Fault identification method and device

Abstract

The invention discloses a fault identification method and device, and the method comprises the steps: obtaining a plurality of device features of a to-be-identified device, each device feature corresponding to a module, and the module being a hardware module or a software module of the device; according to the correlation of different modules, the fusion weight of other device features is determined, the other device features refer to the device features except the target device feature in the multiple device features, and the target device feature is the device feature of the corresponding target module; fusing the other device features with the target device feature based on the fusion weight to obtain a fused device feature; and determining a fault identification result of a target module in the to-be-identified equipment according to the fusion equipment characteristics.

Description

Technical Field

[0001] The present application relates to the field of fault identification technology, and in particular to a fault identification method and device. Background Art

[0002] In related technologies, in order to identify a fault condition of a module of an electronic device, relevant information of the module is generally collected and processed to obtain a corresponding fault identification result.

[0003] The accuracy of the fault identification results obtained in this way is low. Summary of the Invention

[0004] To this end, this application discloses the following technical solutions:

[0005] A first aspect of the present application provides a fault identification method, comprising:

[0006] Obtain multiple device features of the device to be identified, each of the device features corresponding to a module, and the module is a hardware module or a software module of the device;

[0007] Determine the fusion weight of other device features according to the correlation of different modules, where other device features refer to device features other than the target device features among the multiple device features, and the target device features refer to the device features corresponding to the target module;

[0008] fusing the other device features and the target device features based on the fusion weight to obtain a fused device feature;

[0009] Determine a fault identification result of the target module in the device to be identified according to the fusion device characteristics.

[0010] Optionally, fusing the other device features and the target device features based on the fusion weight to obtain a fused device feature includes:

[0011] fusing the other device features that meet the first condition and the target device features based on the fusion weight to obtain a fused device feature;

[0012] The first condition includes:

[0013] The failure frequency of modules corresponding to other device characteristics is greater than the failure frequency of the target module.

[0014] Optionally, determining the fusion weights of other device features based on the correlations between different modules includes:

[0015] Obtaining mutual exclusion strength data reflecting the correlation between the target module and other modules, where the other modules refer to modules corresponding to the other device features;

[0016] Determine the fusion weight of the other device features according to the target device features, the other device features and the mutual exclusion strength data.

[0017] Optionally, the method for determining the mutual exclusion strength data between the target module and other modules includes:

[0018] Obtaining module knowledge data of a target module and module knowledge data of other modules, wherein the module knowledge data of the target module includes sample device features corresponding to the target module, and the module knowledge data of other modules includes sample device features corresponding to other modules;

[0019] Determining a first boundary formed by the module knowledge data of the target module and a second boundary formed by the module knowledge data of other modules in the multidimensional space to which the module knowledge data belongs;

[0020] Based on the overlap degree of the first boundary and the second boundary, mutual exclusion strength data of the target module and other modules is determined.

[0021] Optionally, also include:

[0022] The mutual exclusion strength data of the target module and the other modules are updated according to the target device characteristics and the other device characteristics.

[0023] Optionally, determining the fusion weight of other device features according to the target device feature, the other device features, and the mutual exclusion strength data includes:

[0024] Calculating the target device features, the other device features, and the mutual exclusion strength data of the target module and the other modules according to the gating parameters corresponding to the target module to obtain the fusion weights of the other device features;

[0025] The gating parameters corresponding to the target module are determined according to the characteristics of the sample device.

[0026] Optionally, the fault identification result of the target module includes at least a fault probability value of the target module in the device to be identified. If the fault probability value is greater than a target threshold, it indicates that there is a fault in the target module of the device to be identified. If the fault probability value is less than or equal to the target threshold, it indicates that there is no fault in the target module of the device to be identified.

[0027] Optionally, obtaining multiple device features of the device to be identified includes:

[0028] Processing the device information of the device to be identified according to the preprocessing model in the pre-built data processing model to obtain preprocessing features;

[0029] The pre-processing features are processed respectively according to a plurality of adapters of the data processing model to obtain device features output by each adapter, and each adapter corresponds to a module.

[0030] Optionally, also include:

[0031] Processing sample device features according to the plurality of adapters to obtain a fault identification result of the sample device features;

[0032] The actual fault condition corresponding to the sample device feature is compared with the fault identification result of the sample device feature, so as to update the parameters contained in the plurality of adapters based on the comparison result.

[0033] A second aspect of the present application provides an electronic device, including a memory and a processor;

[0034] The memory is used to store computer programs;

[0035] The processor is configured to execute the computer program to:

[0036] Obtain multiple device features of the device to be identified, each of the device features corresponding to a module, and the module is a hardware module or a software module of the device;

[0037] Determine the fusion weight of other device features according to the correlation of different modules, where other device features refer to device features other than the target device features among the multiple device features, and the target device features refer to the device features corresponding to the target module;

[0038] fusing the other device features and the target device features based on the fusion weight to obtain a fused device feature;

[0039] Determine a fault identification result of the target module in the device to be identified according to the fusion device characteristics. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are merely embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without any creative work.

[0041] Figure 1 This is a flowchart of a fault identification method provided by an embodiment of the present application;

[0042] Figure 2 This is a structural diagram of a data processing model provided in an embodiment of the present application;

[0043] Figure 3 This is a structural diagram of another data processing model provided in an embodiment of the present application;

[0044] Figure 4 This is a flow chart of a method for determining adapter weight parameters provided by an embodiment of the present application;

[0045] Figure 5 This is a structural diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0047] This embodiment provides a fault identification method, see Figure 1 , is a flowchart of the method, which may include the following steps.

[0048] S101, obtaining multiple device features of a device to be identified, each device feature corresponds to a module, and the module is a hardware module or a software module of the device.

[0049] The fault identification method of this embodiment can be executed by any electronic device with corresponding data processing capabilities. The electronic device can be a terminal device for personal use, such as a desktop computer, a laptop computer, or a server-type device.

[0050] The method of this embodiment can be repeatedly executed periodically or irregularly for the device to be identified. For example, multiple device features of the device to be identified can be obtained at regular intervals to determine the fault identification result of the current device to be identified. Alternatively, the method of this embodiment can be used to determine the fault identification result of the device to be identified when an abnormal phenomenon occurs in the device to be identified, such as an abnormal sound or incorrect output data.

[0051] The device to be identified can be any device that needs to be identified as to whether a fault exists and to identify specific fault information.

[0052] As some examples, the device to be identified can be any server device in the server cluster of an enterprise user. By applying the method of this embodiment to these server devices, the current fault identification results of these server devices can be obtained, so that each server device in the server cluster can be inspected and maintained based on the fault identification results.

[0053] The device characteristics of the device to be identified may include device characteristics of a hardware module corresponding to the device to be identified, and device characteristics of a software module corresponding to the device to be identified.

[0054] Hardware modules refer to the physical components of the device being identified, such as the motherboard, central processing unit (CPU), graphics processing unit (GPU), power supply, hard drive, and cooling system. The device features corresponding to hardware modules include, but are not limited to, those corresponding to the motherboard, CPU, and power supply.

[0055] Software modules refer to the different computer programs running on the hardware modules of the device to be identified. For example, the operating system can be considered a software module, each application can be considered a software module, and the database program used for centralized data storage can also be considered a software module. If the device to be identified is a server, the application can be a containerized application, and the database can be a distributed database. The device features corresponding to the software modules include, but are not limited to, the device features corresponding to the operating system, the device features corresponding to the database program, the device features corresponding to a containerized application A, the device features corresponding to a containerized application B, etc.

[0056] The device feature corresponding to a module can be a feature vector that reflects the module's current operating and operational status. Device features can be generated by a related data processing model, and the dimensionality of the device feature can be related to the model structure of the data processing model. The model structure can be designed as needed and is not limited.

[0057] In S101, the specific module device characteristics to be obtained can be set as needed and are not limited. For example, the obtained device characteristics can be the device characteristics of modules that have experienced failures in the recent period, or can be the device characteristics of several modules newly added to the device to be identified in the recent period.

[0058] S102, determining the fusion weight of other device features according to the correlation of different modules, where the other device features refer to device features other than the target device features among the multiple device features, and the target device features are device features corresponding to the target module.

[0059] In this embodiment, the correlation between any two different modules can represent the degree of mutual influence between the two modules during the operation of the device to be identified. For example, if there is a strong correlation between module 1 and module 2, it means that the degree of mutual influence between the two is high. If there is an abnormality or fault in module 1, there is a high probability that module 2 will also have an abnormality or fault, and there is a high probability that there is a causal relationship between the two. If there is a weak correlation between module 1 and module 2, it means that the degree of mutual influence between the two is low. If there is an abnormality or fault in module 1, there is no guarantee that there will be an abnormality or fault in module 2. Even if both modules have abnormalities or faults at the same time, there may not be a causal relationship between them.

[0060] As some examples, when the cooling system fails and cannot dissipate heat normally, the CPU may crash due to overheating, so there may be a strong correlation between the cooling system and the CPU; when the hard disk is damaged, the distributed database that mainly stores data on the hard disk may experience data loss or errors, so there may be a strong correlation between the hard disk and the distributed database; when the CPU fails, the hard disk can still store data normally, so there may be a weaker correlation between the CPU and the hard disk.

[0061] In S102, relevant data can be obtained first and analyzed to determine the correlation between the target module and other modules, and then the above-mentioned fusion weight can be determined based on the correlation; alternatively, the correlation between each two modules can be predetermined before executing the method of this embodiment, and when executing S102, the correlation between the target module and other modules can be directly read to determine the fusion weight according to this step.

[0062] The target module may be any module among the multiple modules corresponding to the multiple device features in S101 , and the other modules refer to the other modules among the multiple modules except the target module.

[0063] In some embodiments, each module may be used as a target module, and the process from S102 to S104 may be executed for each module to obtain a fault identification result for each module.

[0064] For example, assuming that S101 obtains 10 device features corresponding to modules 1 to 10, when executing this method, module 1 can be considered as the target module, and the fault identification result of module 1 can be obtained according to the process of S102 to S104. Module 2 can be considered as the target module, and the fault identification result of module 2 can be obtained according to the process of S102 to S104. Similarly, the fault identification results of modules 3 to 10 can be obtained. When module 1 is considered as the target module, modules 2 to 10 are equivalent to other modules. When module 2 is considered as the target module, module 1 and modules 3 to 10 are equivalent to other modules, and so on.

[0065] The device features corresponding to other modules are referred to as other device features. In S102, for any other device feature, a fusion weight corresponding to the other device feature can be determined based on the correlation between the other module corresponding to the other device feature and the target module. For example, if the target module is module 1 and the other module is module 2, the fusion weight for device feature 2 corresponding to module 2 can be determined based on the correlation between modules 1 and 2.

[0066] S103: Fusing other device features and the target device feature based on the fusion weight to obtain a fused device feature.

[0067] In S103, the target device features and all other device features can be fused to obtain fused device features, or the target device features and part of other device features can be fused to obtain fused device features. For the latter case, in S102, only the fusion weights of the other device features used for fusion can be obtained, and the rest that are not fused do not need to determine the fusion weights, that is, it is possible to first determine which other device features need to be fused, and then determine the fusion weights of these other device features that need to be fused.

[0068] All the above-mentioned other device features refer to all other device features among the multiple device features obtained in S101 except the target device feature. Combined with the above example, if the target device feature is device feature 1 corresponding to module 1, then all the other device features can refer to device features 2 to 10 corresponding to modules 2 to 10.

[0069] There are many ways to merge, which are not limited in this embodiment.

[0070] As some examples, the target device feature and each other device feature used for fusion can be weighted and summed based on the fusion weight to obtain the fused device feature. This fusion method can be expressed as the following formula (1).

[0071]

[0072] Among them, K S represents the fusion device characteristics, K t Indicates the target device characteristics, K i represents the i-th other device feature, G ti Indicates the fusion weight corresponding to the i-th other device feature when the target device feature is the t-th device feature.

[0073] As another example, the weighted sum of the other device features used for fusion can be first performed according to the fusion weight to obtain the spliced ​​device feature, and then the spliced ​​device feature and the target device feature can be spliced ​​together to obtain the fused device feature. This fusion method can be expressed by the following formula (2).

[0074]

[0075] K F Indicates the characteristics of the splicing equipment. The meanings of other symbols are the same as above.

[0076] S104: Determine a fault identification result of a target module in the device to be identified based on the fused device characteristics.

[0077] In S104, various data processing technologies can be used to process the fused device features to obtain a fault identification result of the target module in the device to be identified. The data processing technologies used can be referred to in related technologies and are not limited thereto.

[0078] As an example, the fused device features can be input into a pre-built neural network capable of processing the fused device features. This neural network can then be used to process the fused device features and obtain fault identification results for the target module. The neural network used can be a neural network built based on a self-attention mechanism or other mechanisms used in related technologies.

[0079] Optionally, for different target modules, the neural networks used in S104 may be different. The difference here may be reflected in that the network structures of the neural networks used are the same, but the values ​​of the parameters contained in each are different, such as the values ​​of the weights contained in the neural networks are different.

[0080] In combination with the above example, when module 1 is used as the target module, the fusion device characteristics corresponding to module 1 are obtained according to the above method, and then the fusion device characteristics are processed by neural network 1 corresponding to module 1 to obtain the fault identification result of module 1; when module 2 is used as the target module, the fusion device characteristics corresponding to module 2 are obtained according to the above method, and then the fusion device characteristics are processed by neural network 2 corresponding to module 2 to obtain the fault identification result of module 2. Neural network 1 and neural network 2 have the same network structure, but the values ​​of the network weights contained in each are different.

[0081] The fault identification result of the target module may include several information related to the current fault status of the target module.

[0082] In some embodiments, the fault identification result of the target module includes at least a fault probability value of the target module in the device to be identified. A fault probability value greater than a target threshold indicates that a fault exists in the target module of the device to be identified, and a fault probability value less than or equal to the target threshold indicates that a fault does not exist in the target module of the device to be identified.

[0083] The failure probability value can be a percentage between 0 and 100%, and the target threshold can be a preset value between 0 and 100%. The specific value can be set as needed and is not limited. The target thresholds corresponding to different target modules can be the same or different.

[0084] As an example of different target thresholds, the target threshold corresponding to module 1 can be 70%, and the target threshold corresponding to module 2 can be 75%; when the fault identification result corresponding to module 1 is obtained according to the aforementioned method, if the fault probability value of module 1 is greater than 70%, it can be considered that module 1 of the device to be identified is faulty; when the fault identification result corresponding to module 2 is obtained according to the aforementioned method, if the fault probability value of module 2 is greater than 75%, it can be considered that module 2 of the device to be identified is faulty; if the fault probability value of module 2 is less than or equal to 75%, it can be considered that module 2 of the device to be identified is not faulty.

[0085] Furthermore, if the fault identification result indicates that there is a fault in the target module of the device to be identified, the fault identification result can be output through a variety of output methods to prompt relevant personnel to carry out timely inspection and maintenance, such as voice broadcast, display on a specific terminal device, output of vibration signals or prompt sound signals through wearable devices, etc.

[0086] For further example, the fault identification result of the target module may also include any one or more of the following information: the fault type, fault cause, fault occurrence time, fault severity, etc. of the fault that may currently exist in the target module.

[0087] The beneficial effects of this embodiment are:

[0088] Within a device, a module failure often involves the interaction of multiple factors. For example, high temperatures can cause certain hardware modules to fail or age, leading to the crash of some applications. Related technologies only analyze information related to specific modules and are unable to analyze the interactions between different factors, resulting in low accuracy.

[0089] The fault identification method of this embodiment combines other device features of non-target modules with target device features corresponding to target modules based on the correlation between different modules, thereby analyzing the correlation between different factors during fault identification, which helps to obtain more accurate fault identification results.

[0090] Optionally, a method for obtaining multiple device features of a device to be identified may be:

[0091] Processing the device information of the device to be identified according to the preprocessing model in the pre-built data processing model to obtain preprocessing features;

[0092] The pre-processing features are processed respectively according to a plurality of adapters of the data processing model to obtain the device features output by each adapter, and each adapter corresponds to a module.

[0093] The data processing model used in this embodiment may include a preprocessing model and multiple adapters. The number of adapters is not limited and can be consistent with the number of device features to be obtained.

[0094] As an example, see Figure 2 , is a schematic diagram of the structure of an exemplary data processing model. The data processing model includes three adapters, each of which corresponds to a module of the device to be identified and can output a device feature corresponding to the module.

[0095] When obtaining current device characteristics, we can first collect device information about the device to be identified within the recent period up to the current moment. This device information can include any one or more of the following: device status information, device operation log information, and device load information. It can also include other information related to the device to be identified. Device status information can include parameters such as the temperature, power, and power consumption of each component of the device to be identified, various device status-related data obtained from registers, memory, and hard disks, and other data that can reflect the device status.

[0096] Then, the preprocessing model can be used to process the device information of the device to be identified to obtain the preprocessing features of the device to be identified. The preprocessing features can be a set of feature data that can reflect the overall device status of the device to be identified. Formally, the preprocessing features can be a feature vector, a set of feature vectors or a feature matrix, without limitation.

[0097] Each adapter of the data processing model can obtain the preprocessing features output by the preprocessing model, obtain information about the module corresponding to the adapter from the overall state of the device to be identified reflected by the preprocessing features, and then output the device features corresponding to the module.

[0098] by Figure 2 For example, adapters 1, 2, and 3 process the preprocessing features to obtain the device features of module 1 corresponding to adapter 1, the device features of module 2 corresponding to adapter 2, and the device features of module 3 corresponding to adapter 3. Module 1 can be any hardware module or software module listed in the above examples, and the same applies to modules 2 and 3.

[0099] Optionally, as described in the above embodiment, when obtaining the fused device features, only a portion of the other device features may be fused with the target device features. One fusion method may be:

[0100] Based on the fusion weight, other device features that meet the first condition are fused with the target device feature to obtain a fused device feature;

[0101] The first condition includes:

[0102] The failure frequency of the modules corresponding to other device characteristics is greater than the failure frequency of the target module.

[0103] The first condition here can also be understood as follows: if the failure frequency of another module is greater than the failure frequency of the target module, then the device characteristics corresponding to this other module are other device characteristics that meet the first condition; if the failure frequency of another module is less than or equal to the failure frequency of the target module, then the device characteristics corresponding to this other module are other device characteristics that do not meet the first condition.

[0104] In order to determine the failure frequency of each module, fault information of the device to be identified and / or other devices of the same type as the device to be identified in the past period of time (such as the past week or the past month) can be collected through manual monitoring or other methods, and the number of failures of each module in the device in the past period of time can be counted. The more times a module fails, the higher the failure frequency of the module, and the fewer times a module fails, the lower the failure frequency of the module.

[0105] According to the above-mentioned fusion method, after obtaining multiple device features, the failure frequencies of the modules corresponding to these device features can be compared to determine which other device features meet the first condition relative to the target device feature. Then, only the fusion weights of the other device features that meet the first condition are determined, and then the other device features that meet the first condition and the target device features are fused according to the fusion weights to obtain the fused device features.

[0106] Among them, for a certain target module, if the failure frequency of each other module is less than or equal to that of the target module, then when identifying the failure of the target module, it is not necessary to fuse it according to S102 and S103. The target device feature corresponding to this target module can be directly used as the fused device feature, and identification is performed according to S104 to obtain the failure identification result of the target module.

[0107] Optional, if the application has Figure 2 The data processing model of the structure shown is used to implement the method of this embodiment. The steps of determining the fusion weight and obtaining the fusion device characteristics can be performed by the fusion module contained in the data processing model. Moreover, the method of fusing other device characteristics that meet the first condition in the above embodiment can be determined by the order of the adapters in the data processing model and the connection relationship between the fusion module and the adapter.

[0108] As described above, each adapter of the data processing model corresponds to a module of the device to be identified. In this embodiment, these adapters can be sorted according to the failure frequency of the corresponding modules. The adapter corresponding to the module with the highest failure frequency is ranked first and numbered as adapter 1. The adapter corresponding to the module with the second highest failure frequency is ranked second and numbered as adapter 2, and so on.

[0109] Each adapter in the data processing model except adapter 1 can be configured with a corresponding fusion module. Since the failure frequency of the module corresponding to adapter 1 is greater than the failure frequency of each other module, the device characteristics output by adapter 1 can be directly used as fusion device characteristics for fault identification without fusion processing. Therefore, adapter 1 does not need to be configured with a corresponding fusion module.

[0110] Each adapter is connected to its own corresponding fusion module and the fusion modules corresponding to other adapters in the following order. In this way, the device characteristics output by an adapter can be input into its own corresponding fusion module and the fusion modules corresponding to other subsequent adapters.

[0111] by Figure 2 For example, the device feature 1 output by adapter 1 is input into the subsequent fusion module 2 and fusion module 3 respectively, and the device feature 2 output by adapter 2 is input into its corresponding fusion module 2 and outputs the subsequent fusion module 3.

[0112] According to the above structure, a fusion module can obtain the device characteristics output by the current adapter and the device characteristics output by other adapters that are ranked before the current adapter, and the failure frequency of the modules corresponding to the other adapters is higher than the failure frequency of the modules corresponding to the current adapter. When the fusion module is working, it can use the device characteristics output by the current adapter as the target device characteristics and the device characteristics output by other adapters that are ranked before the current adapter as other device characteristics that meet the first condition, thereby fusing the target device characteristics and other device characteristics that meet the first condition according to the aforementioned method to obtain the fused device characteristics. The current adapter here refers to the adapter corresponding to the fusion module.

[0113] by Figure 2 For example, the fusion module 3 can take device feature 3 as the target device feature, and device feature 1 and device feature 2 as other device features that meet the first condition, and fuse device feature 3 with device feature 1 and device feature 2 in the aforementioned manner to obtain a fused device feature.

[0114] The beneficial effects of this embodiment are:

[0115] First, compared to using all other device features for fusion, selecting other device features that meet the first condition for fusion can reduce the amount of data required to process in determining fusion weights and fusion processing, thereby improving the processing efficiency of the method of this embodiment to a certain extent.

[0116] Secondly, for example, if a failure in module 1 is likely to cause a failure in module 2, then the failure frequency of module 1 should be higher than that of module 2. Conversely, if the failure frequency of module 3 is lower than that of module 2, then the failure of module 2 is unlikely to be the cause of the failure of module 3. In other words, module 1 is more likely to cause a failure in module 2, and module 3 is less likely to cause a failure in the target module.

[0117] Based on the above principles, when identifying the fault of the target module, this embodiment fuses the other device features of modules that are more likely to induce the fault of the target module with the target device features through the fault frequency of each module, and excludes the other device features of modules that are less likely to induce the fault of the target module. The fused device features thus obtained can include information about the target module itself and information about other modules that may become inducing factors; therefore, by fusing in the above manner, it is possible to reduce the amount of data while more comprehensively retaining the information required for fault identification in the fused device features, so that the obtained fault identification results have higher accuracy.

[0118] Optionally, based on the relevance of different modules, the fusion weights of other device features may be determined by:

[0119] Obtaining mutual exclusion strength data reflecting the correlation between the target module and other modules, where the other modules refer to modules corresponding to other device features;

[0120] The fusion weights of other device features are determined based on the target device features, other device features, and mutual exclusion strength data.

[0121] For each other module, the mutual exclusion strength data between the other module and the target module can be obtained. The mutual exclusion strength data is negatively correlated with the correlation between the two modules, that is, the larger the value of the mutual exclusion strength data of the two modules, the weaker the correlation between the two modules, and the smaller the value of the mutual exclusion strength data, the stronger the correlation between the two modules.

[0122] In this embodiment, the mutual exclusion strength data of the target module and other modules can be calculated only when the fusion weight is determined, or the mutual exclusion strength data between every two modules in the device to be identified can be calculated in advance, and the corresponding mutual exclusion strength data can be directly read when determining the fusion weight.

[0123] In the case of pre-calculation, the calculated mutual exclusion strength data between each two modules can be stored in the form of a graph. For example, a graph data structure called a conflict graph can be constructed, in which each node represents a module of the device to be identified, and the mutual exclusion strength data between each two modules is stored as the length of the edge between the two nodes.

[0124] For one other device feature, after obtaining the mutual exclusion strength data of the target module and its corresponding other module, the mutual exclusion strength data, the other device feature and the target device feature may be calculated to obtain a fusion weight of the other device feature.

[0125] Optionally, the calculated fusion weight can be a real number with a value range between greater than or equal to 0 and less than or equal to 1. The larger the fusion weight, the stronger the correlation between other modules corresponding to other device features and the target module. The smaller the fusion weight, the weaker the correlation between other modules corresponding to other device features and the target module.

[0126] Optionally, in order to calculate the fusion weight of other device features, multiple sets of gating parameters can be pre-configured, and each gating parameter corresponds to a module of the device to be identified. On this basis, after determining the target module whose fault needs to be identified, the gating parameters corresponding to the target module can be obtained. Then, according to the gating parameters corresponding to the target module, the target device features, other device features, and the mutual exclusion strength data of the target module and other modules are calculated to obtain the fusion weight of other device features.

[0127] Among them, the gating parameters corresponding to the target module are determined according to the characteristics of the sample device.

[0128] The calculation process can be expressed by the following formula (3).

[0129]

[0130] Sigmoid is a type of activation function. Its expression can be found in related technologies. In some embodiments, Sigmoid can also be replaced by other commonly used activation functions without limitation.

[0131] G ti Indicates the fusion weight corresponding to the i-th other device feature when the target device feature is the t-th device feature, W t Represents the predetermined gating parameter corresponding to the t-th module, which can be a parameter matrix, K t Indicates the target device characteristics, K i represents the i-th other device feature, C ti Represents the mutual exclusion strength data between the target module and the i-th other module.

[0132] Optionally, when the method of this embodiment is based on Figure 2 When the data processing model shown is implemented, the gating parameters of module t can be regarded as at least a part of the weight parameters contained in the fusion module corresponding to module t in the data processing model.

[0133] For example, Figure 2 The fusion module 2 corresponds to module 2 and is used to fuse the device features corresponding to module 2 as target device features. The module gating parameter W2 corresponding to module 2 can be at least a part of the weight parameter contained in the fusion module 2.

[0134] Combine Figure 2 For example, if module 3 corresponding to adapter 3 is regarded as the target module, and device feature 3 corresponding to module 3 is equivalent to the target device feature, then module 1 corresponding to adapter 1 is equivalent to another module, and device feature 1 corresponding to module 1 is equivalent to the first other device feature. When calculating the fusion weight of device feature 1, the gating parameter W3 of module 3 and the mutual exclusion strength data C of module 1 and module 3 can be used. 31 , device feature 3 (i.e. K3) and device feature 1 (i.e. K1) are substituted into formula (3) to calculate the fusion weight of device feature 1. Similarly, the fusion weight of device feature 2 can be calculated. Based on these fusion weights, fusion module 3 can fuse device feature 1, device feature 2 and device feature 3 to obtain the fusion device feature corresponding to module 3.

[0135] Optionally, the method for determining the mutual exclusion strength data of the target module and other modules may be:

[0136] Obtaining module knowledge data of a target module and module knowledge data of other modules, wherein the module knowledge data of the target module includes sample device features corresponding to the target module, and the module knowledge data of other modules includes sample device features corresponding to other modules;

[0137] Determining a first boundary formed by the module knowledge data of the target module and a second boundary formed by the module knowledge data of other modules in the multidimensional space to which the module knowledge data belongs;

[0138] Based on the degree of overlap between the first boundary and the second boundary, mutual exclusion strength data of the target module and other modules are determined.

[0139] The sample device characteristics corresponding to the target module can be obtained by:

[0140] For a device (referred to as a sample device) whose actual fault condition has been confirmed at a certain moment through other means (such as manual inspection), the device information of the sample device at that moment is obtained. This device information is then processed to obtain a feature vector that reflects the operating and operational status of the target module in the sample device at that moment. This vector serves as a sample device feature corresponding to the target module. By repeating this method for several different sample devices, multiple sample device features corresponding to the target module can be obtained. The collection of these sample device features is equivalent to the module knowledge data of the target module.

[0141] Similarly, for a sample device, we can obtain its device information at that moment. This device information is then processed to obtain a feature vector reflecting the operating and operational status of other modules within the sample device at that moment. This feature vector serves as a sample device feature corresponding to the other module. Repeating this method for several different sample devices yields multiple sample device features corresponding to the other modules. The collection of these sample device features is equivalent to the module knowledge data for the other modules.

[0142] The method of processing the device information to obtain the sample device characteristics may be the same as the method of obtaining the device characteristics of the device to be identified in the aforementioned embodiment, and will not be described in detail.

[0143] As previously explained, each device feature can be a feature vector with a certain dimension (let's call it N), which can be 10, 25, or another value. To determine the first boundary, we can use a Gaussian distribution to fit the distribution of each sample device feature contained in the target module's module knowledge data in N-dimensional space. This yields a set of Gaussian distribution parameters that describe the distribution of these sample device features corresponding to the target module. This set of Gaussian distribution parameters is equivalent to the first boundary formed by the target module's module knowledge data in N-dimensional space.

[0144] Similarly, the distribution of each sample device feature contained in the module knowledge data of other modules in the N-dimensional space can be fitted according to the Gaussian distribution to obtain a set of Gaussian distribution parameters that describe the distribution of the sample device features corresponding to these other modules. This set of Gaussian distribution parameters is equivalent to the second boundary formed by the module knowledge data of other modules in the N-dimensional space.

[0145] The method of fitting multiple N-dimensional feature vectors in an N-dimensional space according to Gaussian distribution can be found in related technologies and will not be described in detail.

[0146] In the above method of determining the first boundary and the second boundary, the Gaussian distribution is only used as an example. In other embodiments, the Gaussian distribution can also be replaced by other common probability distributions in related technologies, not limited to the Gaussian distribution.

[0147] After obtaining the first and second boundaries, the divergence between them can be calculated, and based on the result, the mutual exclusion strength data between the target module and the other modules can be determined. The divergence calculated here can be the JS divergence (i.e., Jensen-Shannon Divergence), the KL divergence (i.e., Kullback-Leibler Divergence), or other divergences, without limitation.

[0148] Taking the calculation of JS divergence as an example, the mutual exclusion strength data C of the target module t and other modules i ti The calculation process can be expressed by the following formula (4).

[0149] C ti =1-JensenShannon(P t ||P i ), (4)

[0150] Among them, P t The first boundary calculated based on the module knowledge data of the target module t, P i Represents the second boundary calculated based on the module knowledge data of the i-th other module. JensenShannon (P t ||P i ) indicates the calculation of the JS divergence between two boundaries. The calculation method can be found in relevant technical materials.

[0151] The mutual exclusion strength data C calculated by formula (4) ti The smaller its value is, the more similar the module knowledge data of the target module t is to the module knowledge data of the i-th other module is, and the boundaries between the two are highly overlapped. If they are used for fault identification at the same time, there may be potential identification conflicts. The larger its value is, the clearer the boundary between the module knowledge data of the target module t and the module knowledge data of the i-th other module is, and the mutual exclusivity of the two module knowledge data is strong, and there is no conflict between them.

[0152] In the above embodiments, in order to improve the accuracy of the results, the sample devices may be limited to devices of the same type as the device to be identified. For example, if the device to be identified is a server device in a server cluster, the sample devices may be other server devices in the server cluster.

[0153] In some optional embodiments, when calculating the fusion weights of other device features, the module knowledge data corresponding to the i-th other module can also be added to Ki, that is, Ki can be regarded as a set consisting of the i-th other device feature and the module knowledge data corresponding to the i-th other module. Similarly, the module knowledge data corresponding to the target module t can be added to Kt, that is, Kt can be regarded as a set consisting of the target device feature and the module knowledge data corresponding to the target module t.

[0154] In some optional embodiments, after obtaining the fault identification result of the target module, the following steps may be performed to dynamically update the mutual exclusion strength data:

[0155] Updates the mutual exclusion strength data of the target module and other modules based on the target device characteristics and other device characteristics.

[0156] The module knowledge data of each module collected when determining the mutual exclusion strength data can be stored in a knowledge base respectively. When executing the above steps, the module knowledge data corresponding to the target module can be taken out from the knowledge base, and the target device characteristics can be added to the module knowledge data of the target module. The module knowledge data corresponding to other modules can be taken out from the knowledge base, and the other device characteristics can be added to the module knowledge data of other modules. Then, according to the aforementioned method for calculating the mutual exclusion strength data, based on the module knowledge data with the target device characteristics added and the module knowledge data with the other device characteristics added, the mutual exclusion strength data between the target module and other modules can be recalculated.

[0157] In this way, during the period of identifying a device fault, more accurate mutual exclusion strength data can be obtained by continuously updating the module knowledge data of each module in the knowledge base, thereby facilitating obtaining more accurate fault identification results in subsequent fault identification.

[0158] The following combination Figure 2 The structure of the data processing model shown illustrates an application scenario of the recognition method of this embodiment.

[0159] For a device to be identified whose fault needs to be identified, the current device information of the device to be identified is obtained, and the device information is input into the preprocessing model for processing to obtain preprocessing features;

[0160] If the modules corresponding to adapters 1 to 3 all need to perform fault identification, then the three adapters can work simultaneously. Each adapter processes the input preprocessing features to obtain the device features of the corresponding module. That is, adapter 1 outputs device feature 1 of module 1, adapter 2 outputs device feature 2 of module 2, and adapter 3 outputs device feature 3 of module 3.

[0161] Since device feature 1 does not need to be fused with other device features, it is directly input into the neural network 1 corresponding to module 1 as fused device feature 1. After being processed by the neural network 1, the fault identification result 1 of module 1 is obtained;

[0162] Device feature 2 and device feature 1 are both input into the fusion module 2 corresponding to module 2. Fusion module 2 takes device feature 2 as the target device feature and input device feature 1 as the other device feature. Figure 1The corresponding method determines the fusion weight, fuses the target device feature and other device features based on the fusion weight, obtains the fusion device feature 2 corresponding to module 2, and processes the fusion device feature 2 through the neural network 2 corresponding to module 2 to obtain the fault identification result 2 of module 2;

[0163] Device features 1 to 3 are input into the fusion module 3 corresponding to module 3. Fusion module 3 takes device feature 3 as the target device feature and input device features 1 and 2 as other device features. Figure 1 The corresponding method determines the fusion weights of device features 1 and 2 respectively, fuses the target device features and other device features based on the fusion weights, obtains the fused device feature 3 corresponding to module 3, and processes the fused device feature 3 through the neural network 3 corresponding to module 3 to obtain the fault identification result 3 of module 3.

[0164] In some embodiments, if only fault identification results of a few modules in the device to be identified need to be obtained and all components in the data processing model are not needed, the corresponding components can be disabled to reduce resource consumption of applying the data processing model.

[0165] Still combined Figure 2 For example, if only the fault identification result 2 of module 3 needs to be obtained, then adapters 1 to 3 can work normally and output the device features 1 to 3 required by fusion module 3. The subsequent neural networks 1 and 2, as well as fusion module 2 can be disabled, leaving only fusion module 3 and neural network 3 working to output the fault identification result 3 corresponding to module 3.

[0166] If only the fault identification results of module 1 and module 2 need to be obtained, then adapter 3, fusion module 3 and neural network 3 can be disabled, adapter 1 and neural network 1 work to obtain fault identification result 1, and adapter 2, fusion module 2 and neural network 2 work to obtain fault identification result 2.

[0167] The specific modules whose faults need to be identified can be selected by the relevant users based on actual needs and are not limited.

[0168] Optionally, if several new hardware modules or software modules that need to be identified for faults are added to the device to be identified, adapters, fusion modules and neural networks for identifying the faults of these newly added modules can be added to the fault identification model used in this embodiment, and the adapters corresponding to the newly added modules can be arranged after the original adapters. In this way, when identifying the faults of the newly added modules, the knowledge about identifying the faults of the original modules previously learned by the data processing model can be fully utilized by fusing the features of multiple devices to improve the accuracy of identifying the faults of the newly added modules.

[0169] Combine Figure 2For example, suppose the original modules 1 to 3 are the CPU, hard disk and operating system of the device to be identified. Now it is necessary to add a neural network processor NPU to the device to be identified, and it is necessary to use Figure 2 The model identifies the fault of NPU. At this time, NPU can be used as the module 4 that needs to identify the fault. Figure 2 In the data processing model, below the adapter 3, a new adapter 4 corresponding to the module 4 is added, and after the adapter 4, a subsequent fusion module 4 and a neural network 4 are added to form the following Figure 3 In the data processing model shown, the newly added fusion module 4 can obtain device features 1 to 4 output by adapters 1 to 4, take device feature 4 as the target device feature, take device features 1 to 3 as other device features, and fuse the target device feature and other device features according to the aforementioned method to obtain the fused device feature corresponding to module 4. Then, the neural network 4 can process the fused device feature 4 of module 4 to obtain the fault identification result of module 4.

[0170] Alternatively, if an original module of a device to be identified did not need to be identified for faults before, but as the module ages or is iteratively updated, it is now necessary to identify the faults of the module, the corresponding adapter, fusion module and neural network can be added to the data processing model according to the above method to identify the faults of the module.

[0171] For example, the memory stick of the device to be identified originally does not require fault identification, but after long-term use, it is prone to faults due to aging, so now it is necessary to identify the fault of the memory stick. Therefore, the memory stick can be regarded as module 5, and the corresponding adapter 5 is added below the adapter 4 of the data processing model according to the above method, and the module 5 and the neural network 5 are integrated to identify the fault of the module 5.

[0172] It can be seen from the above embodiments that the application Figure 2 or Figure 3 The data processing model shown has the following advantages for fault identification:

[0173] On the one hand, when new modules require fault identification, by adding new adapters, fusion modules, and neural networks, the data processing model can quickly acquire the ability to identify faults in the newly added modules without changing the existing weight parameters of each component in the data processing model. Furthermore, because this process does not change the existing weight parameters of each component, the data processing model's original ability to identify faults in other modules is fully preserved, effectively solving the problem of "catastrophic forgetting." "Catastrophic forgetting" refers to the problem in which, for a data model composed of a large number of weight parameters, the introduction of new knowledge (i.e., knowledge related to identifying faults in newly added modules) causes the model to forget old knowledge (i.e., knowledge related to identifying faults in existing modules).

[0174] On the other hand, when identifying the fault of a newly added module, the data processing model of this embodiment can fuse the device features corresponding to the newly added module and the device features of the original module, so that when identifying the fault of the newly added module, the knowledge obtained by the data processing model when identifying the fault of the original module can be inherited and reused. For example, when identifying the fault of module 4, by fusing device features 1 to 3 into device feature 4, the knowledge obtained by the data processing model when identifying the fault of modules 1 to 3 can be inherited and reused, thereby improving the accuracy of identifying the fault of the newly added module.

[0175] When the method of this embodiment is implemented based on a data processing model, the preprocessing model, adapter, fusion module and neural network contained in the data processing model can all include a number of weight parameters, wherein the weight parameters contained in the preprocessing model can be predetermined based on conventional training methods and training data in the relevant technical field.

[0176] The adapter, the fusion module and the weight parameters contained in the neural network can be determined before executing the above method based on the pre-collected sample device characteristics and the corresponding actual fault conditions.

[0177] Among them, such as Figure 4 As shown, the method of determining the weight parameters contained in the adapter may include the following steps.

[0178] S401 : Processing sample device features according to multiple adapters to obtain fault identification results of the sample device features.

[0179] S402 : Compare the actual fault situation corresponding to the sample device characteristics with the fault identification result of the sample device characteristics, and update the parameters contained in the plurality of adapters based on the comparison result.

[0180] Referring to the description of the sample device in the aforementioned embodiment, in S401 , multiple groups of sample device information may be obtained respectively, each group of sample device information being device information of a sample device whose actual fault condition has been determined.

[0181] For each set of sample device information, the method of the aforementioned embodiment can be used to process this set of sample device information using a preprocessing model and multiple adapters to obtain multiple sample device features corresponding to multiple modules, and then obtain fault identification results for the multiple modules corresponding to this set of sample device features.

[0182] by Figure 2 For example, a group of sample device information is processed using the preprocessing model and adapters 1 to 3 to obtain sample device features 1 to 3. The sample device features are then further processed in combination with the subsequent fusion module and neural network to obtain the fault identification results of sample device feature 1, the fault identification results of sample device feature 2, and the fault identification results of sample device feature 3. These three fault identification results respectively represent the fault conditions of modules 1 to module 3 of the sample device analyzed by the data processing model.

[0183] In S402 , the fault identification result corresponding to the sample device feature may be compared with the corresponding actual fault condition to obtain a comparison result that can characterize the degree of difference between the two.

[0184] The actual fault condition corresponding to a feature of a sample device represents the actual fault condition of the corresponding module in the sample device detected by manual inspection or other technical means.

[0185] Combine Figure 2 In the example, sample device feature 1 output by adapter 1 corresponds to module 1. Therefore, the actual fault condition corresponding to sample device feature 1 can be understood as the actual fault condition of module 1 in the sample device. The comparison result obtained by comparing the fault identification result corresponding to sample device feature 1 with the corresponding actual fault condition corresponds to module 1. This comparison result reflects the degree of difference between the fault identification result output by the data processing model and the actual fault condition when identifying the fault of module 1.

[0186] Therefore, in S402, the goal is to minimize the degree of difference reflected by the comparison results corresponding to module 1, determine the loss corresponding to adapter 1 based on the comparison results corresponding to module 1, and then adjust the parameters of adapter 1 based on the original initial parameters of adapter 1 according to the loss, and repeat the processes of S401 and S402 after the adjustment until the degree of difference reflected by the comparison results corresponding to module 1 is less than a certain threshold, thereby completing the process of determining the parameters of adapter 1.

[0187] Similarly, the fault identification result corresponding to the sample device feature 2 can be compared with the corresponding actual fault condition to obtain the comparison result corresponding to the module 2. The goal is to minimize the degree of difference reflected by the comparison result. Based on the comparison result, the loss of the adapter 2 is determined, and the parameters of the adapter 2 are repeatedly adjusted according to the above method to finally determine the parameters of the adapter 2.

[0188] The original initial parameters of the adapter can be determined by manual setting, random generation, copying the parameters of the original adapter, etc., without limitation.

[0189] The method of adjusting weight parameters according to loss can be found in related technologies and will not be described in detail.

[0190] The process of determining the parameters of the adapter 3 is as described above and will not be described in detail.

[0191] In this embodiment, the adjustment of parameters of multiple adapters may be performed simultaneously, or, when computing resources are limited, may be performed sequentially in a certain order, without limitation.

[0192] The parameters contained in the adapter are the weight parameters of the adapter mentioned above.

[0193] The weight parameters of each fusion module and each neural network can be determined according to the above method and will not be repeated here.

[0194] To illustrate the beneficial effects of the fault identification method of this embodiment, a comparison of the effects of the method of this embodiment and some related fault identification methods when applied to specific data sets is provided below.

[0195] Table 1

[0196] <![CDATA[A N ]]> F1 FM FWT iCaRL 0.6035 0.1005 21.61% -2.68 DynamicER 0.7966 0.3455 / -5.09 ACL 0.8429 0.3393 / 0.00 This application 0.8606 0.4197 / 2.60

[0197] In Table 1, A N It represents the accuracy of data recognition in a specific data set based on the corresponding method. F1 represents the F1 score, which can represent a harmonic average of the precision and recall rate of the recognition method. N The larger the value of F1 is, the more accurate the fault identification result obtained by the corresponding identification method is.

[0198] FM stands for forgetting measure, which measures the degree of decline in the model's performance on old tasks during continuous learning. A larger value indicates a more severe forgetting of old tasks. In the model used in this embodiment, as previously described, after adding one or more new adapters, fusion modules, and neural networks corresponding to the new modules, the adapters, fusion modules, and neural networks corresponding to the original modules remain intact. Therefore, performance on old tasks does not decline during continuous learning.

[0199] FWT stands for forward transfer. This parameter measures the contribution of previously learned tasks to the performance improvement of the current new task. A positive value indicates that the knowledge of the previous task has a positive effect on the new task, while a negative value indicates that the knowledge of the previous task has a negative effect on the new task.

[0200] The specific calculation methods of the above parameters can be found in relevant technical materials and will not be elaborated on here.

[0201] In Table 1, iCaRL represents the incremental classifier and representation learning model, Dynamic ER represents the dynamically extensible representation model, and ACL represents the adapter-based continuous learning model. The structure and working principle of these models can be found in relevant technical materials and will not be described in detail. Figure 2 The data processing model shown.

[0202] By comparing the parameters in Table 1, it can be seen that the fault identification method of this embodiment and the data processing model used therein can obtain more accurate fault identification results than the various models for fault identification provided in the related art. Moreover, when new modules that need to identify faults are introduced into the device to be identified, new adapters, fusion modules and neural networks are introduced into the original fault identification model for these new modules. This will not affect the accuracy of fault identification of the original modules. At the same time, the recognition ability of the model for the original modules can be combined with the fault identification of the new modules by fusing the device features, thereby improving the accuracy of identifying faults of the new modules.

[0203] This embodiment also provides an electronic device, see Figure 5 , including a memory 501 and a processor 502.

[0204] The memory 501 is used to store computer programs;

[0205] The processor 502 is configured to execute a computer program to:

[0206] Obtain multiple device features of the device to be identified, each device feature corresponds to a module, and the module is a hardware module or a software module of the device;

[0207] According to the correlation of different modules, the fusion weight of other device features is determined. Other device features refer to device features other than the target device features among multiple device features. The target device features are the device features corresponding to the target module.

[0208] Based on the fusion weight, other device features and target device features are fused to obtain fused device features;

[0209] Determine the fault identification result of the target module in the device to be identified based on the fusion device characteristics.

[0210] Optionally, the processor 502 fuses other device features with the target device features based on the fusion weight to obtain a fused device feature, including:

[0211] Based on the fusion weight, other device features that meet the first condition are fused with the target device feature to obtain a fused device feature;

[0212] The first condition includes:

[0213] The failure frequency of the modules corresponding to other device characteristics is greater than the failure frequency of the target module.

[0214] Optionally, the processor 502 determines the fusion weights of other device features based on the correlations between different modules, including:

[0215] Obtaining mutual exclusion strength data reflecting the correlation between the target module and other modules, where the other modules refer to modules corresponding to other device features;

[0216] The fusion weights of other device features are determined based on the target device features, other device features, and mutual exclusion strength data.

[0217] Optionally, the method in which the processor 502 determines the mutual exclusion strength data between the target module and other modules includes:

[0218] Obtaining module knowledge data of a target module and module knowledge data of other modules, wherein the module knowledge data of the target module includes sample device features corresponding to the target module, and the module knowledge data of other modules includes sample device features corresponding to other modules;

[0219] Determining a first boundary formed by the module knowledge data of the target module and a second boundary formed by the module knowledge data of other modules in the multidimensional space to which the module knowledge data belongs;

[0220] Based on the degree of overlap between the first boundary and the second boundary, mutual exclusion strength data of the target module and other modules are determined.

[0221] Optionally, the processor 502 is further configured to:

[0222] Updates the mutual exclusion strength data of the target module and other modules based on the target device characteristics and other device characteristics.

[0223] Optionally, the processor 502 determines the fusion weight of other device features based on the target device features, other device features, and the mutual exclusion strength data, including:

[0224] According to the gating parameters corresponding to the target module, the target device features, other device features, and the mutual exclusion strength data of the target module and other modules are calculated to obtain the fusion weights of other device features;

[0225] Among them, the gating parameters corresponding to the target module are determined according to the characteristics of the sample device.

[0226] Optionally, the fault identification result of the target module includes at least the fault probability value of the target module in the device to be identified. If the fault probability value is greater than the target threshold, it indicates that the target module of the device to be identified is faulty. If the fault probability value is less than or equal to the target threshold, it indicates that the target module of the device to be identified is not faulty.

[0227] Optionally, the processor 502 obtains multiple device features of the device to be identified, including:

[0228] Processing the device information of the device to be identified according to the preprocessing model in the pre-built data processing model to obtain preprocessing features;

[0229] The pre-processing features are processed respectively according to a plurality of adapters of the data processing model to obtain the device features output by each adapter, and each adapter corresponds to a module.

[0230] Optionally, the processor 502 is further configured to:

[0231] Processing sample device features according to the plurality of adapters to obtain a fault identification result of the sample device features;

[0232] The actual fault condition corresponding to the sample device signature is compared with the fault identification result of the sample device signature, so as to update the parameters contained in the plurality of adapters based on the comparison result.

[0233] The working principle of the electronic device of this embodiment can be found in the description of the relevant steps in the fault identification method provided in the above embodiment, and will not be described in detail.

[0234] It should be noted that the various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same and similar parts between the various embodiments can be referenced to each other.

[0235] For the convenience of description, the above systems or devices are described as being divided into various modules or units according to their functions. Of course, when implementing the present application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0236] Through the description of the above embodiments, it can be seen that those skilled in the art can clearly understand that the present application can be implemented by means of software plus the necessary general hardware platform. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a storage medium such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in various embodiments of the present application or certain parts of the embodiments.

[0237] Finally, it should be noted that, in this document, relational terms such as first, second, third, and fourth are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or apparatus comprising the element.

[0238] The above is only a preferred embodiment of the present application. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present application. These improvements and modifications should also be regarded as the scope of protection of the present application.

Claims

1. A fault identification method, comprising: Obtain multiple device features of the device to be identified, each of the device features corresponding to a module, and the module is a hardware module or a software module of the device; Determine the fusion weight of other device features according to the correlation of different modules, where other device features refer to device features other than the target device features among the multiple device features, and the target device features refer to the device features corresponding to the target module; fusing the other device features and the target device features based on the fusion weight to obtain a fused device feature; Determine a fault identification result of the target module in the device to be identified according to the fusion device characteristics.

2. The method according to claim 1, wherein fusing the other device features with the target device features based on the fusion weight to obtain a fused device feature comprises: fusing the other device features that meet the first condition and the target device features based on the fusion weight to obtain a fused device feature; The first condition includes: The failure frequency of modules corresponding to other device characteristics is greater than the failure frequency of the target module.

3. The method according to claim 1, wherein determining the fusion weights of other device features based on the correlations between different modules comprises: Obtaining mutual exclusion strength data reflecting the correlation between the target module and other modules, where the other modules refer to modules corresponding to the other device features; Determine the fusion weight of the other device features according to the target device features, the other device features and the mutual exclusion strength data.

4. The method according to claim 3, wherein the method for determining the mutual exclusion strength data between the target module and other modules comprises: Obtaining module knowledge data of a target module and module knowledge data of other modules, wherein the module knowledge data of the target module includes sample device features corresponding to the target module, and the module knowledge data of other modules includes sample device features corresponding to other modules; Determining a first boundary formed by the module knowledge data of the target module and a second boundary formed by the module knowledge data of other modules in the multidimensional space to which the module knowledge data belongs; Based on the overlap degree of the first boundary and the second boundary, mutual exclusion strength data of the target module and other modules is determined.

5. The method according to claim 4, further comprising: The mutual exclusion strength data of the target module and the other modules are updated according to the target device characteristics and the other device characteristics.

6. The method according to claim 3, wherein determining the fusion weight of other device features based on the target device feature, the other device features, and the mutual exclusion strength data comprises: Calculating the target device features, the other device features, and the mutual exclusion strength data of the target module and the other modules according to the gating parameters corresponding to the target module to obtain the fusion weights of the other device features; The gating parameters corresponding to the target module are determined according to the characteristics of the sample device.

7. According to the method according to claim 1, the fault identification result of the target module includes at least a fault probability value of the target module in the device to be identified, and if the fault probability value is greater than a target threshold, it indicates that there is a fault in the target module of the device to be identified; if the fault probability value is less than or equal to the target threshold, it indicates that there is no fault in the target module of the device to be identified.

8. The method according to claim 1, wherein obtaining multiple device features of the device to be identified comprises: Processing the device information of the device to be identified according to the preprocessing model in the pre-built data processing model to obtain preprocessing features; The pre-processing features are processed respectively according to a plurality of adapters of the data processing model to obtain device features output by each adapter, and each adapter corresponds to a module.

9. The method according to claim 8, further comprising: Processing sample device features according to the plurality of adapters to obtain a fault identification result of the sample device features; The actual fault condition corresponding to the sample device feature is compared with the fault identification result of the sample device feature, so as to update the parameters contained in the plurality of adapters based on the comparison result.

10. An electronic device comprising a memory and a processor; The memory is used to store computer programs; The processor is configured to execute the computer program to: Obtain multiple device features of the device to be identified, each of the device features corresponding to a module, and the module is a hardware module or a software module of the device; Determine the fusion weight of other device features according to the correlation of different modules, where other device features refer to device features other than the target device features among the multiple device features, and the target device features refer to the device features corresponding to the target module; fusing the other device features and the target device features based on the fusion weight to obtain a fused device feature; Determine a fault identification result of the target module in the device to be identified according to the fusion device characteristics.

Images