failures in advance and reduce the risk of accidental data loss and hard disk damage. The fault prediction evaluation result can guide the formulation of maintenance strategies and improve the reliability and stability of solid-state hard disks.

[0169] Therefore, the embodiments should be regarded as illustrative and non-restrictive from all points, and the scope of the present invention is limited by the appended claims rather than the above description, and it is therefore intended that all changes falling within the meaning and range of equivalent elements of the application documents are included in the present invention.

[0170] The above is only a specific embodiment of the present invention, so that those skilled in the art can understand or implement the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features invented herein.

Claims

1. A solid state drive failure prediction method based on artificial intelligence, characterized in that: The following steps are involved: Step S1: fine-grained partitioning of the solid state drive into storage units, and serialization of the storage units to construct a storage unit sequence for each partition; Step S2: obtaining the SSD operation log; performing multi-time point state evolution on the storage unit sequence of each partition based on the SSD operation log, and performing deep dynamic evolution learning to generate the state evolution law of each storage unit; Step S3: Calculate the periodic access load peak value of the solid state drive operation log, and perform data access stress test analysis to generate a simulated access data packet for each storage unit; Step S4: performing dynamic pressure access simulation based on the state evolution law of each storage unit and the simulated access data packet of each storage unit to obtain response state evaluation data of each storage unit; Step S5: mining the state trend evolution of the response state evaluation data of each storage unit, and locating the abnormal mutation partition to obtain the abnormal mutation sector; the specific steps of step S5 are: Step S51: mining the state trend evolution of the response state evaluation data of each storage unit to generate storage unit response state trend evolution data; Step S52: performing time series evolution fitting on the storage unit response state trend evolution data to construct a state trend evolution curve; Step S53: performing state trend mutation identification on the state trend evolution curve, and extracting the state trend mutation point of the storage unit; Step S54: locating abnormal mutation partitions of the solid state hard disk based on the state trend mutation points of the storage units to obtain abnormal mutation sectors; Step S6: predict the fault probability of the abnormal mutation sector and perform a fault risk assessment to obtain a fault prediction assessment result.

2. The artificial intelligence-based solid-state drive fault prediction method according to claim 1, characterized in that: The specific steps of step S1 are: Step S11: performing physical address mapping partition identification on the solid state hard disk, thereby obtaining the sector partition data of the solid state hard disk; Step S12: performing fine-grained division of the solid state hard disk sector partition data into storage units, and extracting the storage units of each partition; Step S13: performing unit sequence coding on the storage unit of each partition to obtain the sequence coding of each storage unit; Step S14: Serialize the storage units based on the sequential coding of each storage unit to construct a storage unit sequence for each partition.

3. The artificial intelligence-based solid-state drive fault prediction method according to claim 1, characterized in that: The specific steps of step S2 are: Step S21: Obtaining the solid state drive operation log; Step S22: performing sliding filtering and smoothing on the solid state drive operation log to generate a filtering and smoothing operation log; Step S23: performing storage unit feature matching on the filtered smoothing operation log based on the storage unit sequence of each partition, thereby obtaining an operation log segment of each storage unit; Step S24: mining the state characteristics of each storage unit on the operation log fragment of each storage unit to generate the state characteristics of each storage unit; Step S25: performing multi-time point state evolution on the state characteristics of each storage unit, thereby obtaining multiple time point state evolution characteristics; Step S26: Perform deep dynamic evolution learning on the state evolution features at multiple time points to generate the state evolution law of each storage unit.

4. The artificial intelligence-based solid-state drive fault prediction method according to claim 3, characterized in that: The specific steps of step S24 are: Calculate the read and write speed of each storage unit for each operation log segment to generate the read and write speed of each storage unit; Perform read and write delay distribution analysis based on the read and write speed of each storage unit to generate storage unit read and write delay distribution data; Based on the storage unit read and write delay distribution data, the data throughput time series change analysis is performed on the operation log fragment of each storage unit to obtain the throughput time series change data of each storage unit; Perform data access behavior detection on the operation log fragments of each storage unit and extract the data access behavior characteristics of each storage unit; Access error identification is performed on the data access behavior characteristics of each storage unit to obtain storage unit access error data; Classifying the error type of storage unit access error data to obtain the access error type; Calculate the access error frequency according to the access error type and generate the access error classification frequency; The storage unit status feature mining is performed on the throughput time series change data and access error classification frequency of each storage unit to generate the status feature of each storage unit.

5. The artificial intelligence-based solid-state drive fault prediction method according to claim 1, characterized in that: The specific steps of step S3 are: Step S31: Calculate the data request waiting queue length for the operation log fragment of each storage unit to obtain the data request waiting queue length of each storage unit; Step S32: Calculate the periodic access load peak value according to the length of the data request waiting queue of each storage unit to obtain the access load peak value of each storage unit; Step S33: performing data access stress test analysis based on the access load peak value of each storage unit to generate an access stress evaluation value of the storage unit; Step S34: Perform data access simulation according to the access pressure evaluation value of the storage unit, thereby generating a simulated access data packet for each storage unit.

6. The artificial intelligence-based solid-state drive fault prediction method according to claim 1, characterized in that: The specific steps of step S4 are: Step S41: Based on the state evolution law of each storage unit, dynamic state mapping modeling is performed on the storage unit sequence of each partition to construct multiple storage unit digital twin models; Step S42: using the simulated access data packet of each storage unit to perform dynamic pressure access simulation on multiple storage unit digital twin models, and collecting storage unit access simulation response data; Step S43: decomposing the storage unit access simulation response data into multiple time periods, and extracting the access simulation response data of multiple time periods; Step S44: performing storage unit response status evaluation on the access simulation response data of multiple time periods to obtain response status evaluation data of each storage unit.

7. The artificial intelligence-based solid-state drive fault prediction method according to claim 6, characterized in that: The specific steps of step S44 are: Calculating the unit access response speed for the access simulation response data of multiple time periods to obtain the unit access response speeds of multiple time periods; Perform access delay analysis on access response speeds of multiple time period units and extract access delay data for each time period; Perform delay variation analysis on the access delay data of each time period to obtain the unit access delay characteristics; Calculate the data access accuracy of access simulation response data in multiple time periods to obtain the access accuracy of each time period; The access accuracy rate of each time period is identified by the time period accuracy distribution to generate multi-time period accuracy distribution data; The storage unit response status is evaluated based on the unit access delay characteristics and multi-period accuracy distribution data to obtain response status evaluation data of each storage unit.

8. The artificial intelligence-based solid-state drive fault prediction method according to claim 1, characterized in that: The specific steps of step S6 are: Step S61: predicting the fault probability of the abnormal mutation sector, thereby obtaining abnormal sector fault prediction data; Step S62: performing dynamic propagation path evolution analysis on the abnormal sector fault prediction data to obtain fault propagation path evolution data; Step S63: performing partition propagation path distribution analysis on the solid state drive according to the fault propagation path evolution data to obtain the solid state drive fault propagation path distribution data; Step S64: performing a fault risk assessment on the fault propagation path distribution data of the solid state drive to obtain a fault prediction assessment result.

9. A solid state drive failure prediction system based on artificial intelligence, characterized in that: Used to execute the artificial intelligence-based solid-state drive fault prediction method as claimed in claim 1, comprising: The storage unit partitioning module is used to perform fine-grained partitioning of the storage units of the solid-state drive, serialize the storage units, and construct a storage unit sequence for each partition; The dynamic evolution learning module is used to obtain the SSD operation log; based on the SSD operation log, the storage unit sequence of each partition is subjected to multi-time point state evolution, and deep dynamic evolution learning is performed to generate the state evolution law of each storage unit; The load peak module is used to calculate the periodic access load peak of the SSD operation log and perform data access stress test analysis to generate simulated access data packets for each storage unit; A dynamic access simulation module is used to perform dynamic pressure access simulation based on the state evolution law of each storage unit and the simulated access data packet of each storage unit to obtain response state evaluation data of each storage unit; The state trend evolution mining module is used to mine the state trend evolution of the response state evaluation data of each storage unit, and locate the abnormal mutation partition to obtain the abnormal mutation sector; The fault prediction module is used to predict the fault probability of abnormal mutation sectors and perform fault risk assessment to obtain fault prediction assessment results.

Images