Fault prediction method and related equipment

Abstract

The invention provides a fault prediction method and related equipment, which are used for improving the accuracy of fault prediction. The fault prediction method comprises the steps of obtaining historical fault data of a storage medium, wherein the historical fault data is used for indicating historical fault information of the storage medium; and converting the historical fault data into image feature data corresponding to the storage medium. And inputting the image feature data into the fault prediction model to obtain a prediction result which is used for predicting the health degree of the storage medium.

Description

[0001] This application claims priority to the Chinese patent application filed with the State Intellectual Property Office on October 17, 2023, with application number 202311350303.1 and invention name “A fault monitoring method, device and computing device cluster”, all contents of which are incorporated by reference in this application. Technical Field

[0002] The present application relates to the field of computers, and in particular to a fault prediction method and related equipment. Background Art

[0003] In today's large-scale data centers, memory failures occur frequently, causing a decrease in the stability and reliability of servers and even the entire information technology (IT) infrastructure, which has a negative impact on the service performance of the business. Therefore, whether memory failures can be predicted in advance is an urgent problem to be solved.

[0004] In one fault prediction method, historical data is screened, and features corresponding to the screened historical data are manually constructed, and then the random forest algorithm is used for training and modeling. In this method, relying on manual data screening and feature construction may miss data, resulting in incomplete correlation between faults and reducing the accuracy of fault prediction. Summary of the invention

[0005] The present application provides a fault prediction method and related equipment. In the fault prediction method, historical fault data is directly converted into image feature data, that is, the full amount of fault information is converted. In other words, the data used to train the model is comprehensive, thereby improving the accuracy of the fault prediction model.

[0006] In a first aspect, the present application provides a fault prediction method, comprising:

[0007] Obtain historical fault data of the storage medium, and the historical fault data is used to indicate the historical fault information of the storage medium. The historical fault information includes information reflecting the specific content of the fault, such as the time when the fault occurred, the type of fault, and the address of the fault. Convert the historical fault data into image feature data corresponding to the storage medium, and the image feature data is used to predict faults. Specifically, the image feature data is input into a fault prediction model to obtain a prediction result, and the prediction result is used to predict the health of the storage medium. In other words, the input of the fault prediction model is image feature data, and the prediction model uses an image recognition algorithm. Optionally, the prediction result can be expressed in a variety of forms, which can be a score for the health of the storage medium, or a prediction of the possibility of a failure of the storage medium. In addition, it can also be other ways that can indicate the health level, such as the health level of the storage medium, etc., which are not limited here.

[0008] In this application, directly converting historical fault data into image feature data means converting the entire amount of fault information, that is, the data used to train the model is comprehensive, which improves the accuracy of the fault prediction model, and thus improves the accuracy of fault prediction. In addition, converting historical fault data into image feature data means that the fault prediction model uses an image classification algorithm, which is a mature algorithm. Applying this algorithm in the field of fault prediction also improves the effectiveness of the algorithm.

[0009] In some optional implementations of the first aspect, converting the historical fault data into image feature data corresponding to the storage medium includes: obtaining the number of faults of each fault address corresponding to the historical fault data. The number of faults corresponding to each fault address mentioned here refers to the number of faults corresponding to each fault address in the historical fault information indicated by the historical fault data. The historical fault data is converted into image feature data according to the address space of the storage medium, each fault address corresponding to the historical fault data, and the number of faults of each fault address.

[0010] In the present application, when converting historical fault data into image feature data corresponding to the storage medium, the number of faults at each fault address corresponding to the historical fault data is combined, which can reflect the distribution of historical faults at different fault addresses, provide a more accurate basis for fault prediction, and further improve the accuracy of fault prediction.

[0011] In some optional implementations of the first aspect, historical fault data is converted into image feature data according to the address space of the storage medium, each fault address, and the number of faults of each fault address. Specifically, the position of each fault address in the address space of the storage medium is determined. It can be understood that the address space of the storage medium can be represented by a matrix form of (rows, columns), and the fault address is also represented in this form. Then, corresponding to the image, the address space of a storage medium has a corresponding image. The image space is divided based on rows and columns, and corresponds to the address space of the storage medium one by one, so that the position of each fault address can be determined on the image. Then, the number of faults of each fault address is filled into the position of each fault address in the address space of the storage medium (that is, the position in the image of the address space of the corresponding storage medium), and the image data to be compressed is obtained. The image data to be compressed is compressed to obtain image feature data.

[0012] In the present application, the association between the address representation of the storage medium and the image is utilized to determine the corresponding position of the address space of the storage medium on the image, thereby determining the position of each fault address on the image, providing technical support for converting historical fault data into image feature data and improving the feasibility of the technical solution of the present application.

[0013] In some optional implementations of the first aspect, compressing the image data to be compressed includes: compressing the rows and / or columns of the image data to be compressed, that is, compressing the rows of the image to be compressed, or compressing the columns of the image to be compressed, or compressing both the rows and columns of the image to be compressed. Wherein, in the image data to be compressed, the mean number of failures or the sum of the number of failures of each area to be compressed is the number of failures of each position corresponding to each area to be compressed in the compressed image data. In other words, in the compressed image data, the number of failures of each position is the mean number of failures or the sum of the number of failures of the area to be compressed corresponding to the position in the image to be compressed. The present application does not limit the size of the compressed image feature data, and it can also be determined according to the needs of the actual application. For example, for a device with good processing performance, the degree of compression can be less than that of a device with poor processing performance. No specific limitation is made here.

[0014] In the present application, there are multiple ways to compress the compressed image data, and there are also multiple possible degrees of compression, which can adapt to different needs and enrich the implementation methods of the technical solutions of the present application.

[0015] In some optional implementations of the first aspect, the historical fault data is fault data within a target time window. In this solution, the number of faults of each fault address refers to the number of faults of each fault address within the target time window. Among them, the specific duration and time period of the target time window can be set according to the needs of actual application, and are not specifically limited here. Exemplarily, the target time window can be one day, one week, or other shorter or longer time periods, which are not specifically limited here.

[0016] In some optional implementations of the first aspect, after inputting the image feature data into the fault prediction model to obtain the prediction result, in the case where the storage medium predicted by the prediction result is unhealthy, the storage medium can also be tested to obtain the test result. Among them, there are many possible situations in which the storage medium predicted by the prediction result is unhealthy, which can be that the score corresponding to the prediction result is greater than the unhealthy threshold, or the score corresponding to the prediction result is less than the healthy threshold, or the probability corresponding to the prediction result is less than the healthy probability threshold. In addition, there can be other situations indicating that the storage medium is unhealthy, which are not limited here. In addition, the unhealthy storage medium means that the storage medium is about to fail.

[0017] In the present application, the storage medium is tested, which can not only timely discover whether the storage medium is indeed faulty, but also prevent the adverse effects caused by misjudgment of the fault prediction model, further improving the practicality of the technical solution of the present application.

[0018] In some optional implementations of the first aspect, the health status indicated by the test result is used as a label of the image feature data and input into the fault prediction model, and the image feature data and the label are used to iteratively train the fault prediction model.

[0019] In this application, the test results are used as labels for image feature data to iteratively train the fault prediction model, which can not only enrich the training data set, but also further improve the accuracy of the fault prediction model because the test results are verified results and represent the actual situation.

[0020] In a second aspect, the present application provides a method for training a fault prediction model, comprising:

[0021] Obtain historical fault data of the storage medium, where the historical fault data is used to indicate historical fault information of the storage medium. Convert the historical fault data into image feature data corresponding to the storage medium. Iterate and train the fault prediction model according to the image feature data and the true value corresponding to the image feature data until the fault prediction model converges or reaches a preset number of iterations.

[0022] In this application, directly converting historical fault data into image feature data means converting the entire amount of fault information, that is, the data used to train the model is comprehensive, which improves the accuracy of the fault prediction model. In addition, converting historical fault data into image feature data means that the fault prediction model uses an image classification algorithm, which is a mature algorithm. Applying this algorithm in the field of fault prediction also improves the effectiveness of the algorithm.

[0023] In some optional implementations of the second aspect, converting the historical fault data into image feature data corresponding to the storage medium includes: obtaining the number of faults of each fault address corresponding to the historical fault data. The number of faults corresponding to each fault address mentioned here refers to the number of faults corresponding to each fault address in the historical fault information indicated by the historical fault data. The historical fault data is converted into image feature data according to the address space of the storage medium, each fault address corresponding to the historical fault data, and the number of faults of each fault address.

[0024] In the present application, when converting historical fault data into image feature data corresponding to the storage medium, the number of faults at each fault address corresponding to the historical fault data is combined, which can reflect the distribution of historical faults at different fault addresses, provide more accurate training samples for the fault prediction model, and further improve the accuracy of the fault prediction model.

[0025] In some optional implementations of the second aspect, historical fault data is converted into image feature data according to the address space of the storage medium, each fault address, and the number of faults of each fault address within the target time window. Specifically, the position of each fault address in the address space of the storage medium is determined. It can be understood that the address space of the storage medium can be represented by a matrix form of (rows, columns), and the fault address is also represented in this form. Then, corresponding to the image, the address space of a storage medium has a corresponding image. The image space is divided based on rows and columns, and corresponds to the address space of the storage medium one by one, so that the position of each fault address can be determined on the image. Then, the number of faults of each fault address is filled into the position of each fault address in the address space of the storage medium (that is, the position in the image of the address space of the corresponding storage medium), and the image data to be compressed is obtained. The image data to be compressed is compressed to obtain image feature data.

[0026] In the present application, the association between the address representation of the storage medium and the image is utilized to determine the corresponding position of the address space of the storage medium on the image, thereby determining the position of each fault address on the image, providing technical support for converting historical fault data into image feature data and improving the feasibility of the technical solution of the present application.

[0027] In some optional implementations of the second aspect, compressing the image data to be compressed includes: compressing the rows and / or columns of the image data to be compressed, that is, compressing the rows of the image to be compressed, or compressing the columns of the image to be compressed, or compressing both the rows and columns of the image to be compressed. Wherein, in the image data to be compressed, the mean number of failures or the sum of the number of failures of each area to be compressed is the number of failures of each position corresponding to each area to be compressed in the compressed image data. In other words, in the compressed image data, the number of failures of each position is the mean number of failures or the sum of the number of failures of the area to be compressed corresponding to the position in the image to be compressed. The present application does not limit the size of the compressed image feature data, and it can also be determined according to the needs of the actual application. For example, for a device with good processing performance, the degree of compression can be less than that of a device with poor processing performance. No specific limitation is made here.

[0028] In the present application, there are multiple ways to compress the compressed image, and there are also multiple possible degrees of compression, which can adapt to different needs and enrich the implementation methods of the technical solutions of the present application.

[0029] In a third aspect, the present application provides a fault prediction device, comprising:

[0030] An acquisition unit, used for acquiring historical fault data of a storage medium, where the historical fault data is used for indicating historical fault information of the storage medium;

[0031] The processing unit is used to convert the historical fault data into image feature data corresponding to the storage medium; input the image feature data into the fault prediction model to obtain the prediction result, and the prediction result is used to predict the health of the storage medium.

[0032] The communication device is used to implement the method shown in the aforementioned first aspect or the first aspect or any possible implementation method of the first aspect. The beneficial effects shown in this aspect are similar to those of the first aspect or any possible implementation method of the first aspect, and will not be repeated here.

[0033] In a fourth aspect, the present application provides a training device for a fault prediction model, characterized in that it includes:

[0034] An acquisition unit, used for acquiring historical fault data of a storage medium, where the historical fault data is used for indicating historical fault information of the storage medium;

[0035] The processing unit is used to convert the historical fault data into image feature data corresponding to the storage medium; iteratively train the fault prediction model according to the image feature data and the true value corresponding to the image feature data until the fault prediction model converges or reaches a preset number of iterations.

[0036] The training device is used to implement the method shown in the aforementioned second aspect or any possible implementation of the second aspect or the first aspect. The beneficial effects shown in this aspect are similar to those of the second aspect or any possible implementation of the second aspect, and will not be repeated here.

[0037] In a fifth aspect, the present application provides an electronic device comprising a processor and a memory, wherein the processor stores instructions. When the instructions stored in the memory are executed on the processor, the method shown in the aforementioned first aspect, any possible implementation of the first aspect, the second aspect, or any possible implementation of the second aspect is implemented.

[0038] In a sixth aspect, the present application provides a computer-readable storage medium, which stores instructions. When the instructions are executed on a processor, the method shown in the aforementioned first aspect, any possible implementation of the first aspect, the second aspect, or any possible implementation of the second aspect is implemented.

[0039] In a seventh aspect, the present application provides a computer program product, which, when executed on a processor, implements the method shown in the aforementioned first aspect, any possible implementation of the first aspect, the second aspect, or any possible implementation of the second aspect.

[0040] The beneficial effects shown in any one of the fifth to seventh aspects are similar to those of the first aspect, any possible implementation of the first aspect, the second aspect, or any possible implementation of the second aspect, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 A schematic diagram of the structure of the artificial intelligence main framework of the embodiment of the present application;

[0042] Figure 2 A schematic diagram of a system architecture of a fault prediction system provided in an embodiment of the present application;

[0043] Figure 3 A schematic diagram of a flow chart of a fault prediction method provided in an embodiment of the present application;

[0044] Figure 4 A schematic diagram of a fault prediction method provided in an embodiment of the present application;

[0045] Figure 5 A flowchart of a method for training a fault prediction model provided in an embodiment of the present application;

[0046] Figure 6 A schematic diagram of the structure of a fault prediction device provided in an embodiment of the present application;

[0047] Figure 7 A schematic diagram of the structure of a training device for a fault prediction model provided in an embodiment of the present application;

[0048] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application;

[0049] Fig. 9 A schematic diagram of the structure of the chip provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] The embodiments of the present application provide a fault prediction method and related devices for improving the accuracy of fault prediction.

[0051] The embodiments of the present application are described below in conjunction with the accompanying drawings. Those skilled in the art will appreciate that, with the development of technology and the emergence of new scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0052] The terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchanged in appropriate circumstances, which is only to describe the distinction mode adopted by the objects of the same attribute in the embodiments of the present application when describing. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, so that the process, method, system, product or equipment containing a series of units need not be limited to those units, but may include other units that are not clearly listed or inherent to these processes, methods, products or equipment. In addition, "at least one" refers to one or more, and "multiple" refers to two or more. "And / or", describes the association relationship of associated objects, indicating that three relationships can exist, for example, A and / or B, can represent: A exists alone, A and B exist simultaneously, and B exists alone, wherein A, B can be singular or plural. The character " / " generally represents that the associated objects before and after are a kind of "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b, or c can mean: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or plural.

[0053] First, artificial intelligence (AI) is briefly explained. The technical solution of this application applies AI technology. AI is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In short, artificial intelligence studies the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making.

[0054] Next, the overall workflow of the artificial intelligence system is described. Figure 1 , Figure 1 This is a structural diagram of the artificial intelligence main framework of the embodiment of the present application. The main framework describes the overall workflow of the artificial intelligence system and is applicable to general artificial intelligence field needs.

[0055] Figure 1The illustrated embodiment includes two dimensions: "intelligent information chain" (horizontal axis) and "IT value chain" (vertical axis). The "intelligent information chain" reflects a series of processes from data acquisition to processing. For example, it can be a general process of intelligent information perception, intelligent information representation and formation, intelligent reasoning, intelligent decision-making, intelligent execution and output. In this process, the data undergoes a condensation process of "data-information-knowledge-wisdom". The "IT value chain" is an industrial ecological process from the underlying infrastructure of human intelligence, information (providing and processing technology implementation) to the system, reflecting the value that artificial intelligence brings to the information technology industry.

[0056] (1) Infrastructure.

[0057] The infrastructure provides computing power support for the artificial intelligence system, enables communication with the outside world, and is supported by the basic platform. The infrastructure communicates with the outside world through sensors; computing power is provided by smart chips (CPU, NPU, GPU, ASIC, FPGA and other hardware acceleration chips); the basic platform includes distributed computing frameworks and networks and other related platform guarantees and support, which can include cloud storage and computing, interconnected networks, etc. For example, sensors communicate with the outside world to obtain data, and these data are provided to the smart chips in the distributed computing system provided by the basic platform for calculation.

[0058] (2)Data.

[0059] The data on the upper layer of the infrastructure is used to represent the data sources in the field of artificial intelligence. The data involves graphics, images, voice, text, and IoT data of traditional devices, including business data of existing systems and perception data such as force, displacement, liquid level, temperature, and humidity.

[0060] (3) Data processing

[0061] Data processing usually includes data training, machine learning, deep learning, search, reasoning, decision-making and other methods.

[0062] Among them, machine learning and deep learning can symbolize and formalize data for intelligent information modeling, extraction, preprocessing, and training.

[0063] Reasoning refers to the process of simulating human intelligent reasoning in computers or intelligent systems, using formalized information to perform machine thinking and solve problems based on reasoning control strategies. Typical functions are search and matching.

[0064] Decision-making refers to the process of making decisions after intelligent information is reasoned, usually providing functions such as classification, sorting, and prediction.

[0065] (4) General ability.

[0066] After the data has undergone the data processing mentioned above, some general capabilities can be further formed based on the results of the data processing, such as an algorithm or a general system, for example, translation, text analysis, computer vision processing, speech recognition, image recognition, etc.

[0067] (5) Smart products and industry applications.

[0068] Smart products and industry applications refer to the products and applications of artificial intelligence systems in various fields. They are the encapsulation of the overall artificial intelligence solution, which productizes intelligent information decision-making and realizes practical application. Its application areas mainly include: smart manufacturing, smart transportation, smart home, smart medical care, smart security, autonomous driving, safe cities, smart terminals, etc.

[0069] The following is a description of the fault prediction system provided by the embodiment of the present application. Figure 2 , Figure 2 A schematic diagram of the system architecture of a fault prediction system provided in an embodiment of the present application.

[0070] like Figure 2 As shown, the fault prediction system includes an execution module 210, a training module 220, a database 230, a target device 240, and a data storage system 250. The execution module includes a calculation module 211 and an input / output (I / O) interface 212.

[0071] The training module 220 is used to train the model 201. When the model 201 is trained, the execution module 210 runs the model 201, performs reasoning, and obtains reasoning results. In the embodiment of the present application, the model 201 is a fault prediction model, which is used to predict whether the storage medium of the device will fail, or to predict the health of the storage medium.

[0072] During the training process, the training module can obtain a training data set from the database 230 and train the model 201 based on the training data set. The specific training process will be described later and will not be expanded here.

[0073] During the reasoning process, the calculation module 211 included in the execution module 210 runs the model 201 to obtain the reasoning result. The reasoning result is sent to the first device through the I / O interface 212 to inform the user. Optionally, the reasoning result can also be stored in the data storage system 250, which is not limited here.

[0074] It should be noted that the execution module 210 and the training module 220 can be included in the same device or located in different devices, which is not limited here. The execution module 210 can predict the health of the storage medium of the device where it is located, and can also predict the health of the storage medium of other devices, such as Figure 2 The first device shown is not specifically limited here.

[0075] In addition, the training module 220 can configure the model 201 in the execution module 210 by sending it through wireless communication or wired communication. The image generation model 201 can also be configured in the execution module 210 through a removable storage device. The actual configuration method is selected according to the needs of the actual application and is not limited here.

[0076] Another thing that needs to be explained is that Figure 2 In the illustrated embodiment, the target device 240 and the execution module 210 and the training module 220 may be independent devices respectively, or they may be a whole in actual applications, which is not limited here.

[0077] It should be noted that Figure 2 It is only a schematic diagram of the architecture of the fault prediction system provided in the embodiment of the present application, and does not constitute any limitation on the positional relationship between the devices and components shown in the figure.

[0078] The following is an explanation of the fault prediction provided by the embodiment of the present application. Figure 3 , Figure 3 A schematic flow chart of a fault prediction method provided in an embodiment of the present application includes:

[0079] 301. Obtain historical fault data of a storage medium, where the historical fault data is used to indicate historical fault information of the storage medium.

[0080] The first device can obtain historical fault data from the log data, that is, the historical fault data is data in the log data related to the fault of the storage medium. In other words, the first device can filter the log data to obtain the historical fault data. Optionally, the data in the log data related to the fault of the storage medium can be marked by a fault mark, so that the first device can filter the data accordingly. The fault mark can specifically define a target field in the log data and store the fault information in the target field. By filtering the log data, data in the log data that is not related to the fault of the storage medium can be removed, reducing invalid inputs of the algorithm.

[0081] The log data may be recorded by a baseboard management controller (BMC) or collected by other plug-ins, which is not specifically limited here.

[0082] The historical fault information indicated by the historical fault data is used to describe the specific fault conditions of the storage medium, including: fault type, fault occurrence event, fault address, fault duration, fault recovery time and other parameters used to mark the fault information of the storage medium, which are not specifically limited here.

[0083] In the embodiment of the present application, there are many possibilities for the storage medium, which may be a hard disk, a solid state disk (SSD), an optical disk, a magnetic tape, a memory, and other media for storing data. The present application does not limit the type of storage medium used by the predicted device (hereinafter referred to as the target device), which can be determined according to the needs of the actual application. It should be noted that the first device and the target device can be the same device or independent devices, which is not specifically limited here.

[0084] 202. Convert the historical fault data into image feature data corresponding to the storage medium.

[0085] In summary, the first device first obtains the number of faults of each fault address corresponding to the historical fault data, and then converts the historical fault data into image feature data according to the address space of the storage medium, each fault address, and the number of faults of each fault address within the target time window.

[0086] The number of faults of each fault address corresponding to the historical fault data refers to the number of faults corresponding to each fault address in the historical fault data. Optionally, the historical fault data includes fault data within a target time window, or the historical fault data includes fault data whose number of faults reaches a threshold number.

[0087] In some optional implementations, the historical fault data includes fault data within a target time window. In this solution, the number of faults of each fault address corresponding to the historical fault data refers to the number of faults of each fault address within the target time window. The target time window refers to the collection time window of the historical fault data. For example, if the historical fault data is recorded from 8:00 to 20:00, then the target time window is 8:00 to 20:00.

[0088] It should be noted that the embodiment of the present application does not limit the duration or period of the target time window. For example, the duration of the target time window can be one day, one week, or other longer or shorter time periods. For example, the period corresponding to the target time window can also be the working time period of the target device, that is, the period when the target device is in the on state.

[0089] In an embodiment of the present application, when converting historical fault data into image feature data corresponding to the storage medium, the number of faults at each fault address corresponding to the historical fault data is combined, which can reflect the distribution of historical faults at different fault addresses, provide a more accurate basis for fault prediction, and further improve the accuracy of fault prediction.

[0090] Furthermore, the first device converts the historical fault data into image feature data according to the address space of the storage medium, each fault address, and the number of faults of each fault address within the target time window. This can be understood as performing temporal aggregation and spatial aggregation on the historical fault data to obtain the image feature data.

[0091] In summary, it includes filling the number of failures of each failure address into the position of each failure address in the address space of the storage medium to obtain the image data to be compressed. Then the image data to be compressed is compressed to obtain the image feature data. The following is a detailed description with reference to the schematic diagram. Figure 4 , Figure 4 A schematic diagram of a fault prediction method provided in an embodiment of the present application.

[0092] It is understandable that the address space of the storage medium is often expressed in the form of a (row, column) matrix, and each fault address naturally corresponds to a position in the matrix. Corresponding to the image, the address space of a storage medium has a corresponding image, and the image is divided into rows and columns to achieve a one-to-one correspondence with each address in the address space of the storage medium, so that the position of each fault address on the image can be determined.

[0093] exist Figure 4 In the embodiment shown, a 16GB memory stick is taken as an example. The large rectangles shown at t1, t2, and t3 in the figure represent a small area in the memory stick. Each small rectangle represents an address of the memory stick. Considering the three channels of RGB, there are three large rectangles at each moment. Figure 4 As shown in the figure, the address space of the storage medium is 3×65536×1024. The black part shown in the figure represents the fault address.

[0094] That is, the first device will first determine the position of each fault address in the image corresponding to the storage medium, and then fill the fault count of each fault address into the position of each fault address in the address space of the storage medium (that is, fill in the position in the image of the address space of the corresponding storage medium), to obtain the image data to be compressed. Figure 4 In the illustrated embodiment, this process is time aggregation.

[0095] It should be noted that, regardless of whether the historical fault data includes fault data within the target time window or fault data with the number of faults reaching the number threshold, the historical fault data has a corresponding duration. Then, the number of faults at each fault address in the historical fault data can be understood as the number of faults at each fault address within this period of time, that is, time aggregation of the historical fault data.

[0096] In an embodiment of the present application, the representation form of the address of the storage medium is associated with the image to determine the position corresponding to the address space of the storage medium on the image, thereby determining the position of each fault address on the image, thereby providing technical support for converting historical fault data into image feature data and improving the feasibility of the technical solution of the present application.

[0097] Furthermore, compressing the image data to be compressed can be understood as spatially aggregating the compressed image, specifically, compressing the rows and / or columns of the compressed image data, that is, compressing the rows of the compressed image, or compressing the columns of the compressed image, or compressing both the rows and columns of the compressed image.

[0098] exist Figure 4 In the illustrated embodiment, the number of rows and columns of the image data to be compressed is 65536 and 1024, respectively. After spatial aggregation, the number of rows and columns of the image feature data obtained are both 128. Figure 4 In the illustrated embodiment, the rows corresponding to the image data to be compressed are compressed 512 times, and the columns corresponding to the communication data to be compressed are compressed 8 times.

[0099] In addition, the mean value of the number of failures or the sum of the number of failures of each area to be compressed corresponding to the image data to be compressed is the number of failures of each area to be compressed at each position corresponding to the image feature data. For example, assuming that the size of the image data to be compressed is 1024×1024, and the rows and columns corresponding to the image data to be compressed are compressed 8 times, then the data in the image to be compressed includes 64 areas to be compressed, and these 64 areas to be compressed correspond to the 64 positions of the image feature data one by one.

[0100] It should also be noted that the embodiments of the present application do not limit the degree of compression of the image data to be compressed or the size of the image feature data, which can be determined according to the needs of the actual application. For example, for a device with good processing performance, the degree of compression can be lower than that of a device with poor processing performance; or the degree of compression for a scene with high prediction accuracy can be lower than that for a scene with low prediction accuracy requirements. No specific limitations are made here.

[0101] In the present application, there are multiple ways to compress the compressed image data, and there are also multiple possible degrees of compression, which can adapt to different needs and enrich the implementation methods of the technical solutions of the present application.

[0102] 303. Input the image feature data into the fault prediction model to obtain a prediction result, which is used to predict the health of the storage medium.

[0103] The fault prediction model uses an image classification algorithm to process the image feature data to obtain a prediction result. This application does not limit the specific algorithm used by the fault prediction model, including but not limited to convolutional neural networks (CNN), sequence models based on attention mechanisms (transformers), and other algorithms that can achieve image classification or image recognition, which are not specifically limited here.

[0104] The prediction results can be expressed in various forms to indicate the health of the storage medium. It can be a score for the health of the storage medium, or a prediction of the possibility of failure of the storage medium. In addition, it can also be other ways that can indicate the health, such as the health level of the storage medium, etc., which are not limited here.

[0105] In the embodiment of the present application, directly converting historical fault data into image feature data means converting the entire amount of fault information, that is, the data used to train the model is comprehensive, which improves the accuracy of the fault prediction model, and thus improves the accuracy of fault prediction. In addition, converting historical fault data into image feature data means that the fault prediction model uses an image classification algorithm, which is a mature algorithm. Applying this algorithm in the field of fault prediction also improves the effectiveness of the algorithm.

[0106] In some optional implementations, after obtaining the prediction result, the first device may also test the prediction result. In the case where the storage medium predicted by the prediction result is unhealthy, the first device tests the storage medium to obtain a test result.

[0107] Among them, the storage medium is unhealthy means that the storage medium is about to fail. The prediction result predicts that the storage medium is unhealthy in various ways:

[0108] Optionally, the score corresponding to the prediction result may be greater than the unhealthy threshold. For example, if the score corresponding to the prediction result is 80 and the unhealthy threshold is 60, then the prediction result of 80 points not only indicates that the storage medium is unhealthy, but also indicates that the storage medium is highly unhealthy and has a high probability of failure.

[0109] Optionally, the score corresponding to the prediction result may be less than the health threshold. For example, if the prediction result score is 55 and the health threshold is 60, the prediction result of 55 not only indicates that the storage medium is unhealthy, but also indicates that the degree of unhealthiness of the storage medium is low, and the probability of failure is low.

[0110] Optionally, the probability corresponding to the prediction result may be less than the health probability threshold. For example, the probability corresponding to the prediction result is 10%, and the health probability threshold is 90%. The prediction result of 10% means that the storage medium has only a 10% probability of being healthy, so the storage medium is also unhealthy.

[0111] Optionally, the probability corresponding to the prediction result may be greater than the unhealthy probability threshold. For example, the probability corresponding to the prediction result is 80%, and the healthy probability threshold is 50%. The prediction result of 90% means that the storage medium has a 90% probability of being unhealthy, and if it is greater than 50%, the storage medium is considered to be unhealthy.

[0112] In addition to the above examples, there may be other situations indicating that the storage medium is unhealthy, which are not specifically limited here.

[0113] In some optional implementations, regardless of the prediction result obtained by the fault prediction model, the first device may periodically detect the storage medium and then compare the prediction result to implement random inspection of the fault prediction model.

[0114] In some optional implementations, when the storage medium predicted by the prediction result is unhealthy, the first device may also issue an alarm to prompt the user or administrator.

[0115] In the present application, the storage medium is tested, which can not only timely discover whether the storage medium is indeed faulty, but also prevent the adverse effects caused by misjudgment of the fault prediction model, further improving the practicality of the technical solution of the present application.

[0116] In some optional implementations, the first device may also use the health status indicated by the test result as a label of the image feature data and input it into the fault prediction model, and the image feature data and the label are used to iteratively train the fault prediction model.

[0117] In this application, the test results are used as labels for image feature data to iteratively train the fault prediction model, which can not only enrich the training data set, but also further improve the accuracy of the fault prediction model because the test results are verified results and represent the actual situation.

[0118] The following is an explanation of the training method of the fault prediction model provided in the embodiment of the present application. Figure 5 , Figure 5 A flowchart of a method for training a fault prediction model provided in an embodiment of the present application includes:

[0119] 501. Obtain historical fault data of a storage medium, where the historical fault data is used to indicate historical fault information of the storage medium.

[0120] Similar to the first device, the second device can obtain historical fault data from the log data, that is, the historical fault data is data in the log data related to the fault of the storage medium. The specific implementation process of step 501 is similar to step 201, the only difference is that the execution subject is not the first device but the second device, as shown in the previous text, and will not be repeated here.

[0121] 502. Convert the historical fault data into image feature data corresponding to the storage medium.

[0122] In summary, step 502 includes obtaining the number of faults of each fault address corresponding to the historical fault data. Then, according to the address space of the storage medium, each fault address, and the number of faults of each fault address, the historical fault data is converted into image feature data. Further, it includes filling the number of faults of each fault address into the position of each fault address in the address space of the storage medium to obtain the image data to be compressed. Then, the image data to be compressed is compressed to obtain the image feature data.

[0123] Compressing the image data to be compressed includes: compressing the rows and / or columns of the image data to be compressed, wherein the mean value or sum of the number of failures of each area to be compressed corresponding to the image data to be compressed is the number of failures of each area to be compressed at each position corresponding to the image feature data.

[0124] The specific implementation process of step 502 is similar to that of step 202, the only difference being that the execution subject is not the first device but the second device, as described above, and will not be repeated here.

[0125] 503. According to the image feature data and the true value corresponding to the image feature data, iteratively train the fault prediction model until the fault prediction model converges or reaches a preset number of iterations.

[0126] When training the fault prediction model, the second device also obtains the true value corresponding to the image feature data, that is, the real health of the storage medium. This true value is also input into the fault prediction model as a training sample, and its function is to iteratively train the fault prediction model.

[0127] When the fault prediction model converges or reaches the preset number of iterations, the fault prediction model training is considered complete and can be applied in the inference stage, that is, Figure 2 The process shown. The preset number of iterations can be limited according to the needs of the actual application. For example, for a scene with higher accuracy requirements, the preset number of iterations is greater; or, for a second device with good performance, the preset number of iterations is greater than that of a second device with poor performance, etc., which is not specifically limited here.

[0128] Figure 5The beneficial effects of the embodiment shown are similar to those of the aforementioned Figure 3 to Figure 4 The embodiments shown are similar, as shown in the previous text, and will not be repeated here.

[0129] Next, the relevant equipment provided in the embodiments of the present application is described.

[0130] In summary, the relevant equipment provided in the embodiments of the present application is used to implement the fault prediction method or the training method of the fault prediction model shown above, and the device includes the corresponding units or modules of the above method. The units or modules included in the device can be implemented by software and / or hardware. The device can be, for example, a device, a module of a device (such as a chip, etc.), or a logical node, a logical module or software that can realize all or part of the functions of the device, etc., which is not specifically limited here.

[0131] See also Figure 6 , Figure 6 This is a schematic diagram of the structure of the fault prediction device provided in the embodiment of the present application. Figure 6 As shown, the communication prediction device 600 includes an acquisition unit 601 and a processing unit 602 .

[0132] In some optional implementations, the acquisition unit 601 is used to acquire historical fault data of the storage medium, and the historical fault data is used to indicate historical fault information of the storage medium. The processing unit 602 is used to convert the historical fault data into image feature data corresponding to the storage medium; input the image feature data into the fault prediction model to obtain a prediction result, and the prediction result is used to predict the health of the storage medium.

[0133] In some optional implementations, the historical fault data is fault data within a target time window. The processing unit 602 is specifically configured to obtain the number of faults of each fault address corresponding to the historical fault data within the target time window; and convert the historical fault data into image feature data according to the address space of the storage medium, each fault address, and the number of faults of each fault address within the target time window.

[0134] In some optional embodiments, the processing unit 602 is specifically used to fill the number of failures of each fault address within the target time window into the position of each fault address in the address space of the storage medium to obtain image data to be compressed; and compress the image data to be compressed to obtain image feature data.

[0135] In some optional embodiments, the processing unit 602 is specifically used to compress rows and / or columns of the image data to be compressed, wherein the mean number of failures or the sum of the number of failures for each area to be compressed corresponding to the image data to be compressed is the number of failures for each position of each area to be compressed corresponding to the image feature data.

[0136] In some optional implementations, the processing unit 602 is further configured to detect the storage medium to obtain a test result if the storage medium predicted by the prediction result is unhealthy.

[0137] In some optional implementations, the processing unit 602 is further configured to use the health status indicated by the test result as a label of the image feature data and input it into the fault prediction model, and the image feature data and the label are used to iteratively train the fault prediction model.

[0138] The fault prediction device 600 is used to implement the fault prediction method, specifically to implement the aforementioned Figure 1 The functions of the execution module in the embodiment shown, or the aforementioned Figures 2 to 4 The function of the first device in the illustrated embodiment is detailed in the foregoing text and will not be repeated here.

[0139] See also Figure 7 , Figure 7 A schematic diagram of the structure of a training device for a fault prediction model provided in an embodiment of the present application. Figure 7 As shown, the training device 700 for the fault prediction model includes an acquisition unit 701 and a processing unit 702 .

[0140] Includes the following:

[0141] In some optional implementations, the acquisition unit 701 is used to acquire historical fault data of the storage medium, and the historical fault data is used to indicate historical fault information of the storage medium. The processing unit 702 is used to convert the historical fault data into image feature data corresponding to the storage medium; and iteratively train the fault prediction model according to the image feature data and the true value corresponding to the image feature data until the fault prediction model converges or reaches a preset number of iterations.

[0142] In some optional implementations, the processing unit 702 is specifically used to obtain the number of failures of each failure address corresponding to the historical failure data; and convert the historical failure data into image feature data according to the address space of the storage medium, each failure address, and the number of failures of each failure address.

[0143] In some optional implementations, the processing unit 702 is specifically used to fill the number of failures of each fault address into the position of each fault address in the address space of the storage medium to obtain image data to be compressed; and compress the image data to be compressed to obtain image feature data.

[0144] In some optional embodiments, the processing unit 702 is specifically used to compress rows and / or columns of the image data to be compressed, wherein the mean number of failures or the sum of the number of failures for each area to be compressed corresponding to the image data to be compressed is the number of failures for each position of each area to be compressed corresponding to the image feature data.

[0145] The fault prediction model training device 700 is used to implement the fault prediction model training method, specifically to implement the aforementioned Figure 1 The function of the training module in the embodiment shown, or the aforementioned Figure 5 The function of the second device in the illustrated embodiment is as described above and will not be described in detail here.

[0146] See also Figure 8 , Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device 800 includes a processor 801 , a memory 802 , a communication interface 803 and a bus 804 .

[0147] The processor 801, the memory 802, and the communication interface 803 communicate through the bus 804, and can also communicate through other means such as wireless transmission. The memory 802 stores program code, and the processor 801 can call the program code stored in the memory 802 to execute the aforementioned Figure 1 The operations performed by the execution module in the embodiment shown, or the aforementioned Figures 2 to 4 The operation performed by the first device in the embodiment shown; or for implementing the aforementioned Figure 1 The operations performed by the training module in the embodiment shown, or the aforementioned Figure 5 The operations performed by the second device in the illustrated embodiment will not be described in detail here.

[0148] It should be understood that in the embodiment of the present application, the processor 801 may be a CPU, and the processor 801 may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0149] The memory 802 may include a read-only memory and a random access memory, and provide instructions and data to the processor 801. The memory 802 may also include a non-volatile random access memory. For example, the memory 802 may also store information about the device type.

[0150] The memory 802 may be a volatile memory or a nonvolatile memory, or may include both volatile and nonvolatile memories. Among them, the nonvolatile memory may be a read-only memory (ROM), a programmable ROM (PROM), an erasable PROM (EPROM), an electrically erasable programmable ROM (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

[0151] In addition to the data bus, the bus 804 may also include a power bus, a control bus, and a status signal bus. However, for the sake of clarity, various buses are labeled as bus 804 in the figure. Bus 840 may be a Peripheral Component Interconnect Express (PCIe) bus, or an extended industry standard architecture (EISA) bus, a unified bus (Ubus or UB), a compute express link (CXL), a cache coherent interconnect for accelerators (CCIX), etc. Bus 840 may be divided into an address bus, a data bus, a control bus, etc.

[0152] The communication device 800 may also include one or more communication interfaces, one or more operating systems, such as Windows Server 2000 and Windows Server 2003. TM , Mac OS X TM , UnixTM ,Linux TM , FreeBSD TM wait.

[0153] See also Fig. 9 , Fig. 9 It is a schematic diagram of the structure of the chip provided by the present application, which can be expressed as a neural network processor 900. The neural network processor 900 is mounted on the main CPU as a coprocessor, and the host CPU assigns tasks. The core part of the neural network processor 900 is the operation circuit 903, and the controller 904 can control the operation circuit 903 to extract the data in the weight memory 902 or the input memory 901 and perform operations.

[0154] In some implementations, the operation circuit 903 includes multiple processing units (process engines, PEs) inside. In some implementations, the operation circuit 903 can be a two-dimensional systolic array. The operation circuit 903 can also be a one-dimensional systolic array or other electronic circuits capable of performing mathematical operations such as multiplication and addition. In some implementations, the operation circuit 903 is a general-purpose matrix processor.

[0155] For example, assume there is an input matrix A, a weight matrix B, and an output matrix C. The operation circuit takes the corresponding data of matrix B from the weight memory 902 and caches it on each PE in the operation circuit. The operation circuit takes the matrix A data from the input memory 901 and performs matrix operation with matrix B, and the partial result or final result of the matrix is ​​stored in the accumulator 908.

[0156] The unified memory 906 is used to store input data and output data. The weight data is directly transferred to the weight memory 902 through the direct memory access controller (DMAC) 905. The input data is also transferred to the unified memory 906 through the DMAC.

[0157] The bus interface unit 910 (BIU) can be used to implement the interaction between the main CPU, DMAC and instruction fetch buffer 909 (IFB) through the bus. The instruction fetch buffer 909 is used to store instructions used by the controller 904.

[0158] The vector calculation unit 907 includes multiple operation processing units, and further processes the output of the operation circuit when necessary, such as vector multiplication, vector addition, exponential operation, logarithmic operation, size comparison, etc. It is mainly used for non-convolutional / fully connected layer network calculations in neural networks, such as pixel-level summation, upsampling of feature planes, etc.

[0159] in, Figures 1 to 5 The relevant operations of the fault prediction model shown in the corresponding embodiment can be performed by the operation circuit 903 or the vector calculation unit 907.

[0160] The processor mentioned in any of the above places may be a central processing unit, a microprocessor, or one or more integrated circuits for controlling the execution of the program of the above-mentioned first aspect method.

[0161] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0162] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0163] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0164] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0165] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

Claims

1. A fault prediction method, characterized in that: include: Acquire historical fault data of a storage medium, where the historical fault data is used to indicate historical fault information of the storage medium; Converting the historical fault data into image feature data corresponding to the storage medium; The image feature data is input into a fault prediction model to obtain a prediction result, and the prediction result is used to predict the health of the storage medium.

2. The method according to claim 1, characterized in that: The converting the historical fault data into image feature data corresponding to the storage medium includes: Obtaining the number of faults of each fault address corresponding to the historical fault data; The historical fault data is converted into the image feature data according to the address space of the storage medium, each fault address, and the number of faults of each fault address.

3. The method according to claim 2, characterized in that The converting the historical fault data into the image feature data according to the address space of the storage medium, each fault address, and the number of faults of each fault address includes: Fill the position of each fault address in the address space of the storage medium with the fault number of each fault address to obtain the image data to be compressed; The image data to be compressed is compressed to obtain the image feature data.

4. The method according to claim 3, characterized in that The compressing the image data to be compressed comprises: The rows and / or columns of the image data to be compressed are compressed, wherein the mean value or the sum of the number of failures of each area to be compressed corresponding to the image data to be compressed is the number of failures of each area to be compressed at each position corresponding to the image feature data.

5. The method according to any one of claims 1 to 4, characterized in that After inputting the image feature data into the fault prediction model to obtain the prediction result, the method further includes: If the storage medium predicted by the prediction result is unhealthy, the storage medium is tested to obtain a test result.

6. The method according to claim 5, characterized in that The method further comprises: The health status indicated by the test result is used as a label of the image feature data and input into the fault prediction model. The image feature data and the label are used to iteratively train the fault prediction model.

7. A fault prediction device, characterized in that: include: An acquisition unit, used for acquiring historical fault data of a storage medium, wherein the historical fault data is used for indicating historical fault information of the storage medium; A processing unit, used for converting the historical fault data into image feature data corresponding to the storage medium; The processing unit is further used to input the image feature data into a fault prediction model to obtain a prediction result, and the prediction result is used to predict the health of the storage medium.

8. The device according to claim 7, characterized in that The processing unit is specifically used for: Obtaining the number of faults of each fault address corresponding to the historical fault data; The historical fault data is converted into the image feature data according to the address space of the storage medium, each fault address, and the number of faults of each fault address.

9. The device according to claim 8, characterized in that The processing unit is specifically used for: Fill the number of failures of each failure address into the position of each failure address in the address space of the storage medium to obtain image data to be compressed; The image data to be compressed is compressed to obtain the image feature data.

10. The device according to claim 9, characterized in that The processing unit is specifically used for: The rows and / or columns of the image data to be compressed are compressed, wherein the mean value or the sum of the number of failures of each area to be compressed corresponding to the image data to be compressed is the number of failures of each area to be compressed at each position corresponding to the image feature data.

11. The device according to any one of claims 7 to 10, characterized in that The processing unit is further used for: If the storage medium predicted by the prediction result is unhealthy, the storage medium is tested to obtain a test result.

12. The device according to claim 11, characterized in that The processing unit is further used for: The health status indicated by the test result is used as a label of the image feature data and input into the fault prediction model. The image feature data and the label are used to iteratively train the fault prediction model.

13. An electronic device, characterized in that: comprising a processor coupled to a memory; Instructions are stored in the memory, and when the instructions are executed on the processor, the electronic device implements the method according to any one of claims 1 to 6.

14. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores instructions, and when the instructions are executed on a processor, the method according to any one of claims 1 to 6 is implemented.

15. A computer program product, characterized in that When the computer program product is executed on a computer, the method according to any one of claims 1 to 6 is implemented.

Images