A potential fault judgment method and system for enterprise middle platform

Abstract

The embodiment of the present invention provides a potential fault judgment method and system for an enterprise middle station, which belongs to the field of artificial intelligence technology. The method includes: obtaining a historical state matrix and a historical fault matrix of an enterprise middle station; combining the historical state matrix and the historical fault matrix to obtain a sample set; obtaining a potential fault prediction network, and constructing a population of attention mechanisms in the current scenario; using each attention mechanism in the population to weight the samples of the sample set respectively; using the weighted sample sets to test the potential fault prediction network respectively to obtain the prediction accuracy of the potential fault prediction network; judging whether the current number of iterations is greater than or equal to a preset number threshold; when judging that the number of iterations is greater than or equal to the number threshold, adding the attention mechanism with the highest prediction accuracy to the input layer of the potential fault prediction network to obtain an updated potential fault prediction network.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a potential fault judgment method and system for an enterprise middle station. Background Art

[0002] The information system of the enterprise's middle office is an important guarantee for maintaining the stable operation of the enterprise. The influencing factors of the stable operation of the information system include hardware, software, management workflow, environment and many other factors. To ensure the continuous and stable operation of the information system of the enterprise's middle office, it is necessary to regularly diagnose the potential faults of the information system during system operation and maintenance.

[0003] When diagnosing information systems, the traditional method is to manually observe system data to determine possible anomalies. This method is relatively subjective and relies on the experience of the staff. With the development of artificial intelligence technology, some AI models have emerged for automatic potential fault judgment of information systems. However, this model is still in a relatively general stage and cannot adapt to the identification of potential faults in specific environments. Due to the large size of such models, if they need to be retrained for a specific environment, it will cause greater computing power and labor costs. Summary of the invention

[0004] The purpose of the embodiments of the present invention is to provide a potential fault judgment method and system for an enterprise middle station, which can complete the adaptive iteration of the potential fault judgment network while using relatively small resources.

[0005] In order to achieve the above object, an embodiment of the present invention provides a potential fault judgment method for an enterprise middle station, including:

[0006] Obtain the historical status matrix and historical fault matrix of the enterprise middle platform;

[0007] Combining the historical state matrix and the historical fault matrix to obtain a sample set;

[0008] Obtain the potential fault prediction network and build a population of attention mechanisms for the current scenario;

[0009] Using each of the attention mechanisms in the population to weight the samples of the sample set respectively;

[0010] Using the weighted sample sets to test the potential fault prediction network respectively to obtain the prediction accuracy of the potential fault prediction network;

[0011] Determine whether the current number of iterations is greater than or equal to a preset number threshold;

[0012] When it is determined that the number of iterations is greater than or equal to the number threshold, adding the attention mechanism with the highest prediction accuracy to the input layer of the potential fault prediction network to obtain an updated potential fault prediction network;

[0013] When it is determined that the number of iterations is less than the number threshold, each of the attention mechanisms in the population is updated, and the step of weighting the samples in the sample set using the attention mechanism is returned to be executed.

[0014] Optionally, obtain the historical status matrix and historical fault matrix of the enterprise middle station, including:

[0015] Determine multiple state quantities of the middle office system, wherein the state quantities include at least one of a server load rate, a network fluctuation value, a retention time of each node, a cumulative completion time of a single demand, and an accumulated amount of a demand pool;

[0016] Performing a normalization operation on the state quantity to obtain a corresponding standardized value;

[0017] Classify each historical fault to obtain multiple fault classes;

[0018] Calculate the quantity difference of each fault class respectively;

[0019] Determine whether the quantity difference is greater than or equal to a preset quantity threshold;

[0020] In the case where it is determined that the quantity threshold is greater than or equal to the quantity difference, returning to execute the step of classifying each historical fault to obtain a plurality of fault categories;

[0021] When it is determined that the quantity threshold is less than the quantity difference, the faults of the same category are taken as the same row, and the code 0 is used to indicate no fault and the code 1 is used to indicate the presence of a fault, so as to construct the historical fault matrix.

[0022] Optionally, each historical fault is classified to obtain multiple fault classes, including:

[0023] An unselected classification basis is selected from multiple classification bases for classification to obtain multiple fault classes.

[0024] Optionally, obtain a potential fault prediction network and construct a population of attention mechanisms for the current scenario, including:

[0025] The attention mechanism is determined according to formula (1):

[0026]

[0027] Where F is the attention mechanism, a ijis the weight of the code of the jth fault in the ith category, i∈[1,m], j∈[1,n], m is the number of fault categories, and n is the number of fault types under the same category.

[0028] Optionally, the potential fault prediction network includes:

[0029] Input layer;

[0030] A first convolutional layer, one end of the first convolutional layer is connected to the input layer;

[0031] A first pooling layer, wherein one end of the first pooling layer is connected to the other end of the first convolutional layer;

[0032] A second convolutional layer, one end of the second convolutional layer is connected to the other end of the first pooling layer;

[0033] A second pooling layer, one end of the second pooling layer is connected to the other end of the second convolutional layer;

[0034] A third convolutional layer, one end of the third convolutional layer is connected to the other end of the second pooling layer;

[0035] A third pooling layer, one end of the third pooling layer is connected to the other end of the third convolutional layer;

[0036] a fourth convolutional layer, one end of the fourth convolutional layer being connected to the other end of the third pooling layer;

[0037] a fourth pooling layer, one end of the fourth pooling layer being connected to the other end of the fourth convolutional layer;

[0038] a flattening layer, one end of which is connected to the other end of the fourth pooling layer;

[0039] A first fully connected layer, one end of the first fully connected layer is connected to the other end of the flattened layer;

[0040] A second fully connected layer, one end of the second fully connected layer is connected to the other end of the first fully connected layer;

[0041] An output layer, one end of the output layer is connected to the other end of the second fully connected layer.

[0042] Optionally, obtain a potential fault prediction network and construct a population of attention mechanisms for the current scenario, including:

[0043] Anchoring multiple weights in the attention mechanism according to a preset anchor selection box;

[0044] The multiple weights selected by the anchor are marked as the first-level weights to be updated to obtain the attention mechanism.

[0045] Optionally, updating each of the attention mechanisms in the population comprises:

[0046] Determine whether the difference between the prediction accuracy corresponding to the attention mechanism and the prediction accuracy of the previous round is positive;

[0047] When it is determined that the difference between the prediction accuracy corresponding to the attention mechanism and the prediction accuracy of the previous round is positive, the weight of the attention mechanism is updated according to formula (1) and formula (2):

[0048] a i ′=a i +Δa0×ln(Δy+1) / (i+1), (1)

[0049]

[0050] Among them, a i ″ is the updated weight, i is the weight a i ″ is the maximum row and column distance from the first-level weight to be updated, Δa0 is the update gradient, and Δy is the difference.

[0051] Optionally, updating each of the attention mechanisms in the population comprises:

[0052] When it is determined that the difference between the prediction accuracy corresponding to the attention mechanism and the prediction accuracy of the previous round is negative, the weight of the attention mechanism is updated according to formula (3):

[0053]

[0054] Among them, a′ is the updated weight, a is the weight before the update, and Δa is the updated value of weight a in the previous iteration.

[0055] On the other hand, the present invention also provides a potential fault judgment system for an enterprise middle station, the system comprising a processor, and the processor is configured to execute any of the methods described above.

[0056] Through the above technical scheme, the embodiment of the present invention provides a potential fault judgment method and system for an enterprise middle station. The method and system improve the input layer of the potential fault prediction network and use the pointer attention mechanism to improve the sensitivity of the potential fault prediction network to different factors, thereby improving the prediction accuracy of the potential fault prediction network for potential fault diagnosis of information systems in different scenarios.

[0057] Other features and advantages of the embodiments of the present invention will be described in detail in the subsequent detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] The accompanying drawings are used to provide a further understanding of the embodiments of the present invention and constitute a part of the specification. Together with the following specific implementations, they are used to explain the embodiments of the present invention, but do not constitute a limitation on the embodiments of the present invention. In the accompanying drawings:

[0059] Figure 1 is a flow chart of a potential fault determination method for an enterprise mid-terminal according to an embodiment of the present invention;

[0060] Figure 2 is a flow chart of a method for determining a historical state matrix and a historical fault matrix according to one embodiment of the present invention;

[0061] Figure 3 4 is a structural block diagram of a potential fault prediction network according to an embodiment of the present invention. DETAILED DESCRIPTION

[0062] The specific implementation of the embodiment of the present invention is described in detail below in conjunction with the accompanying drawings. It should be understood that the specific implementation described here is only used to illustrate and explain the embodiment of the present invention, and is not used to limit the embodiment of the present invention.

[0063] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application are in compliance with the relevant provisions of national laws and regulations. In the embodiments of this application, some existing solutions in the industry such as certain software, components, and models may be mentioned, which should be considered as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of this application, but it does not mean that the applicant has or will necessarily use the solution.

[0064] like Figure 1 The figure is a flow chart of a potential fault determination method for an enterprise mid-terminal according to an embodiment of the present invention. Figure 1 In the method, the method may include the following steps:

[0065] In step S10, the historical status matrix and historical fault matrix of the enterprise middle station are obtained;

[0066] In step S11, the historical state matrix and the historical fault matrix are combined to obtain a sample set;

[0067] In step S12, a potential fault prediction network is obtained, and a population of attention mechanism in the current scenario is constructed;

[0068] In step S13, each attention mechanism in the population is used to weight the samples in the sample set respectively;

[0069] In step S14, the potential fault prediction network is tested using the weighted sample sets to obtain the prediction accuracy of the potential fault prediction network;

[0070] In step S15, it is determined whether the current number of iterations is greater than or equal to a preset number threshold;

[0071] In step S16, when it is determined that the number of iterations is greater than or equal to the number threshold, the attention mechanism with the highest prediction accuracy is added to the input layer of the potential fault prediction network to obtain an updated potential fault prediction network;

[0072] In step S17, when it is determined that the number of iterations is less than the number threshold, each attention mechanism in the population is updated, and the process returns to the step of weighting the samples in the sample set using the attention mechanism, that is, returns to step S13.

[0073] In this Figure 1 In the method shown, step S10 can be used to obtain the historical state matrix and historical fault matrix of the enterprise middle station. The historical state matrix may include a variety of state quantities known to those skilled in the art. In one example of the present invention, the historical state matrix may include at least one of the server load rate, network fluctuation value, retention time of each node, cumulative completion time of a single demand, and accumulation of the demand pool. Further, the method of determining the historical state matrix and the historical fault matrix may include the following: Figure 2 In the steps shown in Figure 2 In the method of determining the historical state matrix and the historical fault matrix, the method may include the following steps:

[0074] In step S20, multiple state quantities of the middle station system are determined, wherein the state quantities include at least one of a server load rate, a network fluctuation value, a retention time of each node, a cumulative completion time of a single demand, and an accumulated amount of a demand pool;

[0075] In step S21, the state quantity is normalized to obtain a corresponding standardized value;

[0076] In step S22, each historical fault is classified to obtain multiple fault categories. Specifically, step S22 may be to select an unselected classification basis from multiple classification bases for classification to obtain multiple fault categories.

[0077] In step S23, the quantity difference of each fault class is calculated respectively;

[0078] In step S24, it is determined whether the quantity difference is greater than or equal to a preset quantity threshold;

[0079] In step S25, when it is determined that the quantity threshold is greater than or equal to the quantity difference, the process returns to the step of classifying each historical fault to obtain multiple fault categories;

[0080] In step S26, when it is determined that the quantity threshold is less than the quantity difference, faults of the same category are taken as the same row, with code 0 indicating no fault and code 1 indicating a fault, to construct a historical fault matrix.

[0081] Step S12 can be used to obtain a potential fault prediction network and construct a population of attention mechanisms in the current scenario. The potential fault prediction network can be in various forms known to those skilled in the art. In one example of the present invention, Figure 3 As shown, the potential fault prediction network may include an input layer 01, a first convolutional layer 02, a first pooling layer 03, a second convolutional layer 04, a second pooling layer 05, a third convolutional layer 06, a third pooling layer 07, a fourth convolutional layer 08, a fourth pooling layer 09, a flattening layer 10, a first fully connected layer 11, a second fully connected layer 12, and an output layer 13. Among them, one end of the first convolutional layer 02 may be connected to the input layer 01. One end of the first pooling layer 03 may be connected to the other end of the first convolutional layer 02. One end of the second convolutional layer 04 may be connected to the other end of the first pooling layer 03. One end of the second pooling layer 05 may be connected to the other end of the second convolutional layer 04. One end of the third convolutional layer 06 may be connected to the other end of the second pooling layer 05. One end of the third pooling layer 07 may be connected to the other end of the third convolutional layer 06. One end of the fourth convolutional layer 08 may be connected to the other end of the third pooling layer 07. One end of the fourth pooling layer 09 may be connected to the other end of the fourth convolutional layer 07. One end of the flattening layer 10 may be connected to the other end of the fourth pooling layer 09. One end of the first fully connected layer 11 may be connected to the other end of the flattening layer 10. One end of the second fully connected layer 12 may be connected to the other end of the first fully connected layer 11. One end of the output layer 13 may be connected to the other end of the second fully connected layer 12.

[0082] The attention mechanism may be in various forms known to those skilled in the art. In one example of the present invention, the attention mechanism may be determined using the following formula (1):

[0083]

[0084] Among them, F is the attention mechanism, a ij is the weight of the code of the jth fault in the ith category, i∈[1,m], j∈[1,n], m is the number of fault categories, and n is the number of fault types under the same category.

[0085] As for the method of generating the population of the attention mechanism, in one example of the present invention, multiple weights may be anchored in the attention mechanism according to a preset anchor selection box, and then the multiple anchored weights are marked as first-level weights to be updated to obtain an attention mechanism, and finally the method is repeatedly executed to obtain a population composed of multiple attention mechanisms.

[0086] Step S14 can be used to test the potential fault prediction network using the weighted sample sets to obtain the prediction accuracy of the potential fault prediction network. The prediction accuracy can be obtained by calculating the average value. Specifically, for example, 500 groups of samples are randomly selected for prediction, and the average value of the prediction accuracy of the 500 groups of samples is calculated as the final prediction accuracy.

[0087] When it is determined that the number of iterations is less than the number threshold, each attention mechanism in the population can be updated through step S17. The specific method for updating the attention mechanism can be a variety of forms known to those skilled in the art. In one example of the present invention, when the prediction accuracy after the update is higher than the prediction accuracy before the update, the difference between the prediction accuracy corresponding to the attention mechanism and the prediction accuracy of the previous round is positive, and the following formulas (1) and (2) can be used to update the weight of the attention mechanism:

[0088] a i ′=a i +Δa0×ln(Δy+1) / (i+1), (1)

[0089]

[0090] Among them, a i ″ is the updated weight, i is the weight a i ″ is the maximum row and column distance from the first-level weight to be updated, Δa0 is the update gradient, and Δy is the difference.

[0091] On the contrary, if the prediction accuracy after the update is lower than the prediction accuracy before the update, the difference between the prediction accuracy corresponding to the attention mechanism and the prediction accuracy of the previous round is negative, which means that the weight iteration is excessive. Therefore, the following formula (3) can be used to update the weight of the attention mechanism:

[0092]

[0093] Among them, a′ is the updated weight, a is the weight before the update, and Δa is the updated value of weight a in the previous iteration.

[0094] In addition, when the difference between the prediction accuracy corresponding to the attention mechanism and the prediction accuracy of the previous round is 0, it means that the current weight iteration may have reached the optimal solution (which may be the global optimal solution or the local optimal solution). Therefore, the above two update iteration methods can be randomly adopted at this time.

[0095] On the other hand, the present invention also provides a potential fault judgment system for an enterprise middle office, the system comprising a processor, the processor being configured to execute any of the above methods. Specifically, the method may include the following steps:

[0096] In step S10, the historical status matrix and historical fault matrix of the enterprise middle station are obtained;

[0097] In step S11, the historical state matrix and the historical fault matrix are combined to obtain a sample set;

[0098] In step S12, a potential fault prediction network is obtained, and a population of attention mechanism in the current scenario is constructed;

[0099] In step S13, each attention mechanism in the population is used to weight the samples in the sample set respectively;

[0100] In step S14, the potential fault prediction network is tested using the weighted sample sets to obtain the prediction accuracy of the potential fault prediction network;

[0101] In step S15, it is determined whether the current number of iterations is greater than or equal to a preset number threshold;

[0102] In step S16, when it is determined that the number of iterations is greater than or equal to the number threshold, the attention mechanism with the highest prediction accuracy is added to the input layer of the potential fault prediction network to obtain an updated potential fault prediction network;

[0103] In step S17, when it is determined that the number of iterations is less than the number threshold, each attention mechanism in the population is updated, and the process returns to the step of weighting the samples in the sample set using the attention mechanism, that is, returns to step S13.

[0104] In this Figure 1 In the method shown, step S10 can be used to obtain the historical state matrix and historical fault matrix of the enterprise middle station. The historical state matrix may include a variety of state quantities known to those skilled in the art. In one example of the present invention, the historical state matrix may include at least one of the server load rate, network fluctuation value, retention time of each node, cumulative completion time of a single demand, and accumulation of the demand pool. Further, the method of determining the historical state matrix and the historical fault matrix may include the following: Figure 2In the steps shown in Figure 2 In the method of determining the historical state matrix and the historical fault matrix, the method may include the following steps:

[0105] In step S20, multiple state quantities of the middle station system are determined, wherein the state quantities include at least one of a server load rate, a network fluctuation value, a retention time of each node, a cumulative completion time of a single demand, and an accumulated amount of a demand pool;

[0106] In step S21, the state quantity is normalized to obtain a corresponding standardized value;

[0107] In step S22, each historical fault is classified to obtain multiple fault categories. Specifically, step S22 may be to select an unselected classification basis from multiple classification bases for classification to obtain multiple fault categories.

[0108] In step S23, the quantity difference of each fault class is calculated respectively;

[0109] In step S24, it is determined whether the quantity difference is greater than or equal to a preset quantity threshold;

[0110] In step S25, when it is determined that the quantity threshold is greater than or equal to the quantity difference, the process returns to the step of classifying each historical fault to obtain multiple fault categories;

[0111] In step S26, when it is determined that the quantity threshold is less than the quantity difference, faults of the same category are taken as the same row, with code 0 indicating no fault and code 1 indicating a fault, to construct a historical fault matrix.

[0112] Step S12 can be used to obtain a potential fault prediction network and construct a population of the attention mechanism in the current scenario. The potential fault prediction network can be in various forms known to those skilled in the art. In one example of the present invention, the potential fault prediction network can include an input layer 01, a first convolutional layer 02, a first pooling layer 03, a second convolutional layer 04, a second pooling layer 05, a third convolutional layer 06, a third pooling layer 07, a fourth convolutional layer 08, a fourth pooling layer 09, a flattening layer 10, a first fully connected layer 11, a second fully connected layer 12, and an output layer 13. One end of the first convolutional layer 02 can be connected to the input layer 01. One end of the first pooling layer 03 can be connected to the other end of the first convolutional layer 02. One end of the second convolutional layer 04 can be connected to the other end of the first pooling layer 03. One end of the second pooling layer 05 can be connected to the other end of the second convolutional layer 04. One end of the third convolutional layer 06 can be connected to the other end of the second pooling layer 05. One end of the third pooling layer 07 may be connected to the other end of the third convolutional layer 06. One end of the fourth convolutional layer 08 may be connected to the other end of the third pooling layer 07. One end of the fourth pooling layer 09 may be connected to the other end of the fourth convolutional layer 07. One end of the flattening layer 10 may be connected to the other end of the fourth pooling layer 09. One end of the first fully connected layer 11 may be connected to the other end of the flattening layer 10. One end of the second fully connected layer 12 may be connected to the other end of the first fully connected layer 11. One end of the output layer 13 may be connected to the other end of the second fully connected layer 12.

[0113] The attention mechanism may be in various forms known to those skilled in the art. In one example of the present invention, the attention mechanism may be determined using the following formula (1):

[0114]

[0115] Among them, F is the attention mechanism, a ij is the weight of the code of the jth fault in the ith category, i∈[1,m], j∈[1,n], m is the number of fault categories, and n is the number of fault types under the same category.

[0116] As for the method of generating the population of the attention mechanism, in one example of the present invention, multiple weights may be anchored in the attention mechanism according to a preset anchor selection box, and then the multiple anchored weights are marked as first-level weights to be updated to obtain an attention mechanism, and finally the method is repeatedly executed to obtain a population composed of multiple attention mechanisms.

[0117] Step S14 can be used to test the potential fault prediction network using the weighted sample sets to obtain the prediction accuracy of the potential fault prediction network. The prediction accuracy can be obtained by calculating the average value. Specifically, for example, 500 groups of samples are randomly selected for prediction, and the average value of the prediction accuracy of the 500 groups of samples is calculated as the final prediction accuracy.

[0118] When it is determined that the number of iterations is less than the number threshold, each attention mechanism in the population can be updated through step S17. The specific method for updating the attention mechanism can be a variety of forms known to those skilled in the art. In one example of the present invention, when the prediction accuracy after the update is higher than the prediction accuracy before the update, the difference between the prediction accuracy corresponding to the attention mechanism and the prediction accuracy of the previous round is positive, and the following formulas (1) and (2) can be used to update the weight of the attention mechanism:

[0119] a i ′=a i +Δa0×ln(Δy+1) / (i+1), (1)

[0120]

[0121] Among them, a i ″ is the updated weight, i is the weight a i ″ is the maximum row and column distance from the first-level weight to be updated, Δa0 is the update gradient, and Δy is the difference.

[0122] On the contrary, if the prediction accuracy after the update is lower than the prediction accuracy before the update, the difference between the prediction accuracy corresponding to the attention mechanism and the prediction accuracy of the previous round is negative, which means that the weight iteration is excessive. Therefore, the following formula (3) can be used to update the weight of the attention mechanism:

[0123]

[0124] Among them, a′ is the updated weight, a is the weight before the update, and Δa is the updated value of weight a in the previous iteration.

[0125] In addition, when the difference between the prediction accuracy corresponding to the attention mechanism and the prediction accuracy of the previous round is 0, it means that the current weight iteration may have reached the optimal solution (which may be the global optimal solution or the local optimal solution). Therefore, the above two update iteration methods can be randomly adopted at this time.

[0126] Through the above technical scheme, the embodiment of the present invention provides a potential fault judgment method and system for an enterprise middle station. The method and system improve the input layer of the potential fault prediction network and use the pointer attention mechanism to improve the sensitivity of the potential fault prediction network to different factors, thereby improving the prediction accuracy of the potential fault prediction network for potential fault diagnosis of information systems in different scenarios.

[0127] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that include computer-usable program code.

[0128] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0129] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0130] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0131] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0132] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0133] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0134] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0135] The above are only embodiments of the present application and are not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included within the scope of the claims of the present application.

Claims

1. A potential fault judgment method for an enterprise middle station, characterized in that: The method comprises: Obtain the historical status matrix and historical fault matrix of the enterprise middle platform; Combining the historical state matrix and the historical fault matrix to obtain a sample set; Obtain the potential fault prediction network and build a population of attention mechanisms for the current scenario; Using each of the attention mechanisms in the population to weight the samples of the sample set respectively; Using the weighted sample sets to test the potential fault prediction network respectively to obtain the prediction accuracy of the potential fault prediction network; Determine whether the current number of iterations is greater than or equal to a preset number threshold; When it is determined that the number of iterations is greater than or equal to the number threshold, adding the attention mechanism with the highest prediction accuracy to the input layer of the potential fault prediction network to obtain an updated potential fault prediction network; When it is determined that the number of iterations is less than the number threshold, each of the attention mechanisms in the population is updated, and the step of weighting the samples in the sample set using the attention mechanism is returned to be executed.

2. The method according to claim 1, characterized in that: Obtain the historical status matrix and historical fault matrix of the enterprise platform, including: Determine multiple state quantities of the middle office system, wherein the state quantities include at least one of a server load rate, a network fluctuation value, a retention time of each node, a cumulative completion time of a single demand, and an accumulated amount of a demand pool; Performing a normalization operation on the state quantity to obtain a corresponding standardized value; Classify each historical fault to obtain multiple fault classes; Calculate the quantity difference of each fault class respectively; Determine whether the quantity difference is greater than or equal to a preset quantity threshold; In the case where it is determined that the quantity threshold is greater than or equal to the quantity difference, returning to execute the step of classifying each historical fault to obtain a plurality of fault categories; When it is determined that the quantity threshold is less than the quantity difference, the faults of the same category are taken as the same row, and the code 0 is used to indicate no fault and the code 1 is used to indicate the presence of a fault, so as to construct the historical fault matrix.

3. The method according to claim 2, characterized in that Each historical fault is classified to obtain multiple fault categories, including: An unselected classification basis is selected from multiple classification bases for classification to obtain multiple fault classes.

4. The method according to claim 1, characterized in that: Obtain the potential fault prediction network and construct a population of attention mechanisms for the current scenario, including: The attention mechanism is determined according to formula (1): Where F is the attention mechanism, a ij is the weight of the code of the jth fault in the ith category, i∈[1,m], j∈[1,n], m is the number of fault categories, and n is the number of fault types under the same category.

5. The method according to claim 1, characterized in that The potential fault prediction network includes: Input layer; A first convolutional layer, one end of the first convolutional layer is connected to the input layer; A first pooling layer, wherein one end of the first pooling layer is connected to the other end of the first convolutional layer; A second convolutional layer, one end of the second convolutional layer is connected to the other end of the first pooling layer; A second pooling layer, one end of the second pooling layer is connected to the other end of the second convolutional layer; A third convolutional layer, one end of the third convolutional layer is connected to the other end of the second pooling layer; A third pooling layer, one end of the third pooling layer is connected to the other end of the third convolutional layer; a fourth convolutional layer, one end of the fourth convolutional layer being connected to the other end of the third pooling layer; a fourth pooling layer, one end of the fourth pooling layer being connected to the other end of the fourth convolutional layer; a flattening layer, one end of which is connected to the other end of the fourth pooling layer; A first fully connected layer, one end of the first fully connected layer is connected to the other end of the flattened layer; A second fully connected layer, one end of the second fully connected layer is connected to the other end of the first fully connected layer; An output layer, one end of the output layer is connected to the other end of the second fully connected layer.

6. The method according to claim 1, characterized in that Obtain the potential fault prediction network and construct a population of attention mechanisms for the current scenario, including: Anchoring multiple weights in the attention mechanism according to a preset anchor selection box; The multiple weights selected by the anchor are marked as the first-level weights to be updated to obtain the attention mechanism.

7. The method according to claim 6, characterized in that Updating each of the attention mechanisms in the population, including: Determine whether the difference between the prediction accuracy corresponding to the attention mechanism and the prediction accuracy of the previous round is positive; When it is determined that the difference between the prediction accuracy corresponding to the attention mechanism and the prediction accuracy of the previous round is positive, the weight of the attention mechanism is updated according to formula (1) and formula (2): Among them, a i ″ is the updated weight, i is the weight a i ″ is the maximum row and column distance from the first-level weight to be updated, Δa0 is the update gradient, and Δy is the difference.

8. The method according to claim 7, characterized in that Updating each of the attention mechanisms in the population, including: When it is determined that the difference between the prediction accuracy corresponding to the attention mechanism and the prediction accuracy of the previous round is negative, the weight of the attention mechanism is updated according to formula (3): Among them, a ′ is the updated weight, a is the weight before the update, and Δa is the updated value of weight a in the previous iteration.

9. A potential fault judgment system for an enterprise middle station, characterized in that: The system comprises a processor configured to execute the method according to any one of claims 1 to 8.

Images