Disk fault prediction method and device

Abstract

The invention discloses a disk fault prediction method and device, and relates to the technical field of data processing, and the method comprises the steps: obtaining the target feature data of a target disk, and predicting the fault probability of the target disk in a future target time period through a preset disk fault prediction model according to the target feature data of the target disk, outputting the target prediction probability, predicting the fault probability of the target disk in the future target time period, and determining the feature contribution degree of each feature data based on a preset disk fault prediction model and the target feature data when the target prediction probability is determined to be greater than a first preset threshold value; furthermore, a preset fault analysis model is adopted to determine a target fault reason according to the target feature data and the feature contribution degree of each feature data, and the target fault reason is sent to the client, so that a user can grasp the running state of the target disk and the fault reason of which the fault probability exceeds a threshold value in advance, and the interpretability of the disk fault is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and in particular to a disk failure prediction method and device. BACKGROUND

[0002] With the rapid development of new generation information technologies such as Internet of Things and cloud computing, the global data volume is growing explosively. The disk is responsible for the persistent storage of massive data, and its running stability is directly related to the reliability and data security of the entire computer system. Once the disk fails, it may cause data read / write delay and service interruption, or even critical data loss and system crash.

[0003] At present, for the failure identification of the disk, the disk attribute parameters are generally monitored by a static threshold, and the failure alarm is triggered only when the parameter anomaly exceeds the threshold, which leads to insufficient explainability of the disk failure. SUMMARY

[0004] The present application provides a disk failure prediction method and device to at least solve the problem of insufficient explainability of the disk failure caused by monitoring the disk attribute parameters by a static threshold and triggering the failure alarm only when the parameter anomaly exceeds the threshold in the related art.

[0005] In a first aspect, the present application provides a disk failure prediction method, which comprises:

[0006] obtaining target feature data of a target disk; the target feature data comprises a plurality of feature data; the target feature data is used to reflect the current state of the target disk; and the target feature data comprises the disk type of the target disk;

[0007] predicting the target disk based on the target feature data of the target disk by using a preset disk failure prediction model, and outputting a target prediction probability; the target prediction probability is the probability of failure of the target disk in a future target time period;

[0008] if it is determined that the target prediction probability is greater than a first preset threshold, determining the feature contribution degree of each feature data based on the preset disk failure prediction model and the target feature data;

[0009] determining a target failure cause based on the target feature data and the feature contribution degree of each feature data by using a preset failure analysis model, and sending the target failure cause to a client.

[0010] In a second aspect, the present application further provides a disk failure prediction device, which comprises:

[0011] The acquisition module is configured to acquire target feature data of the target disk, wherein the target feature data comprises a plurality of feature data, the target feature data is used to reflect a current state of the target disk, and the target feature data comprises a disk type of the target disk.

[0012] The prediction module is configured to use a preset disk failure prediction model and predict the target disk based on the target feature data of the target disk, and output a target prediction probability, wherein the target prediction probability is a probability of the target disk failing in a future target time period.

[0013] The determination module is configured to determine a feature contribution degree of each feature data based on the preset disk failure prediction model and the target feature data if the target prediction probability is greater than a first preset threshold.

[0014] The determination module is further configured to use a preset failure analysis model and determine a target failure cause based on the target feature data and the feature contribution degree of each feature data, and send the target failure cause to the client.

[0015] In a third aspect, the present application further provides an electronic device, comprising: a memory configured to store a computer program; and a processor configured to execute the computer program to implement the steps of the disk failure prediction method provided in the first aspect.

[0016] In a fourth aspect, the present application further provides a computer readable storage medium, wherein the computer readable storage medium stores a computer program, and the computer program is executed by a processor to implement the steps of the disk failure prediction method provided in the first aspect.

[0017] In a fifth aspect, the present application further provides a computer program product, comprising a computer program, and the computer program is executed by a processor to implement the steps of the disk failure prediction method provided in the first aspect.

[0018] By the disk failure prediction method and device provided in the application, the target feature data of the target disk is automatically acquired, the target feature data of the target disk is input into a preset disk failure prediction model for prediction, so as to output a target prediction probability, the probability of the target disk failing in a future target time period is predicted, and when it is determined that the target prediction probability is greater than a first preset threshold, the feature contribution degrees of each feature data are determined based on the preset disk failure prediction model and the target feature data, the target failure cause is further determined by using a preset failure analysis model and according to the target feature data and the feature contribution degrees of each feature data, and the target failure cause is sent to the client, the failure cause of the disk failure is predicted, compared with determining whether the attribute parameters of the disk exceed a threshold to determine the health state of the disk, the probability of the target disk failing in a future target time period is predicted according to the target feature data of the target disk, and the target failure cause is predicted, so that the user can master the running state of the target disk and the failure cause when the failure probability exceeds the threshold in advance, thereby improving the explainability of the disk failure and helping to improve the user experience. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the embodiments of the present application, the drawings needed in the embodiments will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0020] Figure 1 The application scenario diagram of the disk failure prediction method provided in an embodiment of the present application is shown in the following figure.

[0021] Figure 2 The flowchart of the disk failure prediction method provided in an embodiment of the present application is shown in the following figure.

[0022] Figure 3 The flowchart of the disk failure prediction method provided in another embodiment of the present application is shown in the following figure.

[0023] Figure 4 The structural diagram of the disk failure prediction device provided in an embodiment of the present application is shown in the following figure.

[0024] Figure 5 The structural diagram of the electronic device provided in an embodiment of the present application is shown in the following figure. DETAILED DESCRIPTION

[0025] With reference to the drawings of the embodiments of the present application, the technical solutions in the embodiments of the present application will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all embodiments of the present application. Based on the embodiments of the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0026] It should be noted that, in the description of the present application, the terms “comprise”, “contain” or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or equipment comprising a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or equipment. The terms “first”, “second” and the like in the present application are used to distinguish similar objects, and are not used to describe a specific order or sequence.

[0027] In order for those skilled in the art to better understand the present application, the present application will be further described in detail below with reference to the drawings and specific embodiments.

[0028] In the current era of rapid development of artificial intelligence technology, as a key medium for data storage, the disk bears the task of persistent storage of massive data, and its running stability and reliability are directly related to the normal operation and data security of the entire computer system. Facing the increasing demand for data processing, the disk not only needs to provide efficient data read-write service, but also must ensure the stability of long-time continuous work. Once the disk fails, it may cause data read-write delay and service interruption, affecting the continuity of business processes, or even cause critical data loss and system crash. Therefore, timely and accurate identification of disk health status and early warning of potential failure risks are crucial for ensuring data security and maintaining system stability. At present, the identification of disk failure mainly relies on setting fixed thresholds for disk attribute parameters (such as disk self-monitoring, analysis and reporting technical information), and evaluating the health status of the disk by monitoring whether these parameters exceed the preset thresholds. Since the static threshold is limited, it cannot dynamically adapt to changes under different working environments and load conditions, and can only identify the current disk failure, but cannot determine the cause of the disk failure, resulting in poor explainability of disk failure.

[0029] Therefore, in the face of the above technical problems, instead of only monitoring whether the attribute parameter information of the disk exceeds the preset threshold to identify whether the disk has failed, the target feature data of the target disk is obtained, the probability of the target disk failing in a future target time period is predicted according to the target feature data of the target disk by using a preset disk failure prediction model, and a target prediction probability is output, so as to predict the probability of the target disk failing in the future target time period. When it is determined that the target prediction probability is greater than a first preset threshold, the feature contribution degrees of the feature data are determined based on the preset disk failure prediction model and the target feature data, a target failure cause is further determined by using a preset failure analysis model and according to the target feature data and the feature contribution degrees of the feature data, and the target failure cause is sent to the client, so that the user can master the running state of the target disk in advance, and the failure cause when the failure probability exceeds the threshold, thereby enhancing the explainability of disk failure, enabling the user to understand the potential risks and the root causes of the failure in advance, and improving the user experience.

[0030] Figure 1 An application scenario diagram of the disk failure prediction method provided by an embodiment of the present application is shown in FIG. 1. Figure 1 As shown in FIG. 1, the application scenario provided by the embodiment includes a server device 10 and a client 20. The disk failure prediction method is applied to the server device 10. The server device 10 obtains target feature data of a target disk from a preset database, and uses a preset disk failure prediction model to predict the target disk according to the target feature data of the target disk, and outputs a target prediction probability. The target feature data includes a plurality of feature data, and the target feature data is used to reflect the current state of the target disk. The target feature data includes the disk type of the target disk. Further, if the server device 10 determines that the target prediction probability is greater than a first preset threshold, the feature contribution degrees of the feature data are determined according to the preset disk failure prediction model and the target feature data, a target failure cause is determined according to the target feature data and the feature contribution degrees of the feature data by using a preset failure analysis model, and the target failure cause is sent to the client 20. The client 20 displays the received target prediction probability and target failure cause.

[0031] Figure 2 A flowchart of the disk failure prediction method provided by an embodiment of the present application is shown in FIG. 2. Figure 2 The disk failure prediction method provided by the embodiment is applied to a server device. The disk failure prediction method provided by the embodiment specifically includes the following steps:

[0032] S201: Obtain target feature data of a target disk.

[0033] The target feature data includes a plurality of feature data. The target feature data is used to reflect the current state of the target disk. The target feature data includes a disk type of the target disk.

[0034] The disk type includes a mechanical disk and a solid state disk.

[0035] The target disk is a disk to be subjected to fault prediction.

[0036] Exemplarily, the target feature data of the target disk includes real-time parameter data of the 5th, 187th, 188th, 197th and 198th items in the Self-Monitoring, Analysis, and Reporting Technology (SMART) corresponding to the target disk.

[0037] The target feature data of the target disk further includes real-time temperature, current vibration peak value, current disk read-write operation number per second, error number related to the disk in the system log in the last one hour, growth rate of the number of remapping sectors in the past 7 days, current number of sectors to be repaired, number of sectors to be repaired per hour in the past 1 day, disk temperature in the last 1 hour, vibration peak value mean change of the disk tray acceleration sensor in the past 3 days, change trend of the read-write number per second in the past 12 hours, and error log count growth rate in the last 12 hours.

[0038] The target feature data of the target disk further includes peak difference of the disk temperature in the past 24 hours, vibration peak value in the last 1 hour, standard deviation of the read-write number per second in the last 6 hours, and standard deviation of the fan speed in the last 8 hours. The target feature data of the target disk further includes corresponding disk working temperature in a preset time range, instantaneous vibration acceleration of the disk, cooling fan speed, read-write number per second of the target disk, temperature mean value when the read-write number per second exceeds a preset read-write threshold in the past 1 hour, and response time mean value when the temperature exceeds a preset temperature threshold in the past 1 hour.

[0039] Optionally, the preset read-write threshold and the preset temperature threshold can be autonomously set according to requirements, which are not limited in the embodiment.

[0040] The 5th item is the number of remapping sectors, the 187th item is the number of unrepairable errors, the 188th item is the number of command timeouts, the 197th item is the current number of sectors to be repaired, and the 198th item is the number of offline detection unrepairable sectors.

[0041] Among them, the number of remapped sectors records the number of bad sectors that have been remapped. When the disk finds that a sector is damaged, it will be marked as a bad sector, and the data will be transferred to the reserved spare sector, and this process is sector remapping. The number of uncorrectable errors records the number of read and write errors that cannot be corrected by retry and other methods. The number of command timeouts indicates the number of times the disk has timed out when executing commands. The number of current sectors to be repaired records the number of sectors that have been detected to have problems but have not been remapped. The number of offline detection uncorrectable sectors refers to the number of sectors that cannot be corrected when the disk is offline.

[0042] Specifically, in the embodiment, the server device obtains the target feature data of the target disk from the preset database.

[0043] Optionally, the preset time range can be 1 hour, 1 day, 3 days, etc., which is not limited in the embodiment.

[0044] S202: A preset disk failure prediction model is used to predict the target disk based on the target feature data of the target disk, and a target prediction probability is output.

[0045] Among them, the target prediction probability is the probability of the target disk failing in the future target time period.

[0046] It can be understood that, assuming that the target time period can be 72 hours, the target prediction probability refers to the probability of the target disk failing in the future 72 hours from the time when the current latest target feature data of the target disk is obtained.

[0047] Among them, the target time period corresponds to the time period that the preset disk failure prediction model can predict.

[0048] Specifically, in the embodiment, the server device inputs the target feature data of the target disk into the preset disk failure prediction model, so that the preset disk failure prediction model predicts the target disk according to the target feature data of the target disk, and outputs the target prediction probability.

[0049] S203: If it is determined that the target prediction probability is greater than a first preset threshold, the feature contribution degree of each feature data is determined based on the preset disk failure prediction model and the target feature data.

[0050] Optionally, the first preset threshold can be 60%, or it can be set independently according to requirements, which is not limited in the embodiment.

[0051] Among them, the feature contribution degree refers to the influence degree of each feature data on the disk failure probability.

[0052] Specifically, in this embodiment, the preset contribution degree algorithm is used to calculate the feature contribution degree of each feature data according to the preset disk failure prediction model and the target feature data of the target disk, so as to obtain the feature contribution degree of each feature data.

[0053] Optionally, the preset contribution degree algorithm can be a SHapley Additive exPlanations (SHAP) algorithm. The SHAP algorithm is based on the Shapley value in game theory, and decomposes the prediction result of the machine learning model into the sum of the contribution values of each feature.

[0054] S204: The target failure cause is determined based on the target feature data and the feature contribution degrees of each feature data by using the preset failure analysis model, and is sent to the client.

[0055] Specifically, in this embodiment, the server device inputs the feature contribution degrees of each feature data and the target feature data into the preset failure analysis model, and uses the preset failure analysis model to predict the failure cause of the target disk, and outputs the target failure cause.

[0056] The target failure cause refers to the cause of generating a target prediction probability exceeding a first preset threshold.

[0057] Optionally, the preset failure analysis model can be a Bayesian network, or can be set autonomously according to requirements, which is not limited in this embodiment.

[0058] Specifically, by automatically obtaining the target feature data of the target disk, the target feature data of the target disk is input into the preset disk failure prediction model for prediction, so as to output a target prediction probability, realize prediction of the failure probability of the target disk in a future target time period, and when it is determined that the target prediction probability is greater than a first preset threshold, the feature contribution degrees of each feature data are determined based on the preset disk failure prediction model and the target feature data, and then the target failure cause is determined based on the target feature data and the feature contribution degrees of each feature data by using the preset failure analysis model, and is sent to the client, realizing prediction of the failure cause of the disk failure. Compared with determining whether the attribute parameters of the disk exceed the threshold to determine the health status of the disk, the failure probability of the target disk in the future target time period is predicted according to the target feature data of the target disk, and the target failure cause is predicted, so that the user can master the running state of the target disk in advance, and the failure cause when the failure probability exceeds the threshold, thereby enhancing the explainability of the disk failure, and helping to improve the user experience.

[0059] As an optional implementation, on the basis of the above embodiment, the target feature data of the target disk is obtained, including:

[0060] obtaining original data in a first preset time period; the original data including parameter data of the target disk, target environment data and target load data;

[0061] scaling the original data in the first preset time period by using a preset scaling algorithm to obtain scaled original data;

[0062] performing feature extraction on the scaled original data to obtain target feature data of the target disk.

[0063] The parameter data of the target disk refers to the 5th, 187th, 188th, 197th and 198th items in the SMART information corresponding to the target disk, each of which is an attribute parameter. The target environment data includes the working temperature of the target disk, the vibration acceleration of the target disk and the rotation speed of the cooling fan in the system. The target load data includes the read-write frequency per second of the target disk, the response time of the target disk, the read-write throughput and the error log count of the target disk. The error log count refers to the disk errors recorded by the operating system. The read-write throughput refers to the data transmission amount per second.

[0064] It can be understood that the SMART information corresponding to the disk contains both original values and standardized values. The standardized values are standardized values obtained by calculating the original values by the algorithm built-in by the disk manufacturer. In the embodiment, the standard value refers to the standardized value corresponding to each attribute parameter in the SMART information.

[0065] The current standard value corresponding to each parameter of the target disk refers to the current standard value of each parameter of the target disk at the current time.

[0066] For example, assuming that the current time is February 1st and the first preset time period is 29 days, the current standard value corresponding to each attribute parameter of the target disk refers to the standardized value corresponding to each attribute parameter on February 1st and the standardized value of each attribute parameter of the target disk in the 29 consecutive days before February 1st.

[0067] It can be understood that if 30 days of target feature data are required, the first preset time period is 29 days. The first preset time period is determined by the number of days of the required target feature data.

[0068] Specifically, in the embodiment, the server device obtains the current standard value corresponding to each parameter of the target disk by using a preset interface or a preset command, and obtains the historical standard value of each parameter of the target disk in the first preset time period from a preset database, and then performs scaling processing on the current standard value corresponding to each parameter and the historical standard value of the parameter in the first preset time period by using the corresponding scaling formula in the preset scaling algorithm, thereby obtaining the target value of each parameter after scaling, and taking the target value of each parameter after scaling as the parameter data of the target disk after scaling.

[0069] Optionally, the preset interface or the preset command is a preset setting.

[0070] In the preset scaling algorithm, the corresponding scaling algorithm refers to subtracting the minimum value in the parameter corresponding standard value from the parameter corresponding standard value to obtain a first difference value, and subtracting the minimum value in the parameter corresponding standard value from the maximum value in the parameter corresponding standard value to obtain a second difference value. Then, the ratio of the first difference value to the second difference value is doubled, and the doubled value is subtracted from 1 to obtain the target value corresponding to the parameter.

[0071] It can be understood that only the data in the original data that needs to be scaled is scaled. The data that does not need to be scaled is retained as the original data.

[0072] Further, in the embodiment, the parameter data of the scaled target disk is obtained, and corresponding calculation is performed according to the parameter data of the scaled target disk and the target environment data and the target load data, so as to obtain the target characteristic data of the target disk.

[0073] It can be understood that according to the target characteristic data described above, the corresponding data in the target characteristic data can be obtained by calculating the corresponding data in the original data.

[0074] Exemplarily, the feature extraction refers to performing corresponding calculation according to the original data to obtain corresponding characteristic data. The current vibration peak value refers to the peak value of the real-time vibration of the disk carrier acceleration sensor in the last 1 minute. The current disk read-write operation number per second refers to the average value of the disk read-write operation number per second in the last 1 minute. The growth rate of the remapping sector number in the past 7 days is obtained by subtracting the value of the remapping sector number 7 days ago from the current remapping sector number, and dividing the difference by 7. The recent 1-hour high-speed rate of the disk temperature is obtained by subtracting the temperature value 24 hours ago from the current temperature value, and dividing the difference by 24. The vibration peak value average change of the disk carrier acceleration sensor in the past 3 days is obtained by subtracting the average value of the last 3 days from the average value of the last 1 day, and dividing the difference by 2. The change trend of the read-write frequency per second in the past 12 hours is obtained by subtracting the average value of the read-write frequency per second in the last 12 hours from the average value of the read-write frequency per second in the last 6 hours, and dividing the difference by 6. The error log count growth rate in the last 12 hours is obtained by subtracting the total number of error logs in the last 12 hours from the total number of error logs in the last 6 hours, and dividing the difference by 6.

[0075] Specifically, by scaling the original data in the target disk, the original data is unified to the same dimension, thereby ensuring the uniformity of the data.

[0076] As an optional implementation, on the basis of any one of the above embodiments, a preset disk failure prediction model is adopted to predict the target disk based on the target feature data of the target disk, and a target prediction probability is output, comprising:

[0077] The target feature data of the target disk is input into the preset disk failure prediction model, and the target disk is predicted by using the preset disk failure prediction model, and a target prediction probability is output.

[0078] Optionally, the preset disk failure prediction model is a convolutional neural network, which can also be a convolutional neural network combined with a long short-term memory network model, etc., which is not limited in the embodiment.

[0079] Specifically, in the embodiment, the server device inputs the target feature data of the target disk into the preset disk failure prediction model, so that the preset disk failure prediction model predicts the target disk, thereby obtaining the target prediction probability output by the preset disk failure prediction model.

[0080] Specifically, by inputting the target feature data of the target disk into the preset disk failure prediction model, the probability of the target disk failing in the future target time period is estimated, and then the target disk can be processed according to the target prediction probability, thereby ensuring the stability of the system.

[0081] As an optional implementation, on the basis of any one of the above embodiments, if the target prediction probability is greater than a first preset threshold, the feature contribution degree of each feature data is determined based on the preset disk failure prediction model and the target feature data, comprising:

[0082] The preset disk failure prediction model is input into a preset feature contribution degree interpreter to initialize the preset feature contribution degree interpreter to obtain an initialized preset feature contribution degree interpreter;

[0083] The target feature data is input into the preset feature contribution degree interpreter, and the contribution degree of the feature data in the target feature data is calculated by using the preset feature contribution degree interpreter, and the feature contribution degree corresponding to each feature data is output.

[0084] The preset feature contribution degree interpreter can be a Shapley additive explanations interpreter.

[0085] Specifically, in the embodiment, the preset disk failure prediction model is input into the preset feature contribution degree interpreter to initialize the preset feature contribution degree interpreter, so that the preset feature contribution degree interpreter learns the internal parameters and decision logic of the preset disk failure prediction model. After the initialization of the preset feature contribution degree interpreter is completed, the target feature data is input into the preset feature contribution degree interpreter, so that the preset feature contribution degree interpreter calculates the contribution degree of each feature data in the target feature data, and outputs the feature contribution degree corresponding to each feature data.

[0086] It can be understood that the preset disk failure prediction model is a trained disk failure prediction model.

[0087] In the calculation of the feature contribution value using the Shapley Additive explanations interpreter, the preset disk failure prediction model needs to be input only once (when initializing the Shapley Additive explanations interpreter), and the model does not need to be repeatedly input each time the feature contribution value of the target disk is calculated. The reason for not needing to repeatedly input is that the parameters of the preset disk failure prediction model are fixed, and the Shapley Additive explanations interpreter has "remembered" the decision logic of the preset disk failure prediction model after 1-time initialization, and only needs to input a new feature vector to calculate the contribution value.

[0088] In the embodiment, the internal parameters of the preset disk failure prediction model are weights, convolution kernels, tree structures, and the like.

[0089] Specifically, by quantifying the contribution degrees of the feature data in the target feature data, the key factors and the secondary factors that cause the target prediction probability to exceed the first preset threshold can be distinguished, which is helpful for subsequent analysis and processing of the target failure cause.

[0090] As an optional implementation, on the basis of any one of the above embodiments, the target failure cause is determined by using the preset failure analysis model and based on the target feature data and the feature contribution degrees of the feature data, and the method comprises the following steps.

[0091] The feature contribution degrees of the feature data are sorted in descending order to obtain the sorted feature contribution degrees of the feature data;

[0092] The feature contribution degrees of a preset number of feature data are sequentially obtained from the front to the back in the sorted feature contribution degrees of the feature data;

[0093] The feature contribution degrees of the preset number of feature data and the target feature data are input into the preset failure analysis model, and the preset failure analysis model is used to predict the failure cause of the target disk, and the target failure cause is output.

[0094] Optionally, the preset number can be 3 or other positive integers, which is not limited in the embodiment.

[0095] Specifically, in the embodiment, the server device sorts the feature contribution degrees of the feature data in descending order to obtain the sorted feature contribution degrees of the feature data, then sequentially obtains the feature contribution degrees of a preset number of feature data from the front to the back in the sorted feature contribution degrees of the feature data, and inputs the feature contribution degrees of the preset number of feature data and the target feature data into the preset failure analysis model, uses the preset failure analysis model to predict the failure cause of the target disk, and outputs the target failure cause.

[0096] The preset fault analysis model is a trained Bayesian network.

[0097] For example, the training process of the preset fault analysis model is as follows: a sample set, a preset structure constraint strategy, and an initial fault analysis model are obtained. The initial fault analysis model is an untrained model. The sample set is divided into a training sample set and a verification sample set. The training sample set and the preset structure constraint strategy are input into the initial fault analysis model for structure learning, so that the initial fault analysis model generates a target network structure, and a structure-learned initial fault analysis model is obtained. The target network structure and the training sample set are input into the structure-learned initial fault analysis model for parameter learning, so that the initial fault analysis model generates a target probability table, and a parameter-learned initial fault analysis model is obtained. The parameter-learned initial fault analysis model is verified by using the verification set. The verification sample set is input into the model of the parameter-learned initial fault analysis model for cause reasoning. The probability of each possible fault cause is calculated, and the cause with the highest probability is output. The fault cause recognition accuracy is calculated according to the output result. When the fault cause recognition accuracy is greater than a preset accuracy threshold, the training is completed. The parameter-learned initial fault analysis model is determined as the preset fault analysis model.

[0098] Optionally, the preset structure constraint strategy is a predefined rule or restriction condition for guiding the construction of a network topology. The preset structure constraint strategy can be autonomously set according to requirements, and is not limited in the embodiment.

[0099] Optionally, the preset accuracy threshold can be autonomously set according to requirements, and is not limited in the embodiment.

[0100] The target of the structure learning is to find the causal relationship between nodes.

[0101] The sample set includes fault samples and normal samples. The fault samples include fault data and fault causes of a disk. Specifically, the features with the greatest impact on faults can be focused on by contribution degree sorting and preset number screening. The preset fault analysis model can focus on the core contradictions of the target fault causes, and the target fault causes can be determined by the preset fault analysis model. Therefore, the operation and maintenance personnel can timely solve the faults, and the operation and maintenance efficiency can be improved.

[0102] As an optional implementation, on the basis of any one of the above embodiments, the method further includes:

[0103] determining whether the target prediction probability is greater than a first fault threshold and less than a second fault threshold;

[0104] If the target prediction probability is greater than the first failure threshold and less than the second failure threshold, a first prompt information is generated; the first prompt information is used to prompt to backup data in the target disk;

[0105] If the target prediction probability is greater than or equal to the second failure threshold, a second prompt information is generated; the second prompt information is used to prompt to replace the target disk;

[0106] If the target prediction probability is less than the first failure threshold, it is determined whether the target prediction probability is less than a third failure threshold;

[0107] If the target prediction probability is less than the third failure threshold, a third prompt information is generated, and the third prompt information is used to prompt that the target disk is currently in a healthy state.

[0108] Optionally, the first failure threshold can be 50% or the like, the second failure threshold can be 80% or the like, and the third failure threshold can be 10% or the like, which is not limited in the embodiment.

[0109] The second failure threshold is greater than the first failure threshold, and the first failure threshold is greater than the third failure threshold.

[0110] Exemplarily, in the embodiment, it is assumed that the first failure threshold is 50%, the second failure threshold is 80%, and the third failure threshold is 10%. After the target prediction probability is acquired, the service end device determines whether the target prediction probability is greater than the first failure threshold and less than the second failure threshold. If the target prediction probability is greater than 50% and less than 80%, the first prompt information is generated, and the first prompt information can be: “The disk has a risk of failure, and it is suggested to backup all data as soon as possible and observe the running state of the disk.” If the target prediction probability is greater than or equal to 80%, the second prompt information is generated, and the second prompt information can be: “The disk has a serious risk of failure and may fail at any time. Please backup data and replace the disk immediately.” If the target prediction probability is less than 50%, it is determined whether the target prediction probability is also less than 10%. If the target prediction probability is less than 10%, the third prompt information is generated, and the third prompt information can be: “The disk is in good condition, and it is suggested to backup data regularly.”

[0111] It can be understood that the first prompt information, the second prompt information and the third prompt information can be other text descriptions, which realize the corresponding prompt purposes.

[0112] The first preset threshold and the first failure threshold can be the same value or different values, which is not limited in the embodiment.

[0113] Specifically, the server device divides the risk interval by adopting a double-layer threshold value, triggers a data backup prompt when the target prediction probability is in the middle range, and reserves data migration time for potential failure, directly triggers a disk replacement prompt when the target prediction probability is greater than the second failure threshold value, and thus realizes zero-delay response of high-risk failure, which can greatly reduce the risk of business interruption caused by sudden disk failure. Further, by introducing a third failure threshold value, the low-risk state is confirmed again, and when the probability is lower than the third failure threshold value, the healthy state is explicitly prompted, reducing the user's invalid attention to the normal disk.

[0114] As an optional implementation, on the basis of any one of the above embodiments, the method further comprises:

[0115] inputting the target feature data of the target disk into a preset disk failure prediction model, and performing health scoring on the target disk by using the preset disk failure prediction model, and outputting a target disk health score; the target disk health score is used to identify the health state of the target disk;

[0116] If it is determined that the target disk health score is greater than the second preset threshold value, the feature contribution degree of each feature data is determined based on the preset disk failure prediction model and the target feature data;

[0117] determining the target failure cause by using the preset failure analysis model and based on the target feature data and the feature contribution degree of each feature data, and sending it to the client.

[0118] Specifically, in this embodiment, the server device inputs the target feature data of the target disk into the preset disk failure prediction model, so that the preset disk failure prediction model performs health scoring on the target disk, thereby obtaining the target disk health score output by the preset disk failure prediction model. The target disk health score is compared with the second preset threshold value, and if it is determined that the target disk health score is greater than the second preset threshold value, the preset disk failure prediction model and the target feature data are input into the preset feature contribution degree interpreter, thereby determining the feature contribution degree of each feature data. The target failure cause is determined by using the preset failure analysis model according to the target feature data and the feature contribution degree of each feature data, and then sent to the client.

[0119] Optionally, the second preset threshold value can be autonomously set according to requirements, which is not limited in this embodiment.

[0120] Specifically, the health degree of the target disk is scored by the preset disk failure prediction model, and the prediction of the target failure cause is triggered when the target disk health score exceeds the threshold value, thereby realizing the contribution degree calculation only for the high-risk disk (the score is close to the threshold value), and thus reducing the consumption of computing resources.

[0121] As an optional implementation, based on any one of the above embodiments, before predicting the target disk based on the preset disk failure prediction model and the target feature data of the target disk, the method further comprises:

[0122] obtaining a target sample set and an initial disk failure prediction model;

[0123] dividing the target sample set into a target training set, a target validation set and a target test set;

[0124] training the initial disk failure prediction model based on the target training set using a back propagation algorithm to obtain the initial disk failure prediction model after each round of training;

[0125] determining a target disk failure prediction model based on the target validation set and the initial disk failure prediction model after each round of training;

[0126] determining a preset disk failure prediction model based on the target test set and the target disk failure prediction model.

[0127] The target sample set includes normal disk sample data and failed disk sample data, and the normal disk sample data and the failed disk sample data are labeled with different labels and the health scores of the disk sample data are labeled with labels. The initial disk failure prediction model is an untrained disk failure prediction model.

[0128] Optionally, the division ratio of the target training set, the target validation set and the target test set can be 6:1:3, or other ratios, which are not limited in the embodiment.

[0129] The proportion of the target training set is greater than the proportions of the target validation set and the target test set.

[0130] The target training set is used to train the initial disk failure prediction model, the target validation set is used to monitor the overfitting of the initial disk failure prediction model in each round of training and select the target disk failure prediction model, and the target test set is used to perform final performance evaluation on the target disk failure prediction model.

[0131] The target disk failure prediction model is a model that can be selected for final performance evaluation.

[0132] Specifically, in the embodiment, the target sample set and the initial disk failure prediction model are obtained from a preset database, and the target sample set is divided into a target training set, a target validation set and a target test set according to a preset division ratio. In each round of training, the samples in the target training set are input into the initial disk failure prediction model to obtain the prediction output of the initial disk failure prediction. The loss value between the prediction output and the actual failure label is calculated by using a preset loss function, and the gradient of the loss function to the parameters (weights and biases) of each layer is calculated layer by layer from the output layer by using a back propagation algorithm, so that the parameters of the initial disk failure prediction model are updated according to a preset gradient descent algorithm to reduce the loss value. After each round of training, the model parameters obtained by the current training are saved to form the initial disk failure prediction model after each round of training. For the initial disk failure prediction model after each round of training, the samples in the target validation set are input into the initial disk failure prediction model after each round of training to obtain the validation prediction result, and then the first recall rate of the initial disk failure prediction model after each round of training on the target validation set and the validation loss value are calculated according to the validation prediction result. When the validation loss is less than a preset loss threshold, the initial disk failure prediction model trained in the corresponding round is determined as a candidate disk failure prediction model. The number of candidate disk failure prediction models is at least two. It is determined whether the first recall rate corresponding to each candidate disk failure prediction model is greater than a preset recall rate threshold. If the first recall rate corresponding to each candidate disk failure prediction model is greater than the preset recall rate threshold, the candidate disk failure prediction model with the minimum validation loss among the candidate disk failure prediction models is determined as a target disk failure prediction model. Further, the samples in the target test set are input into the target disk failure prediction model to obtain a test prediction result, and then the second recall rate of the target disk failure prediction model on the target test set and the test loss value are calculated according to the test prediction result. When the test loss value is less than a preset test threshold and the second recall rate is greater than the preset recall rate threshold, the target disk failure prediction model is determined as a preset disk failure prediction model. If any one of the test loss value being less than the preset test threshold and the second recall rate being greater than the preset recall rate threshold is not satisfied, the initial disk failure prediction model is retrained.

[0133] Optionally, the preset loss function can be a cross-entropy loss function, a mean square error, etc., which is not limited in the embodiment.

[0134] Optionally, the preset recall rate threshold can be 95%, or can be set autonomously according to requirements, which is not limited in the embodiment.

[0135] The first recall rate refers to the recall rate determined based on the target validation set. The second recall rate refers to the recall rate determined based on the target test set.

[0136] The recall rate refers to the proportion of the disks successfully warned in advance in the third time period before the failure occurs. The recall rate is the ratio of the number of real failure disks warned in advance in the third time period to the total number of predicted disks.

[0137] Optionally, the preset gradient descent algorithm can be a stochastic gradient descent algorithm, and the embodiment is not limited.

[0138] The stochastic gradient descent is an iterative algorithm for solving optimization problems.

[0139] The back propagation algorithm is a supervised learning algorithm for training neural networks, and the core purpose is to adjust the weight parameters in the neural network so that the output of the network is as close as possible to the actual target value, thereby minimizing the loss function.

[0140] Specifically, by dividing the target sample set into a target training set, a target validation set and a target test set, and training the initial disk failure prediction model multiple times through the back propagation algorithm, the target validation set is used to monitor and screen the initial disk failure prediction model after each training in the training process, which can timely find the overfitting trend of the initial disk failure prediction model after each training, and select the model with better generalization performance as the target disk failure prediction model. Further, the target test set is used to finally evaluate the target disk failure prediction model, and the preset disk failure prediction model is determined, thereby improving the reliability of the preset disk failure prediction model and ensuring that the preset disk failure prediction model has strong generalization ability.

[0141] As an optional implementation, on the basis of any one of the above embodiments, the target sample set is obtained, including:

[0142] The historical data of the plurality of training disks in a second preset time period is obtained; the historical data includes parameter data, historical environment data and historical load data of each training disk;

[0143] The historical data in the second preset time period is scaled based on a preset scaling algorithm to obtain scaled historical data;

[0144] The scaled historical data is feature extracted to obtain training feature data of each training disk;

[0145] It is determined whether each training disk fails in a third preset time period; the third preset time period refers to a time period after the second preset time period and continuous with the second preset time period;

[0146] If each training disk fails in the third preset time period, a first label is generated and added to the training feature data of the corresponding training disk;

[0147] If each training disk does not fail within the third preset time period, a second label is generated and added to the training feature data of the corresponding training disk;

[0148] The historical health scores of each training disk are obtained;

[0149] The third label is generated based on the historical health scores of each training disk;

[0150] The disk type of each training disk is determined;

[0151] The fourth label is generated based on the disk type of each training disk;

[0152] The disk identification information corresponding to each training disk is obtained;

[0153] The disk identification information corresponding to each training disk, the training feature data within the second preset time period, and the corresponding label are determined as the corresponding training disk sample;

[0154] The training disk samples including the first label and the training disk samples including the second label are sampled according to a preset ratio, and the sampled training disk samples are determined as the target sample set.

[0155] Each training disk corresponds to one disk.

[0156] Optionally, the second preset time period can be 30 days, or other days, which is not limited in the embodiment.

[0157] Optionally, the third preset time period can be 3 days, or other days, which is not limited in the embodiment.

[0158] Specifically, in this embodiment, the server device obtains historical data of the plurality of training disks in a second preset time period, scales the parameter data of each training disk in the second preset time period using a preset scaling algorithm, and thus obtains scaled historical data. The scaled historical data is subjected to corresponding feature extraction to obtain training feature data of each training disk. It is determined whether each training disk has failed in a third preset time period. If each training disk has failed in the third preset time period, a first label is generated and added to the training feature data of the corresponding training disk. If each training disk has not failed in the third preset time period, a second label is generated and added to the training feature data of the corresponding training disk. Further, the historical health scores of each training disk are obtained from a preset database, and the historical health scores of each training disk are taken as a third label. The disk types of each training disk are determined, and the disk types of each training disk are taken as a fourth label. The disk identification information corresponding to each training disk is obtained from the preset database, and the disk identification information corresponding to each training disk, the training feature data in the second preset time period, and the corresponding label are determined as a corresponding training disk sample. The training disk samples including the first label and the training disk samples including the second label are sampled in a preset ratio, and the sampled training disk samples are determined as a target sample set.

[0159] Optionally, the preset ratio can be 1:10, i.e., the ratio of the training disk samples including the first label to the training disk samples including the second label is 1:10.

[0160] Illustratively, the first label can be 1 and the second label can be 0. Assuming that the second preset time period is 30 days and the third preset time period is 3 days, and assuming that February 1 is the 30th day, within 3 days after February 1, if the disk fails, it is marked as 1, and if the disk does not fail, it is marked as 0.

[0161] The disk identification information is used to distinguish different training disks.

[0162] Specifically, through the failure label generation mechanism of the third preset time period, a supervised learning framework with clear time continuity is constructed, ensuring that the model learns the fault precursor features of the disk state evolving over time, rather than static attribute associations. Further, the positive and negative samples are sampled in a preset ratio, thereby solving the class imbalance problem naturally existing in the disk failure data.

[0163] Figure 3 A flowchart of a disk failure prediction method provided by another embodiment of the present application is shown in FIG. 2. The disk failure prediction method provided by the present embodiment specifically includes the following steps: Figure 3

[0164] ​S301: Acquire target feature data of a target disk.

[0165] S302: Using a preset disk failure prediction model and based on target feature data of the target disk, predict the target disk, and output a target prediction probability and a target disk health score.

[0166] S303: If the target prediction probability is greater than the first preset threshold or the target disk health score is greater than the second preset threshold, the preset disk failure prediction model is input into the preset feature contribution interpreter for initialization to obtain an initialized preset feature contribution interpreter.

[0167] S304: Input the target feature data into a preset feature contribution interpreter, and use the preset feature contribution interpreter to calculate the contribution of feature data in the target feature data, and output the feature contribution corresponding to each feature data.

[0168] S305: sorting the feature contribution of each feature data from large to small, and obtaining the feature contribution of a preset number of feature data in sequence from front to back after the sorting.

[0169] S306: Inputting the feature contribution of a preset number of feature data and target feature data into a preset fault analysis model, and using the preset fault analysis model to predict the failure cause of the target disk, and outputting the target failure cause.

[0170] Through the description of the above implementation methods, those skilled in the art can clearly understand that the method according to the above embodiment can be implemented by means of software plus the necessary general hardware platform, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method.

[0171] Figure 4 This is a schematic diagram of the structure of a disk failure prediction device provided by an embodiment of the present application. Figure 4 As shown, the disk failure prediction method described above is implemented by a disk failure prediction device. This disk failure prediction device can be implemented via a computer program; it can also be implemented via a medium storing the relevant computer program, such as a USB flash drive and / or optical disk, or it can also be implemented via a physical device, such as an electronic device, that integrates or installs the relevant computer program. The electronic device can be a computer or a server device. The disk failure prediction device provided in this embodiment is located in the electronic device. The disk failure prediction device 40 provided in this embodiment includes: an acquisition module 41, a prediction module 42, and a determination module 43.

[0172] Specifically, the acquisition module 41 is configured to acquire target feature data of a target disk. The target feature data includes a plurality of feature data. The target feature data is used to reflect a current state of the target disk. The target feature data includes a disk type of the target disk. The prediction module 42 is configured to use a preset disk failure prediction model and predict the target disk based on the target feature data of the target disk, and output a target prediction probability. The target prediction probability is a probability of the target disk failing in a future target time period. The determination module 43 is configured to determine feature contribution degrees of the feature data based on the preset disk failure prediction model and the target feature data if it is determined that the target prediction probability is greater than a first preset threshold. The determination module 43 is further configured to use a preset failure analysis model and determine a target failure cause based on the target feature data and the feature contribution degrees of the feature data, and send the target failure cause to a client.

[0173] Optionally, the acquisition module 41 is configured to acquire original data in a first preset time period when acquiring the target feature data of the target disk. The original data includes parameter data, target environment data, and target load data of the target disk. The original data in the first preset time period is scaled by using a preset scaling algorithm to obtain scaled original data. The scaled original data is feature extracted to obtain the target feature data of the target disk.

[0174] Optionally, the prediction module 42 is configured to input the target feature data of the target disk into the preset disk failure prediction model, and predict the target disk by using the preset disk failure prediction model, and output the target prediction probability when the prediction module 42 uses the preset disk failure prediction model and predicts the target disk based on the target feature data of the target disk, and outputs the target prediction probability.

[0175] Optionally, the determination module 43 is configured to input the preset disk failure prediction model into a preset feature contribution degree interpreter to obtain an initialized preset feature contribution degree interpreter, and input the target feature data into the preset feature contribution degree interpreter, and calculate the contribution degrees of the feature data in the target feature data by using the preset feature contribution degree interpreter, and output the feature contribution degrees corresponding to the feature data if the determination module 43 determines that the target prediction probability is greater than the first preset threshold, and determines the feature contribution degrees of the feature data based on the preset disk failure prediction model and the target feature data.

[0176] Optionally, the determining module 43 is configured to, when the target fault cause is determined based on the target feature data and the feature contribution degrees of the feature data by using the preset fault analysis model, sort the feature contribution degrees of the feature data in descending order to obtain the sorted feature contribution degrees of the feature data. The feature contribution degrees of a preset number of feature data are sequentially obtained from the front to the back in the sorted feature contribution degrees of the feature data. The feature contribution degrees of the preset number of feature data and the target feature data are input into the preset fault analysis model, and the preset fault analysis model is used to predict the fault cause of the target disk and output the target fault cause.

[0177] Optionally, the disk fault prediction apparatus further comprises a generating module.

[0178] Correspondingly, the determining module 43 is further configured to determine whether the target prediction probability is greater than the first fault threshold and less than the second fault threshold. The generating module is configured to generate first prompt information if the target prediction probability is greater than the first fault threshold and less than the second fault threshold. The first prompt information is used to prompt to back up data in the target disk. The second prompt information is generated if the target prediction probability is greater than or equal to the second fault threshold. The second prompt information is used to prompt to replace the target disk. The determining module 43 is further configured to determine whether the target prediction probability is less than the third fault threshold if the target prediction probability is less than the first fault threshold. The generating module is further configured to generate third prompt information if the target prediction probability is less than the third fault threshold. The third prompt information is used to prompt that the target disk is currently in a healthy state.

[0179] Optionally, the disk fault prediction apparatus further comprises a scoring module.

[0180] Correspondingly, the scoring module is configured to input the target feature data of the target disk into the preset disk fault prediction model, use the preset disk fault prediction model to score the health of the target disk, and output a target disk health degree score. The target disk health degree score is used to identify the health state of the target disk. The determining module 43 is further configured to determine the feature contribution degrees of the feature data based on the preset disk fault prediction model and the target feature data if the target disk health degree score is greater than a second preset threshold. The target fault cause is determined based on the target feature data and the feature contribution degrees of the feature data by using the preset fault analysis model, and is sent to the client.

[0181] Optionally, the disk fault prediction apparatus further comprises a dividing module and a training model.

[0182] Correspondingly, the acquisition module 41 is configured to acquire a target sample set and an initial disk failure prediction model before predicting the target disk based on the target feature data of the target disk by using the preset disk failure prediction model. The division module is configured to divide the target sample set into a target training set, a target validation set and a target test set. The training model is configured to perform round training on the initial disk failure prediction model based on the target training set by using a back propagation algorithm to obtain an initial disk failure prediction model after each round of training. The determination module 43 is configured to determine the target disk failure prediction model based on the target validation set and the initial disk failure prediction model after each round of training. The preset disk failure prediction model is determined based on the target test set and the target disk failure prediction model.

[0183] Optionally, the acquisition module 41, when acquiring the target sample set, is specifically configured to: acquire historical data of a plurality of training disks within a second preset time period. The historical data includes parameter data, historical environment data and historical load data of each training disk. The historical data is scaled based on the second preset time period by using a preset scaling algorithm to obtain scaled historical data. The scaled historical data is feature extracted to obtain training feature data of each training disk. It is determined whether each training disk fails within a third preset time period. The third preset time period refers to a time period after the second preset time period and continuous with the second preset time period. If each training disk fails within the third preset time period, a first label is generated and added to the training feature data of the corresponding training disk. If each training disk does not fail within the third preset time period, a second label is generated and added to the training feature data of the corresponding training disk. The historical health scores of each training disk are acquired. The third label is generated based on the historical health scores of each training disk. The disk types of each training disk are determined. The fourth label is generated based on the disk types of each training disk. The disk identification information corresponding to each training disk is acquired. The disk identification information corresponding to each training disk, the training feature data within the second preset time period and the corresponding label are determined as the corresponding training disk sample. The training disk samples including the first label and the training disk samples including the second label are sampled according to a preset ratio, and the sampled training disk samples are determined as the target sample set.

[0184] The features of the embodiments of the disk failure prediction device can be referred to the related descriptions of the embodiments of the disk failure prediction method, which will not be repeated here.

[0185] Figure 5 The structural schematic diagram of the electronic device provided by an embodiment of the present application is shown in FIG. 1. As shown in FIG. 1, the electronic device 50 provided by the embodiment of the present application includes a memory 52 and a processor 51. Figure 5

[0186] ​The memory 52 stores a computer program, and the processor 51 is configured to execute the computer program to perform the steps in any of the above-mentioned embodiments of the disk failure prediction method.

[0187] The specific implementation process of the processor 51 can refer to the above-mentioned method embodiments, which have similar implementation principles and technical effects, and will not be described here.

[0188] In the above-mentioned embodiments, it should be understood that the processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), etc. The general-purpose processor can be a microprocessor or any conventional processor, etc. The steps of the method disclosed in the application can be directly embodied as hardware processor execution, or executed by a combination of hardware and software modules in the processor.

[0189] The memory can include a random access memory (RAM), and can also include a non-volatile memory (NVM), such as at least one disk memory.

[0190] The bus can be an industry standard architecture (ISA) bus, a peripheral component (PCI) bus, or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For the convenience of representation, the bus in the drawings of the present application does not limit to only one bus or one type of bus.

[0191] The embodiments of the present application also provide a computer readable storage medium, which stores a computer program, wherein the computer program is configured to execute the steps in any of the above-mentioned embodiments of the disk failure prediction method when running.

[0192] In an example embodiment, the computer readable storage medium described above can include, but is not limited to, a U disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store computer programs.

[0193] Embodiments of the present application also provide a computer program product, which comprises a computer program, and the computer program, when executed by a processor, implements the steps in any of the above-mentioned disk failure prediction method embodiments.

[0194] Embodiments of the present application also provide another computer program product, which comprises a non-volatile computer readable storage medium, and the non-volatile computer readable storage medium stores a computer program, and the computer program, when executed by a processor, implements the steps in any of the above-mentioned disk failure prediction method embodiments.

[0195] The skilled in the art can further realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware, computer software or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been described in the above description in general terms. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. The skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0196] The above describes in detail a device information display method provided by the present application. The principles and implementation manners of the present application are described by applying specific examples. The above description of the examples is only used to help understand the method of the present application and its core idea. It should be pointed out that, for those skilled in the art, without departing from the principles of the present application, some improvements and modifications can be made to the present application, and these improvements and modifications also fall within the protection scope of the claims of the present application.

Claims

1. A method of predicting disk failure, the method comprising: The method comprises: obtaining target feature data of a target disk; the target feature data comprises a plurality of feature data; the target feature data is used to reflect the current state of the target disk; the target feature data comprises the disk type of the target disk; using a preset disk failure prediction model and based on the target feature data of the target disk to predict the target disk, and output a target prediction probability; the target prediction probability is the probability of the target disk failing in a future target time period; if it is determined that the target prediction probability is greater than a first preset threshold, determining the feature contribution degree of each feature data based on the preset disk failure prediction model and the target feature data; using a preset failure analysis model and based on the target feature data and the feature contribution degree of each feature data to determine a target failure cause, and sending the target failure cause to a client.

2. The magnetic disk failure prediction method according to claim 1, characterized by, The method comprises: obtaining original data in a first preset time period; the original data comprises parameter data, target environment data and target load data of the target disk; scaling the original data in the first preset time period using a preset scaling algorithm to obtain scaled original data; performing feature extraction on the scaled original data to obtain the target feature data of the target disk.

3. The magnetic disk failure prediction method according to claim 1, characterized by, The method comprises: inputting the target feature data of the target disk into the preset disk failure prediction model, and using the preset disk failure prediction model to predict the target disk and output the target prediction probability.

4. The magnetic disk failure prediction method of claim 1, wherein, The method comprises: inputting the preset disk failure prediction model into a preset feature contribution degree interpreter to initialize the preset feature contribution degree interpreter to obtain an initialized preset feature contribution degree interpreter; inputting the target feature data into the preset feature contribution degree interpreter, and using the preset feature contribution degree interpreter to calculate the contribution degree of the feature data in the target feature data, and outputting the feature contribution degree corresponding to each feature data.

5. The magnetic disk failure prediction method of claim 1, wherein, The method comprises: sorting the feature contribution degrees of each feature data from large to small to obtain the sorted feature contribution degrees of each feature data; obtaining the feature contribution degrees of a preset number of feature data from the sorted feature contribution degrees of each feature data from front to back; inputting the feature contribution degrees of the preset number of feature data and the target feature data into the preset failure analysis model, and using the preset failure analysis model to predict the failure cause of the target disk, and outputting the target failure cause.

6. The magnetic disk failure prediction method of claim 1, wherein, The method further comprises: determining whether the target prediction probability is greater than a first failure threshold and less than a second failure threshold; generating first prompt information if the target prediction probability is greater than the first failure threshold and less than the second failure threshold; the first prompt information is used to prompt backup of data in the target disk; generating second prompt information if the target prediction probability is greater than or equal to the second failure threshold; the second prompt information is used to prompt replacement of the target disk; determining whether the target prediction probability is less than a third failure threshold if the target prediction probability is less than the first failure threshold; generating third prompt information if the target prediction probability is less than the third failure threshold; the third prompt information is used to prompt that the target disk is currently in a healthy state.

7. The magnetic disk failure prediction method of claim 1, wherein, The method further comprises: inputting the target feature data of the target disk into the preset disk failure prediction model, and using the preset disk failure prediction model to perform health scoring on the target disk, and outputting a target disk health degree score; the target disk health degree score is used to identify the health state of the target disk; determining the feature contribution degree of each feature data based on the preset disk failure prediction model and the target feature data if it is determined that the target disk health degree score is greater than a second preset threshold; determining a target failure cause based on the target feature data, the feature contribution degree of each feature data, and using a preset failure analysis model, and sending to a client.

8. The magnetic disk failure prediction method of claim 1, wherein, Before the preset disk failure prediction model is used to predict the target disk based on the target feature data of the target disk, the method further comprises: obtaining a target sample set and an initial disk failure prediction model; dividing the target sample set into a target training set, a target validation set, and a target test set; using a back propagation algorithm and based on the target training set, performing round training on the initial disk failure prediction model to obtain an initial disk failure prediction model after each round of training; determining a target disk failure prediction model based on the target validation set and the initial disk failure prediction model after each round of training; determining the preset disk failure prediction model based on the target test set and the target disk failure prediction model.

9. The magnetic disk failure prediction method of claim 8, wherein, The target sample set is obtained, comprising: obtaining historical data of a plurality of training disks within a second preset time period; the historical data includes parameter data, historical environment data, and historical load data of each training disk; using a preset scaling algorithm and based on the historical data within the second preset time period to perform scaling processing to obtain scaled historical data; performing feature extraction on the scaled historical data to obtain training feature data of each training disk; determining whether each training disk fails within a third preset time period; the third preset time period refers to a time period after the second preset time period and continuous with the second preset time period; generating a first label and adding it to the training feature data of the corresponding training disk if each training disk fails within the third preset time period; If each of the training disks does not fail within the third preset time period, a second label is generated and added to the training feature data corresponding to the training disk; a historical health score of each of the training disks is obtained; a third label is generated based on the historical health score of each of the training disks; a disk type of each of the training disks is determined; a fourth label is generated based on the disk type of each of the training disks; disk identification information corresponding to each of the training disks is obtained; the disk identification information corresponding to each of the training disks, the training feature data within the second preset time period, and the corresponding label are determined as the corresponding training disk sample; the training disk samples including the first label and the training disk samples including the second label are respectively sampled at a preset ratio, and the sampled training disk samples are determined as the target sample set.

10. An electronic device, comprising: comprising: a memory for storing a computer program; a processor for executing the computer program to implement the steps of the disk failure prediction method according to any one of claims 1 to 9.

Images