Method and apparatus for predicting failure information

Abstract

The application discloses a kind of prediction methods and devices of fault information, it is related to computer technical field, comprising: receiving fault prediction request;According to target operation information, generate the first life information of the target device in the target time period under the influence of the target operation information, according to reference operation information and service association information, generate the second life information of the target device;According to the first life information and the second life information, predict the target fault information of the target device in the target time period, solve the technical problem that the prediction efficiency of device fault information is lower in related art, reach the technical effect of improving the prediction efficiency of device fault information.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computers, and particularly relates to a fault information prediction method and device. BACKGROUND

[0002] With the development and popularization of the Internet of Things technology, device cluster technology is applied, a plurality of functionally similar devices are associated and bound, thereby a device composed of a plurality of devices is constructed, and then the device cluster can handle the business accepted by the plurality of devices together, thereby accelerating the business handling rate on the device cluster. In order to guarantee the business handling efficiency of the device cluster, if the device failure can be perceived in advance, the device failure can be intervened in advance, thereby avoiding the influence of the device failure on the business handled on the server cluster.

[0003] The current commonly used method is a fault prediction based on the running information of the device itself, that is, the possibility of device failure in the future time is preliminarily predicted according to the current running state of the device, but in the device clustering scenario, a plurality of devices are used to execute the business operations of different business stages included in the to-be-handled business, that is, there is a relatively complex influence relationship between the plurality of devices in the business handling process, and in this way, the conventional device fault prediction method cannot obtain accurate device fault information. SUMMARY

[0004] The present application provides a fault information prediction method and device, to at least solve the technical problem of low prediction efficiency of device fault information in the related art.

[0005] The application provides a fault information prediction method, comprising the following steps: receiving a fault prediction request, wherein the fault prediction request is used to request to predict fault information of a target device in a target time period after a current time, the target device being included in a device cluster, and the device cluster being used to perform business operations of different business stages of to-be-handled businesses by a plurality of devices; generating first life information of the target device in the target time period under the influence of target operation information, the target operation information being used to indicate a business handling situation of the target device in a reference time period before the current time; generating second life information of the target device according to reference operation information and business association information, the reference operation information being used to indicate a business handling situation of a reference device in the reference time period, the reference device being different from the target device among the plurality of devices, the second life information being used to indicate an influence of the reference device on a device service life of the target device in the target time period, and the business association information being used to indicate a business association relationship between the plurality of devices in a current business handling process of the device cluster; and predicting target fault information of the target device in the target time period according to the first life information and the second life information.

[0006] The application further provides a fault information prediction device, comprising a receiving module, a processing module and a prediction module, wherein the receiving module is used to receive a fault prediction request, wherein the fault prediction request is used to request to predict fault information of a target device in a target time period after a current time, the target device being included in a device cluster, and the device cluster being used to perform business operations of different business stages of to-be-handled businesses by a plurality of devices; the processing module is used to generate first life information of the target device in the target time period under the influence of target operation information, the target operation information being used to indicate a business handling situation of the target device in a reference time period before the current time; generate second life information of the target device according to reference operation information and business association information, the reference operation information being used to indicate a business handling situation of a reference device in the reference time period, the reference device being different from the target device among the plurality of devices, the second life information being used to indicate an influence of the reference device on a device service life of the target device in the target time period, and the business association information being used to indicate a business association relationship between the plurality of devices in a current business handling process of the device cluster; and the prediction module is used to predict target fault information of the target device in the target time period according to the first life information and the second life information.

[0007] The application further provides an electronic device, comprising a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to execute the computer program to realize steps of any one of the above fault information prediction methods.

[0008] The application further provides a computer readable storage medium, and the computer readable storage medium stores a computer program, wherein the computer program is executed by a processor to realize the steps of the above-mentioned any fault information prediction method.

[0009] The application further provides a computer program product, comprising a computer program, wherein the computer program is executed by a processor to realize the steps of the above-mentioned any fault information prediction method.

[0010] Through the application, after receiving a fault prediction request, on the one hand, the first life information under the target running information is generated according to the target running information of the target device in the reference time period, and on the other hand, the second life information of the target device is generated according to the reference running information of the reference device and the service association information, so as to realize the prediction of the remaining service life of the target device in the future time according to the service handling condition of the target device in the historical time, and the prediction of the life influencing factor of the target device by the reference device in the historical time according to the service handling condition of the reference device in the device cluster except the target device, and then the target fault information of the target device is predicted according to the first life information and the second life information, so as to realize the prediction of the fault information of the target device in combination with the running state of the device in the device cluster in the service handling process, thereby guaranteeing the accuracy and reliability of the target fault information predicted for the target device, and thus the technical problem of low prediction efficiency of the device fault information in the related art can be solved, and the technical effect of improving the prediction efficiency of the device fault information is achieved. BRIEF DESCRIPTION OF DRAWINGS

[0011] In order to more clearly illustrate the embodiments of the application, the drawings needed in the embodiments will be briefly introduced below. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0012] Figure 1 is a hardware structure block diagram of the fault information prediction of the embodiments of the application;

[0013] Figure 2 is a flowchart of the fault information prediction according to the embodiments of the application;

[0014] Figure 3 is an optional BMC cluster management system construction schematic diagram according to the embodiments of the application;

[0015] Figure 4 is an optional BMC cluster management system schematic diagram according to the embodiments of the application;

[0016] Figure 5is an optional fault prediction flowchart according to an embodiment of the present application;

[0017] Figure 6 is a structural block diagram of a fault information prediction device according to an embodiment of the present application. DETAILED DESCRIPTION

[0018] The technical solutions in the embodiments of the present application will be clearly and completely described in connection with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, any other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0019] 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 a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such a process, method, article or device. 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.

[0020] In order for those skilled in the art to better understand the technical solutions of the present application, the present application will be further described in detail below in connection with the drawings and specific embodiments.

[0021] In connection with the specific application environment architecture or the specific hardware architecture on which the execution of the fault information prediction method depends, the specific application environment architecture or the specific hardware architecture is described here.

[0022] The method embodiments provided in the embodiments of the present application can be executed in a server device or similar computing device. Taking the case of running on a server device, Figure 1 is a hardware structural block diagram of the fault information prediction of the embodiments of the present application. As shown in Figure 1 , the server device can include one or more (only one is shown in Figure 1 ) processors 102 (the processor 102 can include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data, wherein the above-mentioned server device can further include a transmission device 106 for communication function and an input and output device 108. Those skilled in the art can understand that Figure 1 The structure shown is only schematic, which does not limit the structure of the above-mentioned server device. For example, the server device can further include more or fewer components than those shown in Figure 1 , or have a structure different from that shown in Figure 1different configurations.

[0023] The memory 104 can be used to store computer programs, such as software programs of application software and modules, such as the computer program corresponding to the prediction of failure information in the embodiments of the present application. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, that is, implements the above-mentioned method. The memory 104 can include a high-speed random access memory, and can also include a non-volatile memory, such as one or more magnetic storage devices, flash memories, or other non-volatile solid-state memories. In some examples, the memory 104 can further include a memory remotely arranged with respect to the processor 102, which can be connected to a server device through a network. Examples of the above-mentioned network include but are not limited to the Internet, an intranet, a local area network, a mobile communication network, and a combination thereof.

[0024] The transmission device 106 is used to receive or send data via a network. The specific examples of the above-mentioned network can include a wireless network provided by a communication provider of a server device. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC) which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (Radio Frequency, RF) module which is used to communicate with the Internet in a wireless manner.

[0025] The embodiments of the present application provide a prediction of failure information, and the method is described in detail in combination with an execution flow of the prediction of failure information.

[0026] In the embodiments of the present application, a prediction of failure information is provided, Figure 2 is a flowchart of the prediction of failure information according to the embodiments of the present application, as Figure 2 shown, the method includes the following steps:

[0027] In step S202, a failure prediction request is received, wherein the failure prediction request is used to request to predict failure information of a target device in a target time period after a current time, the target device being included in a plurality of devices of a device cluster, and the device cluster being used to execute business operations of different business stages of to-be-performed businesses by the plurality of devices.

[0028] S204, generating first life information of the target device in the target time period under the influence of the target operation information according to the target operation information, and generating second life information of the target device according to reference operation information and service association information, wherein the target operation information is used to indicate service handling of the target device in a reference time period before the current time, the reference operation information is used to indicate service handling of a reference device in the reference time period, the second life information is used to indicate an influence of the reference device on the device service life of the target device in the target time period, and the service association information is used to indicate a service association relationship between the plurality of devices in a service handling process currently undertaken by the device cluster;

[0029] S206, predicting target failure information of the target device in the target time period according to the first life information and the second life information.

[0030] Through the above steps, after receiving the failure prediction request, on the one hand, the first life information under the influence of the target operation information is generated according to the target operation information of the target device in the reference time period, and on the other hand, the second life information of the target device is generated according to the reference operation information of the reference device and the service association information, so as to realize the prediction of the remaining service life of the target device in the future time according to the service handling of the target device in the historical time, and the prediction of the life influence factor of the target device by the reference device in the historical time according to the service handling of the reference device in the historical time, and then the target failure information of the target device is predicted according to the first life information and the second life information, so as to realize the prediction of the target failure information of the target device combined with the running state of the device in the service handling process of the device cluster, thereby ensuring the accuracy and reliability of the predicted target failure information of the target device, and thus the technical problem of low prediction efficiency of the device failure information in the related art can be solved, and the technical effect of improving the prediction efficiency of the device failure information is achieved.

[0031] In the embodiment provided in step S202, the device cluster improves the processing efficiency of the device cluster for the service by calling multiple devices to jointly complete the service on the device cluster. That is, when the multiple devices in the device cluster process the service accepted by the device cluster, the multiple devices have an association relationship, which can include but is not limited to a master-slave relationship, a peer-to-peer relationship, a pipeline relationship, a sharding relationship, and a redundancy relationship. In the master-slave relationship, one device acts as a master device to control multiple slave devices, and the slave devices only execute the instructions of the master device. This relationship is suitable for scenarios such as database read-write separation (for example, master database writing and slave database reading) and industrial automation. In the peer-to-peer relationship, all devices have equal roles and can independently process requests or cooperate with each other. This relationship is suitable for scenarios such as blockchain nodes, distributed storage, and edge computing nodes cooperating to process data. In the pipeline relationship, devices process data in a fixed order, and the output of a previous device is the input of a next device. This relationship is suitable for pipeline data processing scenarios. In the sharding relationship, data or tasks are horizontally divided, and each device processes a specific subset. This relationship is suitable for scenarios such as database partitioning and search engine index sharding. In the redundancy relationship, multiple devices provide the same function and achieve high availability through backup. This relationship is suitable for scenarios such as web server clusters (Nginx load balancing backend multiple instances) and Kubernetes multiple replica pods.

[0032] In the embodiment provided in step S204, the target running information is information reflecting the service processing situation of the target device in the reference time period. The target running information can include but is not limited to the service load of the target device in the target time period, the service processing efficiency of the target device in the target time period, the running state of the device component deployed in the target device for processing the service, and the like. The present scheme does not limit this.

[0033] Optionally, in the embodiment of the present application, the first life information is the remaining service life situation under the influence of the information of the service processing situation of the target device in the historical time period. That is, only the device remaining service life situation under the influence of the historical service processing information of the target device is considered. The first life information can be but is not limited to a predicted device remaining service life value or a device remaining life score for representing the device remaining service life. The larger the device remaining life score is, the lower the possibility of device failure such as downtime of the target device in the target time period is. The smaller the device remaining life score is, the higher the possibility of device failure such as downtime of the target device in the target time period is. The present scheme does not limit this.

[0034] Optionally, in the embodiments of the present application, the target operation information can include, but is not limited to, target load information for characterizing the service load of the target device in the reference time period, or can also be target component operation information for characterizing the operation mode of the device component with service handling function deployed in the target device, and then the manner of generating the first life information of the target device in the target time period under the influence of the target operation information can include: predicting the third residual life of the target component in the target time period according to the target load information; determining the third residual life as the first life information of the target device in the target time period. Or can also be detecting the component life decay information of the target component in the target time period according to the target component operation information, converting the fourth residual life of the target device in the target time period according to the component life decay information, and determining the fourth residual life as the first life information of the target device. Or can also be predicting the third residual life of the target component in the target time period according to the target load information, and detecting the component life decay information of the target component in the target time period according to the target component operation information, and converting the fourth residual life of the target device in the target time period according to the component life decay information; weighting and summing the third residual life and the fourth residual life using a target weight parameter to obtain a fifth residual life, and determining the fifth residual life as the first life information, wherein the target weight parameter is used to indicate the influence degree of the load information and the component operation information on the residual life of the device.

[0035] Optionally, in the embodiments of the present application, the second life information is an influencing factor of the operation state of the reference device predicted in the dimension of the reference device on the device residual life of the target device in the target time period, and the second life information can be, but is not limited to, a life compensation value or a life compensation coefficient of the residual service life of the target device, the life compensation value is used to indicate the gain state of the target device on the service life of the target device at the target time, and then the life gain value indicated by the life compensation value can be added to the initial residual life of the target device in the target time period predicted based on the target device itself operation information, so as to obtain the target residual life of the target device at the target time which is fused under the influence of the device service relationship in the device cluster. The life compensation coefficient is used to indicate the compensation mode of the initial residual life of the target device predicted based on the target device itself operation information, that is, the life compensation coefficient is a scaling coefficient for indicating the magnification or reduction multiple of the initial residual life, and then the initial residual life is adjusted according to the scaling coefficient indicated by the life compensation system to obtain the target residual life.

[0036] Optionally, in the embodiment of the present application, the service association information is used to indicate the association relationship between the plurality of devices in the process of handling the current service, that is, in the process of handling different services, the calling mode of the device cluster to call the plurality of devices to complete the current service may be different due to the different service types of the services to be handled, and therefore the service association information may include but is not limited to the data transmission relationship between the plurality of devices in the process of handling the current service, the belonging relationship between the sub-services handled by the devices, and the like, which is not limited in the present scheme.

[0037] Optionally, in the embodiments of the present application, the manner of generating the second life information according to the reference operation information and the service correlation information can be: predicting reference failure information of the reference devices in the target time period according to the reference operation information, and predicting reference service load of the reference devices in the target time period according to the reference operation information, wherein the reference failure information is used to indicate service response capability of the reference devices in the target time period, and the reference service load is a service amount allocated to the reference devices in the target time period by the device cluster according to service processing capability of the reference devices in the reference time period; detecting service load compensation amount between each of the reference devices and the target device according to the service influence relationship between the devices indicated by the service correlation information and the reference service load, wherein the service load compensation amount is used to indicate influence of the operation state of the reference devices in the target time period on the service accepted by the target device in the target time period; further converting device life compensation amount corresponding to each service load compensation amount; summing up the device life compensation amounts corresponding to the plurality of reference devices to obtain total life compensation amount of the reference devices to the target device in the target time period, and the second life information includes the total life compensation amount. In the embodiments, the manner of predicting the reference failure information of the reference devices in the target time period according to the reference operation information can be: converting reference life attenuation amount corresponding to reference device component operation information of the reference devices from component operation information and life attenuation amount having a corresponding relationship, wherein the reference operation information includes the reference device component operation information, the reference device component operation information is used to indicate calling manner of the reference devices to the device components having service handling function deployed by the reference devices in the reference time period or operation manner of device elements, and the reference life attenuation amount is used to indicate attenuation of the life of the reference devices under the influence of the calling manner of the components in the reference time period; converting reference residual life of the reference devices in the target time period using the reference life attenuation amount; and determining the reference failure information corresponding to the reference residual life from residual life and failure information having a corresponding relationship, wherein the reference failure information is used to indicate probability of failure of the reference devices in the service handling process in the target time period.In the embodiment, predicting the reference service load of the reference device in the target time period according to the reference operation information comprises: constructing a service load change curve of the reference device using reference load information of the reference device at each time point in a reference time period, wherein the reference operation information comprises the reference load information, the reference load information is used to indicate a service handling condition of the reference device at the corresponding time point, and the service load change curve records a change condition of the service load of the reference device over time; and converting the reference service load that needs to be handled by the reference service in the target time period using the service load change curve.

[0038] In the embodiment provided in step S206, the target failure information is used to represent a failure occurrence probability of the target device in the service handling process in the target time period, and the manner of predicting the target failure information according to the first life information and the second life information can be that the first life information and the second life information are fused to obtain a target remaining life of the target device in the target time period, and the target failure information corresponding to the target remaining life is converted from the remaining life and the failure information having a corresponding relationship.

[0039] As an optional implementation, the generating the first life information of the target device in the target time period under the influence of the target operation information according to the target operation information comprises:

[0040] predicting a first remaining life of the target device in the target time period according to service response information of the target device in the reference time period, wherein the service response information is used to indicate a service response condition of the target device to the handled service in the reference time period; predicting a second remaining life of the target device in the target time period according to element calling information of a device element deployed in the target device in the reference time period, wherein the element calling information is used to indicate an operation manner of the device element in the reference time period, and the target operation information comprises the service response information and the element calling information;

[0041] converting a reference remaining life of the target device in the target time period using the first remaining life and the second remaining life, wherein the first life information comprises the reference remaining life.

[0042] Optionally, in the embodiment of the application, the service response information can comprise, but is not limited to, a service volume of the service handled by the target device in the reference time period, and / or a service handling rate of the target device, and the like, and the present scheme does not limit this.

[0043] Optionally, in the embodiments of the present application, the device element is an element with a service execution function deployed in the target device, and the element calling information is used to indicate a calling manner of the device element in the process of service execution of the target device within the reference time period. The element calling information can include, but is not limited to, a calling period of the device element, a calling frequency, a device element running parameter (such as running power, rotating speed, output power, temperature), and the like. The present scheme does not limit this.

[0044] Optionally, in the embodiments of the present application, the manner of predicting the first residual life of the target device within the target time period according to the service response information of the target device within the reference time period can be: performing feature extraction on the service response information of each time within the reference time period, so as to obtain an information feature corresponding to each service response information, the information feature carrying a time sequence influence relationship between the service response information of each time, and then performing feature recognition on the information feature within the reference time period, so as to obtain a life influence amount of the service response information of each time on the service life of the target device at the target time, and then fusing the life influence amount of each time, so as to obtain a total life influence amount, and converting the first residual life by using the total life influence amount. In the embodiments, the manner of performing feature extraction on the service response information of each time within the reference time period, so as to obtain an information feature corresponding to each service response information can include, but is not limited to: sorting the service response information of each time within the reference time period according to an implementation order, so as to obtain a service response information sequence, inputting the service response information sequence into a bidirectional LSTM (Bi-LSTM) layer, so as to capture a long-term dependence relationship and context information of the service response information, and so as to obtain the information feature.

[0045] Optionally, in the embodiments of the present application, the manner of predicting the second residual life of the target device within the target time period according to the element calling information of the device element deployed in the target device within the reference time period can be: constructing an element life decay curve of the target device by using the element calling information, wherein the element life decay curve records a decay relationship of the element life with time, and then determining the second residual life of the target device within the target time period according to the decay relationship of the element life recorded in the element life decay curve.

[0046] Optionally, in the embodiments of the present application, the reference residual life can be obtained by merging the first residual life and the second residual life according to the importance of the first residual life and the second residual life, but the present scheme does not limit this.

[0047] Through the above content, by analyzing the service response information and the element calling information, the first residual life and the second residual life are predicted respectively, and then comprehensive evaluation is performed, so as to ensure the comprehensiveness and accuracy of residual life prediction.

[0048] As an optional implementation, the predicting the first residual life of the target device in the target time period according to the service response information of the target device in the reference time period comprises:

[0049] performing feature recognition on the log sequence of the target device in the reference time period to obtain log features of service response logs in the log sequence, the service response logs corresponding to a plurality of time points of the target device in the reference time period, wherein the service response information comprises a plurality of the service response logs of the target device in the reference time period, the service response logs being used to indicate a service execution manner of the target device for the service at the corresponding time point, and the log features being used to indicate a time sequence dependency between the current service response log and other logs in the log sequence except the current service response log;

[0050] sorting the log features corresponding to the plurality of service response logs in the reference time period in a time sequence order to obtain a log feature sequence;

[0051] inputting the log feature sequence into a target conversion model to obtain a life attenuation parameter corresponding to each of the service response logs output by the target conversion model, wherein the life attenuation parameter is used to represent an influence of the service execution manner indicated by the service response log on the service life of the target device in the target time period, and the target conversion model records a conversion relationship between log sequence features and the life attenuation parameters of the service response logs in the log sequence;

[0052] summarizing the life attenuation parameters corresponding to the plurality of service response logs to obtain a first life attenuation parameter, wherein the first life attenuation parameter is used to indicate a service life attenuation of the target device in the target time period caused by the service execution manner of the target device in the reference time period;

[0053] using the first life attenuation parameter to convert the first residual life of the target device.

[0054] Optionally, in the embodiments of the present application, the service response log can record, but is not limited to, the target device service bearing information, which can include, but is not limited to, service bearing capacity, service execution efficiency, and the like, and the present scheme does not limit this.

[0055] By means of the embodiments of the present application, the deep learning model is used to perform feature recognition and life attenuation prediction on the device log, thereby realizing deep analysis of device failure. In principle, by means of time series data processing, the change trend of the device running state is captured, thereby providing a theoretical basis for predicting the remaining life of the device. In terms of effect, the life attenuation of the device is predicted by means of the deep learning model, thereby more accurately evaluating the health state of the device, performing maintenance in advance, and avoiding losses caused by device failure.

[0056] As an optional implementation, the second remaining life of the target device in the target time period is predicted according to element call information of device elements deployed in the target device in the reference time period, and the method comprises the following steps:

[0057] obtaining an element call log of each of the device elements deployed in the target device, wherein the element call log records the running state of the device element at each moment in the reference time period, and the element call information comprises the element call log;

[0058] performing feature extraction on the element call log to obtain time-domain call features and frequency-domain call features of the device element, wherein the time-domain call features are used to indicate the call time of the device element in the reference time period, and the frequency-domain call features are used to indicate the call frequency of the device element at the corresponding call time;

[0059] inputting the time-domain call features and the frequency-domain call features into a reference conversion model to obtain an element life attenuation parameter of the device element output by the reference conversion model, wherein the element life attenuation parameter is used to indicate the influence of the call mode of the target device on the device element in the reference time period on the element life of the device element in the target time period;

[0060] summarizing the element life attenuation parameters of all the device elements deployed in the target device to obtain a second life attenuation parameter of the target device in the target time period, wherein the second life attenuation parameter is used to indicate the influence of the call mode of the target device on the device element in the reference time period on the device life of the target device in the target time period;

[0061] using the second life attenuation parameter to convert the second remaining life of the target device.

[0062] Optionally, in the embodiments of the present application, the element call log can record, but is not limited to, running parameters in the running process of the device element, such as running power, running time, call frequency, output power, running temperature, and the like, and the present solution does not limit this.

[0063] Optionally, in the embodiments of the present application, the reference conversion model can be, but is not limited to, a gradient boosting tree model, which is trained with feature vectors as input and historical equipment failure data or manually labeled labels. The input of the model is the smoothed and normalized hardware sensor time series data (i.e. element call log), and the output is the predicted value (days) and prediction confidence interval (such as "30 days ± 5 days") of the remaining useful life (RUL).

[0064] Through the above, by deeply analyzing the call log of the equipment element, the life attenuation of the equipment element is predicted, and the equipment element level failure prediction is realized. Feature extraction technology is used to analyze the call mode of the equipment element from two dimensions of time domain and frequency domain, which provides data support for predicting the life of the element. In terms of effect, by predicting the remaining life of the equipment element, the element that will fail can be replaced in time, and the equipment downtime caused by element failure is reduced.

[0065] As an optional implementation, the converting the first remaining life and the second remaining life into the reference remaining life of the target device in the target time period comprises:

[0066] obtaining a first weight parameter of the first remaining life and the second remaining life, wherein the first weight parameter is used to indicate the importance of the first remaining life and the second remaining life to representing the remaining life of the target device in the target time period;

[0067] performing weighted sum calculation on the first remaining life and the second remaining life using the first weight parameter to obtain the reference remaining life.

[0068] Through the above, by setting the weight parameter and performing weighted sum calculation on the first remaining life and the second remaining life, the comprehensive evaluation of the equipment remaining life is realized. Through weight parameter adjustment, different types of remaining life prediction results can be flexibly weighted according to the type of equipment and application scene, ensuring the applicability and accuracy of the remaining life prediction. Through comprehensive evaluation of the remaining life of the equipment, the health status of the equipment can be more comprehensively understood, providing a decision basis for equipment maintenance and update.

[0069] As an optional implementation, the generating the second life information of the target device according to the reference operation information and the business association information comprises:

[0070] correspondence relationship between the devices indicated by the service association information, to obtain a target relationship graph of the device cluster at present, wherein the target relationship graph comprises a plurality of nodes and connecting edges connecting the nodes, the plurality of nodes correspond to the running information of the plurality of devices one by one, the nodes comprise the reference running information or the target running information, and the connecting edges are used to indicate the association relationship between the running information in the nodes;

[0071] inputting the target relationship graph into a candidate conversion model to obtain a life compensation parameter output by the candidate conversion model, wherein the life compensation parameter is used to indicate a compensation situation of a service handling state of the reference device to a service life of the target device in the target time period, the second life information comprises the life compensation parameter, and the candidate conversion model records a conversion relationship between a relationship graph and a life compensation parameter.

[0072] Optionally, in the embodiment of the present application, the pointing relationship of the connecting edge in the target relationship graph is constructed according to the data exchange relationship between the plurality of devices in the current service handling process of the device cluster.

[0073] Through the above, by constructing the target relationship graph of the device cluster, analyzing the service association relationship between the devices, and predicting the life compensation parameter of the device, the device cluster level fault prediction is realized. According to the service dependence relationship between the devices, the life compensation of the device is predicted, which provides cluster level analysis for device fault prediction. By predicting the life compensation parameter of the device cluster, the health status of the device cluster can be evaluated, the maintenance and update of the cluster can be planned in advance, and the service interruption time caused by the cluster fault is reduced.

[0074] As an optional implementation, the predicting the target fault information of the target device in the target time period according to the first life information and the second life information comprises:

[0075] obtaining a second weight parameter corresponding to the life compensation parameter of the target device in the target time period, wherein the life compensation parameter is used to indicate a compensation situation of a service handling state of the reference device to a service life of the target device in the target time period, the second life information comprises the life compensation parameter, and the second weight parameter is used to indicate an influence degree of the running state of the reference device on the device service life of the target device;

[0076] performing weighted calculation on the life compensation parameter by using the second weight parameter to obtain a life compensation value;

[0077] compensate the reference residual life of the target device in the target time period using the life compensation value, to obtain a target residual life of the target device in the target time period;

[0078] predict the target failure information of the target device according to the target residual life.

[0079] Optionally, in the embodiments of the present application, the manner of predicting the target failure information of the target device according to the target residual life can include but is not limited to: predicting target service carrying information of the target device at a target time point according to reference service carrying information of the target device in a reference time period, wherein the target operation information includes the reference service carrying information; and predicting the target failure information of the target device according to the target service carrying information and the target residual life. In the embodiments, the manner of predicting the target service carrying information according to the reference service carrying information can be: constructing a service carrying curve of the target device according to the reference service carrying information, wherein the service carrying curve records the relationship between the service carrying amount of the target device and time, and then converting the target service carrying amount of the target device in the target time period using the service carrying curve, wherein the target service carrying information includes the target service carrying amount. In the embodiments, the manner of predicting the target failure information according to the target service carrying information and the target residual life can include but is not limited to: determining the target failure information corresponding to the target service carrying information and the target residual life from service carrying information, residual life and failure information having a corresponding relationship.

[0080] Optionally, in the embodiments of the present application, the target failure information is used to indicate the failure occurrence state of the target device in the target time period service handling process, and the target failure information can include but is not limited to the failure time of the target device and / or the failure type, which is not limited in the present application.

[0081] Through the above, by setting the second weight parameter, the life compensation parameter is weighted and calculated, and the further optimization of the residual life of the device is realized. Through the adjustment of the weight parameter, the life compensation of the device under the service association relationship can be more accurately reflected, and more accurate data support is provided for the device failure prediction. Through the optimization of the residual life prediction of the device, the health status of the device can be more accurately evaluated, measures can be taken in advance, and the loss caused by the device failure can be avoided.

[0082] As an optional implementation, the predicting the target failure information of the target device according to the target residual life includes:

[0083] determining a target failure probability corresponding to the target residual life from residual life and failure probability having a corresponding relationship, wherein the target failure information includes the target failure probability.

[0084] As an optional implementation, the method further comprises:

[0085] In a case where it is determined that the target device is in a fault state, obtaining cluster state information of the device cluster, wherein the cluster state information is used to indicate a running state of each of the plurality of devices included in the device cluster;

[0086] Detecting operation benefits of each of a plurality of fault maintenance operations in the cluster running state indicated by the cluster state information, wherein the operation benefits are used to indicate a benefit situation of performing the fault maintenance operation on the target device to the running quality of the device cluster in the current cluster running state;

[0087] Filtering, from the plurality of fault maintenance operations, a target fault maintenance operation that meets a target benefit condition from the operation benefits;

[0088] Performing the target fault maintenance operation on the target device.

[0089] Optionally, in the embodiments of the present application, in a case where it is detected that the target device is in a fault state, a fault maintenance based on reinforcement learning is designed, and the specific idea is as follows:

[0090] 1. State space (State): defined as the global health state matrix of the cluster, including the risk score, resource utilization, topology state, and the like of all nodes.

[0091] 2. Action space (Action): define a series of repair actions, such as: restarting BMC, restarting the server, migrating the load to the standby node, triggering firmware rollback, isolating the fault node, and the like.

[0092] 3. Reward function (Reward): carefully designed to meet business objectives, for example: successful repair of fault +100, -10 for each action performed (encouraging minimal intervention), -50 for action failure or causing the scope of the fault to expand, and -1000 for causing core services to be unavailable.

[0093] 4. Training and inference: initially use historical fault data for offline training to train a deep Q network (DQN) or policy gradient (PG) model. In online operation, the RL agent selects the optimal action to execute according to the current state, and continuously optimizes the strategy according to the reward after execution, gradually reduces the dependence on preset rules, and realizes true intelligent self-healing.

[0094] The application provides a BMC management system and method for a large server cluster based on artificial intelligence, a system, equipment and a storage medium. The application builds a BMC centralized management platform of a large server cluster to centrally manage the BMCs of all servers in the current local area network. The BMC centralized management platform can automatically scan and identify each server node in the cluster and assign a unique management identifier to each node, ensuring that the administrator can quickly understand the overall architecture of the cluster and reducing the time and workload of manual configuration. The BMC centralized management platform system is built-in with an intelligent fault detection algorithm, which can generate an alarm according to the hardware health status of the server and automatically take repair measures according to a preset strategy. The BMC centralized management platform can remotely monitor the running state of each server node in the server cluster in real time, including hardware failure, temperature, fan speed, voltage and other parameters, so that the administrator can conveniently and quickly understand the running state of the current server cluster.

[0095] Figure 3 is an optional BMC cluster management system construction schematic diagram according to an embodiment of the application, as shown in Figure 3 , a BMC centralized management platform is built as the BMC management center of a large server cluster. When the BMC centralized management platform server starts, the BMC centralized management platform can automatically scan and identify each server node in the cluster according to the configured local area network segment, assign a unique management identifier to each node, and add each server BMC identified to the management list for subsequent unified management of the BMC of each server.

[0096] Figure 4 is an optional BMC cluster management system schematic diagram according to an embodiment of the application, as shown in Figure 4As shown, when the BMC centralized management platform identifies all servers in the current local area network, it will monitor the status of all BMCs through a single interface. The platform can display the device information of each node, including the IP of the BMC, the system IP of the server where the BMC is located, the health status of the server, log information, and real-time running data. Administrators can perform maintenance tasks for individual nodes or the entire cluster through the platform, such as BMC restart, server restart, BMC firmware update, logging into the BMC web interface, server fault repair, etc. The BMC centralized management platform first builds a topology dependency graph, automatically collects the network link, data synchronization relationship, and service dependency of the cluster nodes, such as node A being the backup node of node B, node C relying on the computing power output of node D, constructs a visual topology dependency graph, marks the key dependency path, master-slave synchronization link, and heartbeat link between core data nodes. When upgrading the BMC nodes in the cluster, the BMC centralized management platform dynamically plans the upgrade sequence based on the topology dependency graph, uses the topological sorting algorithm in graph theory, and prioritizes upgrading the nodes at the end of the dependency chain. After the upgrade is completed and confirmed stable, the upstream nodes of the dependency are upgraded; for nodes with bidirectional dependency, a serial strategy of "isolation-upgrade-verification-switch" is used, such as isolating node A first, upgrading node B, verifying and switching traffic to B, and then upgrading A. Before upgrading, the BMC centralized management platform defines a "cluster firmware version consistency baseline" (such as "all nodes need to be unified to V2.0, allowing ±1 small version deviation but compatible"), and pre-stores the compatibility matrix between versions (such as "V2.0 is compatible with V1.5, and incompatible with V1.0"). During the upgrade process, after completing a batch of node upgrades, a version consistency vote is initiated through the multicast protocol within the cluster: all nodes report the current firmware version, if more than 90% of the nodes meet the baseline requirements or the deviation is within the compatible range, then continue to the next batch; if not, trigger a self-healing compensation mechanism to re-push the upgrade package to the nodes that do not meet the requirements, or perform targeted rollback on incompatible nodes that have been upgraded.

[0097] The BMC centralized management platform system has a built-in multi-modal AI fault prediction and self-healing engine, Figure 5 According to an optional fault prediction process diagram of an embodiment of the present application, as Figure 5 shown, the technical implementation details are as follows:

[0098] Multi-source data fusion and preprocessing layer; the engine first constructs a unified time series data lake for aligning and storing data from multiple heterogeneous data sources:

[0099] BMC logs: collected through IPMI commands or Redfish API, and parsed for timestamps, event types (such as CPU_CAT_ERR), sensor numbers, and description information.

[0100] Operating System Log (Syslog / Event Log): Collected through Agent or remote syslog protocol, parse its Facility, Severity, Process ID and Message Body.

[0101] Hardware Sensor Data: Polling at fixed frequency (e.g. 1 minute / interval), including but not limited to: CPU core utilization, temperature, power consumption, Cache ECC Count; Memory utilization, read / write speed, UE Count, CE Count growth rate; Storage SMART data (Reallocation Sector Count, Read / Write Error Rate, Media Wear Indicator), RAID status, I / O latency; Network card throughput, packet loss rate, link state flip count; Fan and power supply fan speed (RPM), power input power, output voltage ripple.

[0102] Topology Relationship Data: Get the service dependency, network dependency, data synchronization relationship between nodes from the dependency graph in the second step.

[0103] Data Processing Flow: All data is collected with nanosecond-level timestamp, through a data cleaning and alignment pipeline, handle clock drift and collection interval inconsistency problem, finally generate regular, with uniform time index, multi-dimensional time series data, stored in time series database.

[0104] Multi-modal AI Analysis Layer:

[0105] This layer is the brain of the engine, composed of multiple parallel AI sub-models, for collaborative analysis:

[0106] Log Anomaly Detection Sub-model based on LSTM-Transformer Hybrid Model:

[0107] 1. Input: Preprocessed, time window (e.g. 10 minutes) sliding log sequence.

[0108] 2. Technical details: First, convert each log message to a high-dimensional vector (embedding layer) through Word2Vec or BERT model; input the vectorized log sequence into the Bi-LSTM layer to capture the long-term dependency and context information between logs; input the output of LSTM into the Transformer encoder layer, use the self-attention (Self-Attention) mechanism to focus on the key log events most related to the fault.

[0109] 3. Output: Output a log anomaly score (0-100) representing the degree of log pattern anomaly in this time period, and identify the key log events that triggered the anomaly.

[0110] Hardware health degree prediction sub-model based on time series decomposition and gradient boosting tree (RUL prediction):

[0111] 1. Input: Smoothed and normalized hardware sensor time series data.

[0112] 2. Technical details: Use STL or Prophet model to decompose sensor data into trend term, seasonal term and residual term, trend term directly reflects the long-term degradation of hardware performance; extract a large number of time domain and frequency domain features (such as mean, variance, slope, FFT spectral energy, approximate entropy, etc.) from the decomposed sequence and original data to form a feature vector; use XGBoost or LightGBM gradient boosting tree model, take the feature vector as input, and train with historical failure data or manually labeled health status as label.

[0113] 3. Output: Output the predicted value (days) and prediction confidence interval (such as "30 days ± 5 days") of the remaining useful life (RUL).

[0114] Topology-dependent risk propagation sub-model based on graph neural network (GNN):

[0115] 1. Input: Server cluster topology dependency graph constructed in the second step, where node features are real-time health state vectors of each server.

[0116] 2. Technical details: Treat servers as nodes in the graph and service / data dependency relationships as edges to construct a directed graph; use graph convolution network (GCN) or graph attention network (GAT), each node updates its state representation by aggregating the state features of its neighbor nodes. For example, if an important upstream service node (such as a database) has a sharp decline in health degree, its risk will quickly affect the risk score of downstream nodes (such as application servers) through the message passing mechanism of GNN.

[0117] 3. Output: Output the external risk score each node faces due to dependency relationships, and simulate the chain failure path triggered in the entire cluster after a single node failure.

[0118] Decision and execution layer:

[0119] Risk fusion and early warning generation:

[0120] Set a weighted fusion module to integrate the outputs of each sub-model:

[0121] ; (where α, β, γ are adjustable weight coefficients).

[0122] When the comprehensive risk score exceeds a preset threshold (such as 70 points), a warning event is generated. The warning event explicitly includes: fault type, associated hardware / service, root cause analysis (provided by explainable AI), recommended handling measures, criticality level.

[0123] Automatic decision and repair based on reinforcement learning (RL):

[0124] 1. State space (State): defined as the global health state matrix of the cluster, including the risk score, resource utilization, topology state, etc. of all nodes.

[0125] 2. Action space (Action): defines a series of repair actions, such as: restarting BMC, restarting server, migrating load to standby node, triggering firmware rollback, isolating faulty nodes, etc.

[0126] 3. Reward function (Reward): carefully designed to meet business objectives, for example: successful repair of fault +100, each action executed -10 (encouraging minimal intervention), action failure or leading to expansion of fault scope -50, leading to unavailability of core services -1000.

[0127] 4. Training and inference: initially use historical fault data for offline training, training a deep Q network (DQN) or policy gradient (PG) model. Online, the RL agent selects the optimal action to execute based on the current state, and continuously optimizes the strategy based on the reward after execution, gradually reducing dependence on preset rules, achieving true intelligent self-healing.

[0128] Step 4, remote diagnosis and maintenance:

[0129] Under the support of BMC and IPMI protocols, administrators can perform remote diagnosis and maintenance. In addition, the system also provides:

[0130] Natural language interaction interface: administrators can query cluster status or perform simple operations through text or voice commands (such as "check the health status of node a" and "summarize the abnormalities in the last 24 hours"), and the system returns the results through the NLP engine.

[0131] AR / VR visual operation interface: supports integration with AR glasses or VR headsets, displaying cluster topology, real-time heat map, fault link, hardware health, etc. in a three-dimensional visual manner, greatly improving the intuitiveness and efficiency of fault location and cluster state perception.

[0132] The above embodiments have at least the following technical effects: the application provides a BMC management system and method, system, device and storage medium for large server clusters based on artificial intelligence. Through the centralized management platform integrated with BMC, the system can realize efficient monitoring and management in large-scale clusters, reducing the number of manual interventions. The fault detection and automatic response mechanism built-in the BMC centralized management platform can discover potential problems in advance and handle them, reducing the cluster failure rate. Through the remote diagnosis and maintenance function of the BMC centralized management platform, administrators can operate without entering the data center, saving a lot of time and labor costs.

[0133] Through the description of the above embodiments, those skilled in the art can clearly understand that the method according to the above embodiments can be realized by means of software and the necessary general hardware platform, of course, it can also be realized by hardware, but in many cases the former is a better embodiment.

[0134] The embodiments of the application also provide a fault information prediction device, Figure 6 is a structural block diagram of a fault information prediction device according to an embodiment of the application, as Figure 6 shown, the device comprises:

[0135] A receiving module is configured to receive a fault prediction request, wherein the fault prediction request is used to request to predict fault information of a target device in a target time period after a current time, and the device cluster is used to perform business operations of different business stages of to-be-handled businesses through the plurality of devices.

[0136] A processing module is configured to generate first life information of the target device in the target time period under the influence of target running information, and generate second life information of the target device according to reference running information and business association information, wherein the target running information is used to indicate a business handling situation of the target device in a reference time period before the current time, the reference running information is used to indicate a business handling situation of a reference device in the plurality of devices in the reference time period, the second life information is used to indicate an influence of the reference device on a device service life of the target device in the target time period, and the business association information is used to indicate a business association relationship between the plurality of devices in a current business handling process of the device cluster.

[0137] A prediction module is configured to predict target fault information of the target device in the target time period according to the first life information and the second life information.

[0138] By means of the above device, after receiving the fault prediction request, on the one hand, the first life information under the target operation information is generated according to the target operation information of the target device in the reference time period, and on the other hand, the second life information of the target device is generated according to the reference operation information of the reference device and the service association information, so as to realize the prediction of the remaining service life of the target device in the future time according to the service handling condition of the target device in the historical time, and the prediction of the life influencing factor of the reference device to the target device according to the service handling condition of the reference device in the historical time in the device cluster except the target device, and then the target fault information of the target device is predicted according to the first life information and the second life information, so as to realize the prediction of the fault information of the target device combined with the running state of the device in the device cluster in the service process, thereby ensuring the accuracy and reliability of the target fault information predicted for the target device, and thus the technical problem of low prediction efficiency of the device fault information in the related art can be solved, and the technical effect of improving the prediction efficiency of the device fault information is achieved.

[0139] Optionally, the processing module comprises:

[0140] The first prediction unit is configured to predict a first remaining life of the target device in the target time period according to service response information of the target device in the reference time period, wherein the service response information is used to indicate a service response condition of the target device to the accepted service in the reference time period; and predict a second remaining life of the target device in the target time period according to element calling information of a device element deployed in the target device in the reference time period, wherein the element calling information is used to indicate an operation mode of the device element in the reference time period, and the target operation information comprises the service response information and the element calling information.

[0141] The conversion unit is configured to convert the first remaining life and the second remaining life to obtain a reference remaining life of the target device in the target time period, wherein the first life information comprises the reference remaining life.

[0142] Optionally, the first prediction unit is configured to:

[0143] characterizing a log sequence of a target device in a reference time period to obtain a log feature including a service response log corresponding to a plurality of time points of the target device in the reference time period, wherein the service response information includes a plurality of the service response logs of the target device in the reference time period, the service response log is used to indicate a service execution mode of the target device at the corresponding time point, and the log feature is used to indicate a time sequence dependency between the current service response log and other logs in the log sequence except the current service response log;

[0144] sorting the log features corresponding to the plurality of service response logs in the reference time period in a time sequence order to obtain a log feature sequence;

[0145] inputting the log feature sequence into a target conversion model to obtain a life attenuation parameter corresponding to each service response log output by the target conversion model, wherein the life attenuation parameter is used to represent an influence of the service execution mode indicated by the service response log on the service life of the target device in the target time period, and the target conversion model records a conversion relationship between log sequence features and life attenuation parameters of service response logs in the log sequence;

[0146] summarizing the life attenuation parameters corresponding to the plurality of service response logs to obtain a first life attenuation parameter, wherein the first life attenuation parameter is used to indicate a service life attenuation of the target device in the target time period caused by the service execution mode of the target device in the reference time period;

[0147] using the first life attenuation parameter to convert the first remaining life of the target device.

[0148] Optionally, the first prediction unit is configured to:

[0149] obtaining an element call log of each device element deployed in the target device, wherein the element call log records a running state of the device element at each time point in the reference time period, and the element call information includes the element call log;

[0150] extracting features from the element call log to obtain a time domain call feature and a frequency domain call feature of the device element, wherein the time domain call feature is used to indicate a call time of the target device to the device element in the reference time period, and the frequency domain call feature is used to indicate a call frequency of the target device to the device element at the corresponding call time;

[0151] inputting the time domain calling feature and the frequency domain calling feature into a reference conversion model to obtain an element life attenuation parameter of the device element output by the reference conversion model, the element life attenuation parameter being used to indicate an influence of a calling manner of the target device on the device element in the reference time period on element life of the device element in the target time period;

[0152]

[0153] using the second life attenuation parameter to convert the second remaining life of the target device.

[0154] Optionally, the conversion unit is configured to:

[0155] obtain a first weight parameter of the first remaining life and the second remaining life, wherein the first weight parameter is used to indicate an importance of the first remaining life and the second remaining life to representing the remaining life of the target device in the target time period;

[0156] perform weighted sum calculation on the first remaining life and the second remaining life using the first weight parameter to obtain the reference remaining life.

[0157] Optionally, the processing module comprises:

[0158] a first processing unit configured to perform association processing on the reference running information and the target running information according to a service association relationship between devices indicated by the service association information to obtain a target relationship graph of the device cluster at present, wherein the target relationship graph comprises a plurality of nodes and a connection edge connecting the nodes, the nodes correspond to the running information of the plurality of devices one by one, the nodes comprise the reference running information or the target running information, and the connection edge is used to indicate an association relationship between the running information in the nodes.

[0159] an input unit configured to input the target relationship graph into a candidate conversion model to obtain a life compensation parameter output by the candidate conversion model, wherein the life compensation parameter is used to indicate a compensation of a service handling state of the reference device on the service life of the target device in the target time period, the second life information comprises the life compensation parameter, and the candidate conversion model records a conversion relationship between a relationship graph and a life compensation parameter. ​

[0160] Optionally, the prediction module comprises:

[0161] an acquisition unit, configured to acquire a second weight parameter corresponding to a life compensation parameter of the target device in the target time period, wherein the life compensation parameter is used to indicate a compensation situation of a service handling state of the reference device to a service life of the target device in the target time period, the second life information comprises the life compensation parameter, and the second weight parameter is used to indicate an influence degree of an operation state of the reference device to a service life of the target device;

[0162] a calculation unit, configured to perform weighted calculation on the life compensation parameter by using the second weight parameter to obtain a life compensation value;

[0163] a compensation unit, configured to compensate the reference residual life of the target device in the target time period by using the life compensation value to obtain a target residual life of the target device in the target time period;

[0164] a second prediction unit, configured to predict the target fault information of the target device according to the target residual life.

[0165] Optionally, the second prediction unit is configured to:

[0166] determine a target fault probability corresponding to the target residual life from residual lives and fault probabilities having a corresponding relationship, wherein the target fault information comprises the target fault probability.

[0167] Optionally, the apparatus further comprises:

[0168] an acquisition module, configured to acquire cluster state information of the device cluster in a case where it is determined that the target device is in a fault state, wherein the cluster state information is used to indicate an operation state of each of the plurality of devices included in the device cluster;

[0169] a detection module, configured to detect an operation benefit of each of a plurality of fault maintenance operations in a cluster operation state indicated by the cluster state information, wherein the operation benefit is used to indicate a benefit situation brought by performing the fault maintenance operation on the target device to a running quality of the device cluster in the current cluster operation state;

[0170] a screening unit, configured to screen a target fault maintenance operation meeting a target benefit condition from the plurality of fault maintenance operations in the operation benefits;

[0171] a processing unit, configured to perform the target fault maintenance operation on the target device.

[0172] The features of the embodiments of the prediction device of fault information can be referred to the related descriptions of the embodiments of the prediction method of fault information, which will not be repeated here.

[0173] The embodiments of the present application also provide an electronic device, comprising a memory and a processor, the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any of the above-mentioned embodiments of the prediction method of fault information.

[0174] 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 prediction method of fault information when running.

[0175] In an example embodiment, the above-mentioned computer readable storage medium 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.

[0176] The embodiments of the present application also provide a computer program product, which comprises a computer program, and the computer program is executed by a processor to implement the steps in any of the above-mentioned embodiments of the prediction method of fault information.

[0177] The 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 is executed by a processor to implement the steps in any of the above-mentioned embodiments of the prediction method of fault information.

[0178] The skilled person 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 the two. In order to clearly illustrate the interchangeability of hardware and software, the components and steps of the examples have been described in general terms in the above description. Whether the functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person 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.

[0179] The above describes in detail the method and device for predicting fault information provided by the present application. The principles and implementation manners of the present application are described by using specific examples, and the above description of the embodiments is only applicable to helping understand the method and core idea of the present application. It should be pointed out that, for those skilled in the art, without departing from the principles of the present application, the present application can be improved and modified in several ways, and these improvements and modifications also fall within the protection scope of the claims of the present application.

Claims

1. A method of predicting failure information, characterized by, The method comprises: receiving a failure prediction request, wherein the failure prediction request is used to request failure information of a target device in a target time period after a current time in a plurality of devices included in a device cluster, and the device cluster is used to perform business operations of different business stages of to-be-handled businesses by the plurality of devices; generating first life information of the target device in the target time period under the influence of target operation information according to the target operation information, generating second life information of the target device according to reference operation information and business association information, wherein the target operation information is used to indicate business handling conditions of the target device in a reference time period before the current time, the reference operation information is used to indicate business handling conditions of a reference device in the plurality of devices in the reference time period, the second life information is used to indicate an influence of the reference device on a device service life of the target device in the target time period, and the business association information is used to indicate a business association relationship between the plurality of devices in a current business handling process of the device cluster; predicting target failure information of the target device in the target time period according to the first life information and the second life information; the generating of the first life information of the target device in the target time period under the influence of the target operation information comprises: predicting first residual life of the target device in the target time period according to business response information of the target device in the reference time period, wherein the business response information is used to indicate business response conditions of the target device to the handled business in the reference time period; predicting second residual life of the target device in the target time period according to element calling information of a device element deployed in the target device in the reference time period, wherein the element calling information is used to indicate an operation mode of the device element in the reference time period, the target operation information comprises the business response information and the element calling information; and converting reference residual life of the target device in the target time period by using the first residual life and the second residual life, wherein the first life information comprises the reference residual life.

2. The method of claim 1, wherein the predicting of the first residual life of the target device in the target time period according to the business response information of the target device in the reference time period comprises: characteristics of the log sequence, wherein the service response information comprises a plurality of the service response logs of the target device in the reference time period, the service response log is used to indicate the service execution mode of the target device at the corresponding time, and the log characteristics are used to indicate the time sequence dependence between the current service response log and other logs in the log sequence except the current service response log; sorting the log characteristics corresponding to the plurality of service response logs in the reference time period in a time sequence order to obtain a log characteristic sequence; inputting the log characteristic sequence into a target conversion model to obtain a life attenuation parameter corresponding to each service response log output by the target conversion model, wherein the life attenuation parameter is used to represent the influence of the service execution mode indicated by the service response log on the service life of the target device in the target time period, and the target conversion model records the conversion relationship between the log sequence characteristics and the life attenuation parameters of the service response logs in the log sequence; summarizing the life attenuation parameters corresponding to the plurality of service response logs to obtain a first life attenuation parameter, wherein the first life attenuation parameter is used to indicate the service life attenuation of the target device in the target time period caused by the service execution mode of the target device in the reference time period; using the first life attenuation parameter to convert the first remaining life of the target device.

3. The method of claim 1, wherein the predicting the second remaining life of the target device in the target time period according to the element call information of the device elements deployed in the target device in the reference time period comprises: element call logs of each of the device elements deployed in the target device, wherein the element call log records the running state of the device element at each time in the reference time period, and the element call information comprises the element call log; extracting characteristics from the element call log to obtain time domain call characteristics and frequency domain call characteristics of the device element, wherein the time domain call characteristics are used to indicate the call time of the device element in the target device in the reference time period, and the frequency domain call characteristics are used to indicate the call frequency of the device element in the target device at the corresponding call time; inputting the time domain call characteristics and the frequency domain call characteristics into a reference conversion model to obtain an element life attenuation parameter of the device element output by the reference conversion model, wherein the element life attenuation parameter is used to indicate the influence of the call mode of the device element in the target device in the reference time period on the element life of the device element in the target time period. ​ aggregate the element life attenuation parameters of all the device elements deployed in the target device to obtain a second life attenuation parameter of the target device in the target time period, where the second life attenuation parameter is used to indicate an influence of a calling manner of device elements on the target device in the reference time period on a device life of the target device in the target time period; convert the second remaining life of the target device using the second life attenuation parameter.

4. The method of claim 1, wherein the converting the reference remaining life of the target device in the target time period using the first remaining life and the second remaining life comprises: obtaining a first weight parameter of the first remaining life and the second remaining life, where the first weight parameter is used to indicate an importance of the first remaining life and the second remaining life to representing the remaining life of the target device in the target time period; and performing a weighted sum calculation on the first remaining life and the second remaining life using the first weight parameter to obtain the reference remaining life.

5. The method of claim 1, wherein the generating the second life information of the target device according to the reference running information and the service association information comprises: performing an association processing on the reference running information and the target running information according to a service association relationship between devices indicated by the service association information to obtain a target relationship graph of the device cluster, where the target relationship graph includes a plurality of nodes and a connection edge connecting between the nodes, the plurality of nodes correspond to the running information of the plurality of devices one by one, the nodes include the reference running information or the target running information, and the connection edge is used to indicate an association relationship between the running information in the nodes; and inputting the target relationship graph into a candidate conversion model to obtain a life compensation parameter output by the candidate conversion model, where the life compensation parameter is used to indicate a compensation of a service handling state of the reference device on a service life of the target device in the target time period, the second life information includes the life compensation parameter, and the candidate conversion model records a conversion relationship between a relationship graph and a life compensation parameter.

6. The method of claim 1, wherein the predicting the target fault information of the target device in the target time period according to the first life information and the second life information comprises: obtaining a second weight parameter corresponding to a life compensation parameter of the target device in the target time period, where the life compensation parameter is used to indicate a compensation of a service handling state of the reference device on a service life of the target device in the target time period, the second life information includes the life compensation parameter, the second weight parameter is used to indicate an influence of the running state of the reference device on the device service life of the target device, and the target fault information is predicted according to the first life information, the second life information, and the second weight parameter. ​ ​ ​ weighting the life compensation parameter using the second weight parameter to obtain a life compensation value; compensating a reference residual life of the target device in the target time period using the life compensation value to obtain a target residual life of the target device in the target time period; predicting the target failure information of the target device according to the target residual life.

7. The method of claim 6, wherein the predicting the target failure information of the target device according to the target residual life comprises: determining a target failure probability corresponding to the target residual life from residual life and failure probability having a corresponding relationship, wherein the target failure information comprises the target failure probability.

8. The method of claim 1, wherein, The method further comprises: in a case where it is determined that the target device is in a failure state, obtaining cluster state information of the device cluster, wherein the cluster state information is used to indicate a running state of each of the plurality of devices included in the device cluster; detecting an operation benefit of each of a plurality of failure maintenance operations in a cluster running state indicated by the cluster state information, wherein the operation benefit is used to indicate a benefit to the running quality of the device cluster brought by executing the failure maintenance operation on the target device under the current cluster running state; screening a target failure maintenance operation meeting a target benefit condition from the plurality of failure maintenance operations according to the operation benefit; executing the target failure maintenance operation on the target device.

9. A failure information predicting apparatus characterized by comprising: comprises: a receiving module configured to receive a failure prediction request, wherein the failure prediction request is used to request to predict failure information of a target device in a target time period after a current time, the target device being included in a device cluster, and the device cluster being used to execute business operations of different business stages of to-be-handled businesses by a plurality of devices; a processing module configured to generate first life information of the target device in the target time period under the influence of target running information according to the target running information, and generate second life information of the target device according to reference running information and business association information, wherein the target running information is used to indicate a business handling situation of the target device in a reference time period before the current time, the reference running information is used to indicate a business handling situation of a reference device in the reference time period, the reference device being different from the target device among the plurality of devices, the second life information is used to indicate an influence of the reference device on a device service life of the target device in the target time period, and the business association information is used to indicate a business association relationship between the plurality of devices in a current business handling process of the device cluster; a prediction module configured to predict target failure information of the target device in the target time period according to the first life information and the second life information. The processing module comprises: a first prediction unit configured to predict a first residual life of the target device in the target time period according to service response information of the target device in a reference time period, wherein the service response information is used to indicate a service response condition of the target device to the service in the reference time period; predict a second residual life of the target device in the target time period according to element calling information of a device element deployed in the target device in the reference time period, wherein the element calling information is used to indicate a running mode of the device element in the reference time period, and the target running information comprises the service response information and the element calling information; and a conversion unit configured to convert a reference residual life of the target device in the target time period by using the first residual life and the second residual life, wherein the first life information comprises the reference residual life.

Images