Equipment fault detection method and device, electronic equipment and storage medium

Abstract

The invention discloses an equipment fault detection method and device, electronic equipment and a storage medium. The equipment fault detection method comprises the following steps: acquiring equipment operation data; performing fault detection on the equipment operation data based on a pre-trained fault detection model to obtain a target fault detection result; the fault detection model comprises a first processing model, a second processing model and a decision fusion module; the first processing model is obtained by training the initial training model in a transfer learning mode, and fault detection is performed on the equipment operation data to obtain a first fault detection result; the second processing model is obtained by training the initial training model in a meta-learning mode, and fault detection is carried out on the equipment operation data to obtain a second fault detection result; and the decision fusion module is connected with the first processing model and the second processing model, and fuses the first fault detection result and the second fault detection result to obtain a target fault detection result, so that the accuracy of equipment fault detection is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a method, device, electronic device and storage medium for detecting equipment faults. Background Art

[0002] With the popularization and application of industrial equipment, equipment fault detection plays a vital role in ensuring equipment safety and improving production efficiency.

[0003] In the process of equipment fault diagnosis, due to the small amount of equipment operation data, the machine learning method is used to perform a simple linear transformation on the equipment operation data to extract the features in the equipment operation data. This cannot fully express the potential fault characteristics, resulting in low accuracy of equipment fault diagnosis. Summary of the Invention

[0004] The present invention provides a method, apparatus, electronic device and storage medium for detecting equipment failure, so as to solve the problem of low accuracy in equipment failure prediction when there is a small amount of equipment operation data in the prior art.

[0005] According to one aspect of the present invention, a method for detecting a device failure is provided, the method comprising:

[0006] Obtain equipment operation data;

[0007] Perform fault detection on equipment operation data based on a pre-trained fault detection model to obtain target fault detection results; wherein the fault detection model includes a first processing model, a second processing model, and a decision fusion module;

[0008] The first processing model is obtained by training the initial training model through transfer learning, and performs fault detection on the equipment operation data to obtain a first fault detection result;

[0009] The second processing model is obtained by training the initial training model through a meta-learning method, and performs fault detection on the equipment operation data to obtain a second fault detection result;

[0010] The decision fusion module is connected to the first processing model and the second processing model respectively, and fuses the first fault detection result and the second fault detection result to obtain a target fault detection result.

[0011] According to another aspect of the present invention, there is provided a device for detecting equipment failure, the device comprising:

[0012] Equipment operation data acquisition module, used to obtain equipment operation data;

[0013] A target fault detection result determination module is used to perform fault detection on the equipment operation data based on a pre-trained fault detection model to obtain a target fault detection result; wherein the fault detection model includes a first processing model, a second processing model and a decision fusion module;

[0014] The first processing model is obtained by training the initial training model through transfer learning, and performs fault detection on the equipment operation data to obtain a first fault detection result;

[0015] The second processing model is obtained by training the initial training model through a meta-learning method, and performs fault detection on the equipment operation data to obtain a second fault detection result;

[0016] The decision fusion module is connected to the first processing model and the second processing model respectively, and fuses the first fault detection result and the second fault detection result to obtain a target fault detection result.

[0017] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0018] at least one processor; and

[0019] a memory communicatively connected to at least one processor; wherein,

[0020] The memory stores a computer program that can be executed by at least one processor. The computer program is executed by the at least one processor so that the at least one processor can perform the device failure detection method of any embodiment of the present invention.

[0021] According to another aspect of the present invention, a computer-readable storage medium is provided. The computer-readable storage medium stores computer instructions, which are used to enable a processor to implement the device failure detection method of any embodiment of the present invention when executed.

[0022] The technical solution of the embodiment of the present invention can provide data support for the analysis and processing of subsequent tasks by obtaining equipment operation data, ensuring that various tasks can be executed efficiently and accurately; fault detection is performed on the equipment operation data based on a pre-trained fault detection model to obtain a target fault detection result, thereby realizing equipment fault detection; wherein, the fault detection model includes a first processing model, a second processing model and a decision fusion module; the first processing model is obtained by training the initial training model through a transfer learning method, and fault detection is performed on the equipment operation data to obtain a first fault detection result; the second processing model is obtained by training the initial training model through a meta-learning method, and fault detection is performed on the equipment operation data to obtain a second fault detection result; the decision fusion module is connected to the first processing model and the second processing model respectively, and the first fault detection result and the second fault detection result are fused to obtain the target fault detection result, which is conducive to improving the accuracy of the fault detection model prediction, realizing equipment fault detection, and solving the problem of low accuracy of equipment fault prediction in the case of a small amount of equipment operation data in the prior art, thereby improving the accuracy of equipment fault prediction.

[0023] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present invention, nor is it intended to limit the scope of the present invention. Other features of the present invention will become readily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 This is a flow chart of a method for detecting device failure provided in Example 1 of the present invention;

[0026] Figure 2 is a comparison diagram of mapping features provided by an embodiment of the present invention;

[0027] Figure 3 1 is a schematic diagram of the structure of a residual neural network provided by an embodiment of the present invention;

[0028] Figure 4 is a comparison result diagram of a compression feature provided by an embodiment of the present invention;

[0029] Figure 5 This is a comparison diagram of the results of nonlinear mapping of compressed features by different activation functions provided by an embodiment of the present invention;

[0030] Figure 6 is a flowchart of a training method of a first processing model and a second processing model provided by Embodiment Two of the present application;

[0031] Figure 7 is a structural schematic diagram of a fault detection model provided by the present application;

[0032] Figure 8 is a structural schematic diagram of a device fault detection apparatus provided by Embodiment Three of the present application;

[0033] Figure 9 is a structural schematic diagram of an electronic device provided by Embodiment Four of the present application. DETAILED DESCRIPTION

[0034] In order to make the personnel in the technical field better understand the present application scheme, the technical scheme in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by the person skilled in the art without creative labor should belong to the scope of protection of the present application.

[0035] It should be noted that the terms "first", "second" and the like in the specification and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily have to describe a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device including a series of steps or units does not have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0036] Embodiment One

[0037] Figure 1 is a flowchart of a device fault detection method provided by Embodiment One of the present application. The present embodiment can be applicable to the case of detecting the fault of the device. The method can be executed by a device fault detection apparatus, which can be realized in the form of hardware and / or software. The device fault detection apparatus can be configured in the electronic device provided by the present application. The electronic device can be a server, a computer or a mobile terminal, etc. For example, the mobile terminal can be a mobile phone, a tablet computer, etc. As shown in the figure, the method comprises: Figure 1

[0038] ​S110, acquire equipment operation data.

[0039] In this embodiment, the equipment for fault detection in different fields is different, including but not limited to compressor, fan, rolling bearing, hydraulic pump and air turbine engine, etc., and the corresponding equipment can be selected according to the demand. The equipment operation data is various data information generated by the equipment in the running process, which can reflect the working state, performance and running environment of the equipment. Optionally, the equipment operation data includes but is not limited to the vibration signal, temperature and current of the equipment. The equipment operation data can be measured by a sensor. The equipment operation data can also be obtained by data. The database stores the equipment operation data of different equipment, and the equipment operation data of the equipment is matched in the database through the identification of the equipment. The identification of the equipment can be a unique code or other unique identification of the equipment, including but not limited to the production batch, model and other information of the equipment and the corresponding image recognition code, including but not limited to bar code and two-dimensional code. By acquiring the equipment operation data, data support can be provided for subsequent analysis and processing of tasks, ensuring that each task can be efficiently and accurately executed.

[0040] S120, based on the pre-trained fault detection model, the equipment operation data is detected to obtain the target fault detection result.

[0041] The fault detection model is a model for detecting faults of equipment operation data, which is used to determine whether the equipment has a fault and the corresponding fault type. The fault detection model includes but is not limited to a deep learning model, which can be selected according to the user's demand, and is not limited here. The target fault detection result is data representing whether the equipment has a fault and the corresponding fault type. The target fault detection result can be obtained according to the fault detection model. For example, the equipment operation data is input into the trained fault detection model to perform fault detection, and the target fault detection result is obtained.

[0042] In this embodiment, the fault detection model includes a first processing model, a second processing model and a decision fusion module; the first processing model is obtained by training the initial training model through the transfer learning method, and is used for fault detection of the equipment operation data to obtain the first fault detection result; the second processing model is obtained by training the initial training model through the meta-learning method, and is used for fault detection of the equipment operation data to obtain the second fault detection result; the decision fusion module is connected with the first processing model and the second processing model respectively, and the first fault detection result and the second fault detection result are fused to obtain the target fault detection result.

[0043] The first processing model is a structure obtained by training an initial training model through a transfer learning manner. The first processing model includes but is not limited to a machine learning model, and the structure of the initial training model is not limited herein. Optionally, the first processing model can be a residual neural network model. The initial training model is a basic model for constructing a fault detection model, and the initial training model includes but is not limited to a deep learning model. Optionally, the initial training model can be a residual neural network model. The first fault detection result is a result obtained by performing fault detection on the equipment operation data by using the first processing model, and is used to represent whether the equipment corresponding to the equipment operation data has a fault. The first fault detection result can include a fault type of the equipment and corresponding probability data.

[0044] The second processing model is a structure obtained by training the initial training model through a meta learning manner. The second processing model includes but is not limited to a machine learning model. Optionally, the second processing model can be a residual neural network model. The first processing model and the second processing model can have the same structure, or can have different structures, which are set according to requirements and are not limited herein. The second fault detection result is a result obtained by performing fault detection on the equipment operation data by using the second processing model, and is used to represent whether the equipment corresponding to the equipment operation data has a fault. The second fault detection result can include a fault type of the equipment and corresponding probability data.

[0045] Optionally, the first processing model and the second processing model have the same structure, and model parameters of the first processing model and model parameters of the second processing model are different. The model parameters of the first processing model and the model parameters of the second processing model can be set according to requirements and are not limited herein.

[0046] The decision fusion module is a structure for fusing the first fault detection result and the second fault detection result. The decision fusion module is connected with the first processing model and the second processing model respectively, and the first fault detection result corresponding to the first processing model and the second fault detection result corresponding to the second processing model are input into the decision fusion module for fusion to obtain a target fault detection result. The target fault detection result is the final fault judgment conclusion obtained by the fault detection model after fusing the outputs of the first processing model and the second processing model. The target fault detection result includes but is not limited to the final fault type of the equipment and the corresponding probability data. For example, the probability data corresponding to the fault type output by the first processing model and the second processing model can be subjected to a weighted summation operation to obtain a plurality of weighted fault detection results, and the target fault detection result is formed according to the probability data with the largest value in the weighted fault detection result and the corresponding fault type. For another example, the decision fusion module includes a decision fusion model, and the first fault detection result corresponding to the first processing model and the second fault detection result corresponding to the second processing model are input into the decision fusion model in the decision fusion module for fusion processing to obtain the target fault detection result.

[0047] The first processing model is used for fault detection on the equipment operation data to obtain a first fault detection result, the second processing model is used for fault detection on the equipment operation data to obtain a second fault detection result, and the decision fusion module is used for fusing the first fault detection result and the second fault detection result to obtain a target fault detection result, thereby realizing detection of equipment faults.

[0048] Optionally, the first processing model includes the following structure: a quantum entanglement kernel mapping unit, configured to perform feature mapping on the equipment operation data to obtain mapping features in a nonlinear space, the dimension of the mapping features being higher than the dimension of the equipment operation data; a dynamic kernel dimension compression unit, configured to perform compression processing on the mapping features to obtain compressed features; an activation function unit, configured to perform nonlinear mapping on the compressed features; a feature fusion unit, configured to perform weighted fusion on the features output by the activation function unit to obtain fused features; and a classification decision unit, configured to perform classification processing on the fused features to obtain the first fault detection result.

[0049] The quantum entanglement kernel mapping unit is a structure for performing feature mapping on the equipment operation data. The quantum entanglement kernel mapping unit is used to perform feature mapping on the equipment operation data to obtain mapping features in a nonlinear space. The equipment operation data includes complex nonlinear relationships, and the quantum entanglement kernel mapping unit can map the equipment operation data to a high-dimensional nonlinear space to better display the fault features in the equipment operation data.

[0050] The dynamic core dimension compression unit is a structure for compressing the mapping feature. The dynamic core dimension compression unit obtains the compressed feature important for fault detection by compressing the mapping feature. The compressed feature is the mapping feature compressed by the dynamic core dimension compression unit. The dynamic core dimension compression unit compresses the mapping feature to obtain the compressed feature important for fault detection, reduces the complexity of calculation, and is conducive to ensuring the integrity of fault information.

[0051] The activation function unit is a structure for nonlinearly mapping the compressed feature. In the activation function unit, the compressed feature is nonlinearly mapped to a more discriminative nonlinear space, which is conducive to improving the prediction accuracy of the first processing model.

[0052] The feature fusion unit is a structure for weighted fusion of the feature output by the activation function unit. The feature fusion unit obtains the fusion feature by weighted fusion of the feature output by the activation function unit. The fusion feature is a comprehensive feature vector obtained by weighted fusion of the feature output by the activation function unit by the feature fusion unit. The feature fusion unit integrates the feature output by the activation function unit into a representative fusion feature by assigning different weights to features of different dimensions and different importance.

[0053] The classification decision unit is a structure for classifying the fusion feature. The classification decision unit obtains the first fault detection result by classifying the fusion feature.

[0054] Optionally, the first processing model and the second processing model have the same structure. The second processing model includes the following structures: a quantum entangled core mapping unit configured to map the device running data to obtain mapping features in a nonlinear space, the dimension of the mapping features being higher than the dimension of the device running data; a dynamic core dimension compression unit configured to compress the mapping features to obtain compressed features; an activation function unit configured to nonlinearly map the compressed features; a feature fusion unit configured to weightedly fuse the features output by the activation function unit to obtain fusion features; and a classification decision unit configured to classify the fusion features to obtain a second fault detection result.

[0055] Optionally, the quantum entangled core mapping unit is specifically configured to: perform power mapping on the device running data to obtain high-dimensional features corresponding to the device running data; perform quantum entanglement processing on the high-dimensional features of each order to obtain entangled features corresponding to each order respectively; and perform weighted processing on the entangled features corresponding to each order respectively based on an entanglement weight coefficient to obtain the mapping features, wherein the entanglement weight coefficient is a learning parameter in the first processing model or the second processing model.

[0056] High-dimensional features are feature vectors obtained by power-mapping the device operation data using the quantum entanglement kernel mapping unit. The dimensionality of high-dimensional features is higher than that of the device operation data. Entangled features are features obtained by quantum entanglement processing of high-dimensional features of various orders and are used to characterize the correlation between high-dimensional features of different orders. The entanglement weight coefficient is a learning parameter in the first or second processing model that characterizes the importance of the entangled features. Entangled features of different orders correspond to different entanglement weight coefficients.

[0057] Specifically, the quantum entanglement core mapping unit maps the low-dimensional equipment operation data into high-dimensional features by performing power mapping on the equipment operation data, performs quantum entanglement processing on the high-dimensional features of each order, and obtains the entangled features corresponding to each order. The entangled features of different orders are weighted according to the entanglement weight coefficients corresponding to the entangled features of different orders to obtain mapping features, which is conducive to better displaying the fault characteristics in the equipment operation data and providing accurate data support for subsequent analysis and processing.

[0058] For example, the calculation formula of the mapping feature is as follows:

[0059]

[0060] Among them, y represents the mapping feature; x n Represents the high-dimensional features obtained by power mapping the equipment operation data; represents quantum entanglement; α n Represents the entanglement weight coefficient; N represents the maximum number of mappings of the equipment operation data. In order to improve the distribution characteristics and similarity of equipment operation data under different working conditions, the entanglement weight coefficient can be adaptively updated to highlight important fault characteristics. The update formula of the entanglement weight coefficient is as follows:

[0061]

[0062] Among them, α n represents the entanglement weight coefficient, x m Represents the operating data of the mth device, where m is a positive integer; x n Represents the nth device operation data, where n is a positive integer; γ m Represents the adaptive coefficient, which is used to characterize the weighting strategy of equipment operation data under different working conditions; <x m ,x n > represents the inner product, which is used to characterize the similarity between the mth device operation data and the nth device operation data; ‖x k ‖ 2 is the square of the modulus of the kth device running data, where k is a positive integer. Preferably, γ m Can be set to the variance of the equipment operation data.

[0063] For example, refer to Figure 2 , Figure 2 is a contrastive map of mapping features provided by an embodiment of the present application. In order to verify the nonlinear mapping capability of the quantum entanglement kernel mapping unit on the equipment operation data, the linear mapping, the wavelet transform and the quantum entanglement kernel mapping unit are respectively used to map the equipment operation data for the same equipment operation data, the mapping results of the above three methods are visualized and compared, and different colors of data points represent different fault types. In the mapping features obtained by the quantum entanglement kernel mapping unit, the data points of different fault types form a clear cluster distribution, and in the mapping features obtained by the linear mapping and the wavelet transform, there are a large number of overlapping areas of the data points of different fault types. It is illustrated that the feature mapping of the equipment operation data by the quantum entanglement kernel mapping unit can decouple the feature mapping into separable mapping features.

[0064] In some embodiments of the present application, the mapping features can be obtained by processing the equipment operation data by the residual neural network. The process of processing the equipment operation data by the residual neural network to obtain the mapping features is as follows: inputting the equipment operation data into the convolution layer of the residual neural network for processing to obtain mapping features of different orders; inputting the mapping features into the self-attention mechanism layer to calculate the entanglement features corresponding to the mapping features of different orders; calculating the entanglement weight coefficients corresponding to the mapping features of different orders, and inputting the entanglement weight coefficients and the entanglement features into the full connection layer for weighted processing to obtain the mapping features. For example, refer to Figure 3 , Figure 3 is a structure diagram of a residual neural network provided by an embodiment of the present application.

[0065] Optionally, the dynamic kernel dimension compression unit is specifically configured to: compress and process the mapping features based on a dynamic compression matrix, wherein the dynamic compression matrix is a learnable parameter matrix in the first processing model or the second processing model.

[0066] The dynamic compression matrix is a learnable parameter matrix in the first processing model or the second processing model, and is used for compressing and processing the mapping features. The dynamic compression matrix is used for compressing and processing the mapping features to obtain compressed features, thereby realizing the compression processing of the mapping features and providing data support for subsequent analysis.

[0067] Optionally, the dynamic kernel dimension compression unit includes a convolution layer. The size of the convolution kernel of the convolution layer can be 1*1, the size of the convolution kernel can also be 3*3, and the size of the convolution kernel can be set according to requirements, which is not limited here. The convolution layer also includes a weight matrix of the convolution kernel, and the dynamic compression matrix can be used as the weight matrix of the convolution kernel.

[0068] Optionally, during the training of the first processing model, the compression matrix corresponding to the t-th iterative training is obtained based on the compression matrix corresponding to the t-1-th iterative training, the learning rate, and the gradient of the first loss function corresponding to the t-1-th iterative training.

[0069] The first loss function is used to optimize the dynamic compression matrix and model parameters. Optionally, the calculation formula of the first loss function is as follows:

[0070]

[0071] Among them, L com represents the first loss function; y represents the mapping feature; Q represents the dynamic compression matrix; z t-1 Represents the compressed features corresponding to the t-1th iteration training; represents the square of the Frobenius norm; λ represents the regularization parameter, preferably, λ can be set to 0.3.

[0072] Optionally, during the training of the second processing model, the compression matrix corresponding to the t-th iterative training is obtained based on the compression matrix corresponding to the t-1-th iterative training, the learning rate, and the gradient of the first loss function corresponding to the t-1-th iterative training.

[0073] For example, the calculation formula of the compression feature is as follows:

[0074] z=Q·y;

[0075] Where z represents the compressed feature obtained after compression processing; Q represents the dynamic compression matrix; and y represents the mapping feature. In order to reduce the impact of redundant features unrelated to the fault characteristics, the dynamic compression matrix can be iteratively updated. The iterative update calculation formula of the dynamic compression matrix is ​​as follows:

[0076]

[0077] Among them, Q t represents the dynamic compression matrix corresponding to the t-th iteration training; Q t-1 represents the dynamic compression matrix corresponding to the t-1th iteration training; η t Indicates the learning rate corresponding to the t-th iteration training; represents the gradient of the first loss function.

[0078] In some embodiments of the present invention, the mapping features can be processed by a residual neural network to obtain compressed features. The process of the residual neural network processing the mapping features to obtain compressed features is as follows: the mapping features are input into the convolution layer of the residual neural network for convolution processing to obtain compressed features, wherein the convolution kernel size in the convolution layer is 1*1 and the weight matrix of the convolution kernel in the convolution layer is Q. During the training of the first processing model, a first loss function is set, which can be During each iterative training, the gradient of the first loss function corresponding to the current iterative training is calculated according to the first loss function corresponding to the current iterative training Using the back propagation mechanism, according to The learning rate and the first loss function corresponding to the current iterative training are used to calculate the weight matrix corresponding to the next iterative training, so as to update the weight matrix of the convolution kernel. The next iterative training is performed based on the updated weight matrix, and so on until all iterative training is completed.

[0079] For example, see Figure 4 , Figure 4 This is a comparison chart of compression features provided by an embodiment of the present invention. To verify the dynamic kernel dimensional compression unit's ability to compress mapping features, the same mapping feature was compressed using principal component analysis (PCA) and the dynamic kernel dimensional compression unit, and the compressed features obtained using the different methods were visually compared. As the dimension of the mapping feature increases and the compression ratio increases, the compressed features obtained using the dynamic kernel dimensional compression unit retain more features related to device failures.

[0080] Optionally, the activation function unit is specifically used to: perform nonlinear mapping on the compressed features through an adaptive activation function, wherein the adaptive activation function includes an activation adjustment coefficient; the activation adjustment coefficient is a learning parameter in the first processing model or the second processing model.

[0081] The adaptive activation function includes, but is not limited to, a ReLU activation function. Optionally, the adaptive activation function may be a Sigmoid activation function. The activation adjustment coefficient is a learnable parameter in the adaptive activation function, used to dynamically adjust the response strength of the adaptive activation function. Mapping the compressed features to a nonlinear space using the adaptive activation function is beneficial for improving the ability of the first processing model and the second processing model to distinguish compressed features under different operating conditions, thereby improving the accuracy of fault detection by the first processing model and the second processing model.

[0082] Optionally, during the training of the first processing model, the activation adjustment coefficient corresponding to the t-th iterative training is obtained based on the activation adjustment coefficient corresponding to the t-1-th iterative training, the learning rate, and the gradient of the second loss function corresponding to the t-1-th iterative training.

[0083] The second loss function is used to optimize the activation adjustment coefficient and the data of the model parameters. Optionally, the calculation formula of the second loss function is as follows:

[0084]

[0085] Among them, L act represents the second loss function; a represents the output feature of the activation function unit; Represents the expected output characteristics of the activation function unit output; μ represents the regularization coefficient; Represents the activation adjustment coefficient corresponding to the t-1th iteration training.

[0086] For example, the compressed features are nonlinearly mapped by an adaptive activation function, and the calculation formula of the features output by the activation function unit is as follows:

[0087]

[0088] Among them, a represents the feature output by the activation function unit; z represents the compressed feature; γ as represents the activation adjustment coefficient; f(·) indicates that the adaptive activation function is used to map the compressed features to the Sigmoid function. In order to improve the resolution of the adaptive activation function in high-noise environments or under various load conditions, the activation adjustment coefficient can be updated. The calculation formula for the activation adjustment coefficient update is as follows:

[0089]

[0090] in, Represents the activation adjustment coefficient corresponding to the t-th iteration training; represents the activation adjustment coefficient corresponding to the t-1th iteration training; η γ represents the learning rate of the activation function; represents the gradient of the adaptive activation function with respect to γ; L act Can be set to in, is the expected output of the adaptive activation function, μ represents the regularization coefficient. Preferably, η γ It can be set to 0.01 and μ can be set to 0.1.

[0091] Optionally, in the training process of the second processing model, the activation adjustment coefficient corresponding to the tth iteration training is obtained based on the activation adjustment coefficient corresponding to the (t-1)th iteration training, the learning rate, and the gradient of the second loss function corresponding to the (t-1)th iteration training.

[0092] For example, referring to Figure 5 , Figure 5 is a comparative result diagram of different activation functions for nonlinear mapping of compressed features provided by an embodiment of the application. Among them, the red fold represents the result obtained by using the adaptive activation function in the application for nonlinear mapping of compressed features, the black dotted line represents the compressed features, and the blue dotted line represents the result obtained by using the static activation function for nonlinear mapping of compressed features. As Figure 5 shown, the result obtained by using the adaptive activation function for nonlinear mapping of compressed features is more similar to the compressed features, which shows that the adaptive activation function can be adaptively adjusted according to different working conditions, which is beneficial to more accurately nonlinearly mapping the compressed features.

[0093] The feature fusion unit is a structure for weighted fusion of the features output by the activation function unit. The features output by the activation function unit are weighted fused by the feature fusion unit to obtain the fused features. The fused features are data obtained by weighted fusion of the features output by the activation function unit. The feature fusion unit includes a random forest, and the random forest is used to determine the weight coefficients corresponding to the features output by the activation function unit.

[0094] For example, the calculation formula of the fused features is as follows:

[0095]

[0096] Among them, s represents the fused features; β m represents the weight coefficient, which represents the importance of the mth-dimensional activation function unit output feature; a m represents the mth-dimensional component of the activation function unit output feature, and m is a positive integer. In order to improve the accuracy of the weight coefficient, the weight coefficient can be adaptively adjusted according to the coupling degree between the features output by the activation function unit. The calculation formula of the weight coefficient is as follows:

[0097]

[0098] Among them, β m represents the weight coefficient; δ n represents the feature importance factor, which is used to represent the intrinsic coupling between a m and a n ; a m represents the mth-dimensional activation component in the feature output by the activation function unit; a nrepresents the n-th dimension activation component in the feature output by the activation function unit, n represents a positive integer; m n represents an inner product, used to measure the correlation between features; k 2 represents the square of the length of the k-th dimension activation component in the feature output by the activation function unit, k is a positive integer. Preferably, δ n is set by calculating the feature importance of the feature output by the activation function unit through a preset random forest. The feature output by the activation function unit is input into the random forest for processing to obtain the feature importance factor δ n .

[0099] The classification decision unit is a structure for classifying the fusion feature. In the first processing model, the fusion feature is input into the classification decision unit for classification processing to obtain the first fault detection result, realizing the determination of the first fault detection result. Optionally, the classification decision unit can include a Softmax function.

[0100] For example, the calculation of the first fault detection result is as follows:

[0101]

[0102] wherein, represents the first fault detection result; s represents the fusion feature; W represents a classification weight matrix, used to represent the mapping relationship between the fusion feature and the fault type; soft(·) represents a Softmax function.

[0103] The technical scheme of the embodiment provides data support for subsequent task analysis and processing by obtaining the device operation data, ensuring that each task is efficiently and accurately executed; the target fault detection result is obtained by performing fault detection on the device operation data based on the pre-trained fault detection model, realizing the fault detection of the device; wherein the fault detection model includes a first processing model, a second processing model and a decision fusion module; the first processing model is obtained by training the initial training model through a transfer learning method, and performs fault detection on the device operation data to obtain the first fault detection result; the second processing model is obtained by training the initial training model through a meta-learning method, and performs fault detection on the device operation data to obtain the second fault detection result; the decision fusion module is connected with the first processing model and the second processing model, respectively, and fuses the first fault detection result and the second fault detection result to obtain the target fault detection result, which is conducive to improving the accuracy of the fault detection model prediction, realizing the fault detection of the device, solving the problem of low accuracy of device fault prediction in the prior art under the condition of a small amount of device operation data, and improving the accuracy of device fault prediction. ​​

[0104] Example 2

[0105] Figure 6 This is a flow chart of a method for training a fault detection model provided by the second embodiment of the present invention. The training process of the fault detection model includes a pre-training stage and a fine-tuning stage. Figure 6 As shown, the method includes:

[0106] S210. In the pre-training stage, transfer learning is performed on the initial training model based on the source domain data with batches as training units to obtain a first pre-training model, and meta-learning is performed on the initial training model based on the source domain data with fault detection tasks as training units to obtain a second pre-training model.

[0107] The source domain data is data related to, but not identical to, the application scenario. The source domain data includes operational data from multiple devices under different operating conditions and is characterized by large scale. The first pre-trained model is a model obtained by training the initial training model through transfer learning. The first pre-trained model includes, but is not limited to, a deep learning model. Optionally, the first pre-trained model may be a residual neural network model. The second pre-trained model is a model obtained by training the initial training model through meta-learning. The second pre-trained model includes, but is not limited to, a deep learning model. Optionally, the second pre-trained model may be a residual neural network model.

[0108] Specifically, during the transfer learning process, the source domain data is divided into batches as training units, and the initial training model is trained for transfer learning according to the batches of the training units to obtain a first pre-trained model.

[0109] During meta-learning, the source domain data is divided into training units based on the fault detection task. The equipment operation data corresponding to each fault type is used as a training unit. The initial model is trained using the source domain data trained using the fault detection task as a training unit to obtain a second pre-trained model. For example, if the number of fault types in the source domain data is B, the source domain data is divided into B training units based on the number of fault types. Each training unit is used to train the initial training model, resulting in B second pre-trained models.

[0110] S220. In the fine-tuning stage, transfer learning is performed on the first pre-trained model based on the target domain data with batches as training units to obtain a first processing model, and meta-learning is performed on the second pre-trained model based on the target domain data with fault detection tasks as training units to obtain a second processing model.

[0111] The target domain data is directly related to the application scenario and includes the operating data of multiple devices. The target domain data is characterized by its small size.

[0112] Specifically, in the process of transfer learning of the first pre-trained model, the target domain data is divided into batches into multiple training units, transfer learning is performed on the first pre-trained model according to the multiple training units, and the model parameters of the first pre-trained model are adjusted. When the first pre-trained model reaches the convergence condition, the training of the first pre-trained model is ended, and the trained first pre-trained model is used as the first processing model, thereby realizing the training of the first processing model, which is conducive to improving the accuracy of fault detection of the first processing model.

[0113] In the process of meta-learning the second pre-trained model, the target domain data is divided according to the fault detection task, and the target domain data corresponding to the same fault type is converted into a training unit. The corresponding second pre-trained model is meta-learned according to different training units to obtain a trained second pre-trained model. The trained second pre-trained model is used as the second processing model to realize the training of the second processing model, which is conducive to improving the accuracy of fault detection of the second processing model.

[0114] Optionally, the meta-learning process in the pre-training stage includes: obtaining sample data corresponding to each fault detection task, each fault type corresponds to a fault detection task; training the initial training model based on the sample data corresponding to each fault detection task to obtain model parameters corresponding to each fault detection task; determining the multi-task loss function based on the model parameters corresponding to each fault detection task, and adjusting the parameters of the initial training model based on the multi-task loss function.

[0115] The sample data is the device operation data corresponding to each fault detection task, which is obtained by dividing the source domain data into training units based on fault detection tasks. The multi-task loss function is used to measure the training effectiveness of the initial training model. A smaller multi-task loss function indicates a better initial training model; a larger multi-task loss function indicates a worse initial training model. The multi-task loss function can be determined based on the loss functions of the initial training models corresponding to each fault detection task. For example, the multi-task loss function can be obtained by performing a weighted summation of the loss functions of the initial training models corresponding to each fault detection task. The parameters of the initial training model are adjusted based on the multi-task loss function, and the adjusted initial training model is used as the second pre-trained model.

[0116] Specifically, each fault type corresponds to a fault detection task, and the sample data of the same fault type is used as a training unit. The initial training model is trained and processed according to the sample data corresponding to each fault detection task to obtain model parameters of the initial training models corresponding to different fault types, that is, model parameters corresponding to different fault detection tasks. The corresponding loss function is calculated according to the model parameters corresponding to each fault detection task, and a weighted summation is performed based on the loss functions corresponding to multiple fault detection tasks to obtain a multi-task loss function. The parameters of the initial training model are adjusted according to the multi-task loss function. When the multi-task loss function meets the preset threshold, the training of the initial training model is stopped to obtain the initial training model with adjusted parameters. The initial training model with adjusted parameters is used as the second pre-training model, thereby realizing pre-training of the second pre-training model, which is beneficial to improving the accuracy of the second processing model and providing accurate model support for subsequent analysis and processing.

[0117] Optionally, the meta-learning process in the fine-tuning stage includes: obtaining sample data corresponding to each fault detection task, each fault type corresponds to a fault detection task; training the second pre-trained model based on the sample data corresponding to each fault detection task to obtain model parameters corresponding to each fault detection task; determining a multi-task loss function based on the model parameters corresponding to each fault detection task, and adjusting the parameters of the second pre-trained model based on the multi-task loss function.

[0118] Specifically, each fault type corresponds to a fault detection task, and the sample data of the same fault type is used as a training unit. The second pre-trained model is trained and processed according to the sample data corresponding to each fault detection task to obtain model parameters of the second pre-trained model corresponding to different fault types, that is, model parameters corresponding to different fault detection tasks. The corresponding loss function is calculated according to the model parameters corresponding to each fault detection task, and a weighted summation calculation is performed according to the loss functions corresponding to multiple fault detection tasks to obtain a multi-task loss function. The parameters of the second pre-trained model are adjusted according to the multi-task loss function. When the multi-task loss function meets the preset threshold, the training of the second pre-trained model is stopped to obtain the second pre-trained model with adjusted parameters. The second pre-trained model with adjusted parameters is used as the second processing model, thereby realizing pre-training of the second processing model, which is beneficial to improving the accuracy of the second processing model, and further beneficial to improving the accuracy of fault detection of the fault detection model.

[0119] For example, see Figure 7 , Figure 7This is a training diagram of a fault detection model provided by an embodiment of the present invention. The source domain task represents the source domain data, the target task represents the sample data in the transfer learning process, and the auxiliary task represents the sample data in the meta-learning process. The training process of the fault detection model includes a pre-training stage and a fine-tuning stage. In the pre-training stage, the source domain task is used as a training unit for transfer learning of the initial training model to obtain a first pre-training model; the source domain task is used as a training unit for the fault detection task to perform meta-learning on the initial training model to obtain a second pre-training model. The initial training model is a residual neural network model. In the fine-tuning stage, the target task is used as a training unit for parameter adjustment of the first pre-training model to obtain a first processing model; the auxiliary task is used as a training unit for the fault detection task to perform meta-learning on the second pre-training model to obtain a second processing model, thereby realizing the fault detection task as a training unit and performing meta-learning on the second pre-training model to obtain a second processing model.

[0120] The technical solution of this embodiment is as follows: in the pre-training stage, based on the source domain data with batches as the training unit, the initial training model is transferred learned to obtain a first pre-training model, and based on the source domain data with fault detection tasks as the training unit, the initial training model is meta-learned to obtain a second pre-training model, thereby achieving accurate determination of the first pre-training model and the second pre-training model, and providing accurate model support for subsequent analysis and processing; in the fine-tuning stage, based on the target domain data with batches as the training unit, the first pre-training model is transferred learned to obtain a first processing model, and based on the target domain data with fault detection tasks as the training unit, the second pre-training model is meta-learned to obtain a second processing model, thereby achieving accurate determination of the first processing model and the second processing model, which is conducive to improving the accuracy of fault detection of the fault detection model.

[0121] Example 3

[0122] Figure 8 This is a schematic diagram of the structure of a device failure detection device provided by the third embodiment of the present invention. Figure 8 As shown, the device includes:

[0123] The device operation data acquisition module 310 is used to acquire device operation data;

[0124] A target fault detection result determination module 320 is configured to perform fault detection on the equipment operation data based on a pre-trained fault detection model to obtain a target fault detection result; wherein the fault detection model includes a first processing model, a second processing model, and a decision fusion module;

[0125] The first processing model is obtained by training the initial training model through transfer learning, and performs fault detection on the equipment operation data to obtain a first fault detection result;

[0126] The second processing model is obtained by training the initial training model through a meta-learning method, and performs fault detection on the equipment operation data to obtain a second fault detection result;

[0127] The decision fusion module is connected to the first processing model and the second processing model respectively, and fuses the first fault detection result and the second fault detection result to obtain a target fault detection result.

[0128] The technical solution of this embodiment can provide data support for the analysis and processing of subsequent tasks by acquiring equipment operation data, ensuring that various tasks can be executed efficiently and accurately; fault detection is performed on the equipment operation data based on a pre-trained fault detection model to obtain a target fault detection result, thereby realizing equipment fault detection; wherein, the fault detection model includes a first processing model, a second processing model and a decision fusion module; the first processing model is obtained by training the initial training model through a transfer learning method, and fault detection is performed on the equipment operation data to obtain a first fault detection result; the second processing model is obtained by training the initial training model through a meta-learning method, and fault detection is performed on the equipment operation data to obtain a second fault detection result; the decision fusion module is connected to the first processing model and the second processing model respectively, and the first fault detection result and the second fault detection result are fused to obtain the target fault detection result, which is conducive to improving the accuracy of the fault detection model prediction, realizing equipment fault detection, and solving the problem of low accuracy of equipment fault prediction in the case of a small amount of equipment operation data in the prior art, thereby improving the accuracy of equipment fault prediction.

[0129] Based on the above embodiments, optionally, the first processing model or the second processing model includes the following structure: a quantum entangled kernel mapping unit, used to perform feature mapping on the device operation data to obtain mapping features in a nonlinear space, where the dimension of the mapping features is higher than the dimension of the device operation data; a dynamic kernel dimension compression unit, used to perform compression processing on the mapping features to obtain compressed features; an activation function unit, used to perform nonlinear mapping on the compressed features; a feature fusion unit, used to perform weighted fusion on the features output by the activation function unit to obtain fused features; a classification decision unit, used to perform classification processing on the fused features to obtain a first fault detection result or a second fault detection result.

[0130] Optionally, the quantum entanglement kernel mapping unit is specifically used to: perform power mapping on the device operation data to obtain high-dimensional features corresponding to the device operation data; perform quantum entanglement processing on the high-dimensional features of each order to obtain entangled features corresponding to each order; perform weighted processing on the entangled features corresponding to each order based on the entanglement weight coefficient to obtain mapping features, wherein the entanglement weight coefficient is a learning parameter in the first processing model or the second processing model.

[0131] Optionally, the dynamic kernel dimension compression unit is specifically configured to compress the mapping features based on a dynamic compression matrix, wherein the dynamic compression matrix is a learning parameter in the first processing model or the second processing model.

[0132] Optionally, in the training process of the first processing model or the second processing model, the compression matrix corresponding to the tth iteration training is obtained based on the compression matrix corresponding to the (t-1)th iteration training, a learning rate, and a gradient of the first loss function corresponding to the (t-1)th iteration training.

[0133] Optionally, the activation function unit is specifically configured to perform non-linear mapping on the compressed features by an adaptive activation function, wherein the adaptive activation function includes an activation adjustment coefficient; and the activation adjustment coefficient is a learning parameter in the first processing model or the second processing model.

[0134] Optionally, in the training process of the first processing model or the second processing model, the activation adjustment coefficient corresponding to the tth iteration training is obtained based on the activation adjustment coefficient corresponding to the (t-1)th iteration training, a learning rate, and a gradient of the second loss function corresponding to the (t-1)th iteration training.

[0135] Optionally, the training process of the first processing model and the second processing model respectively includes a pre-training stage and a fine-tuning stage.

[0136] Optionally, in the pre-training stage, the initial training model is subjected to transfer learning based on source domain data in batches as training units to obtain the first pre-training model; and in the fine-tuning stage, the first pre-training model is subjected to transfer learning based on target domain data in batches as training units to obtain the first processing model.

[0137] Optionally, in the pre-training stage, the initial training model is subjected to meta-learning based on source domain data in fault detection tasks as training units to obtain the second pre-training model; and in the fine-tuning stage, the second pre-training model is subjected to meta-learning based on target domain data in fault detection tasks as training units to obtain the second processing model.

[0138] Optionally, the meta-learning process in the pre-training stage includes: obtaining sample data corresponding to each fault detection task, each fault type corresponding to a fault detection task; training the initial training model based on the sample data corresponding to each fault detection task to obtain model parameters corresponding to each fault detection task; determining a multi-task loss function based on the model parameters corresponding to each fault detection task, and adjusting the parameters of the initial training model based on the multi-task loss function.

[0139] Optionally, the meta-learning process in the pre-training stage includes: obtaining sample data corresponding to each fault detection task, each fault type corresponds to a fault detection task; training the second pre-training model based on the sample data corresponding to each fault detection task to obtain model parameters corresponding to each fault detection task; determining a multi-task loss function based on the model parameters corresponding to each fault detection task, and adjusting the parameters of the second pre-training model based on the multi-task loss function.

[0140] Optionally, the meta-learning process in the fine-tuning stage includes: obtaining sample data corresponding to each fault detection task, each fault type corresponds to a fault detection task; training the initial training model based on the sample data corresponding to each fault detection task to obtain model parameters corresponding to each fault detection task; determining the multi-task loss function based on the model parameters corresponding to each fault detection task, and adjusting the parameters of the initial training model based on the multi-task loss function.

[0141] Optionally, the meta-learning process in the fine-tuning stage includes: obtaining sample data corresponding to each fault detection task, each fault type corresponds to a fault detection task; training the second pre-trained model based on the sample data corresponding to each fault detection task to obtain model parameters corresponding to each fault detection task; determining a multi-task loss function based on the model parameters corresponding to each fault detection task, and adjusting the parameters of the second pre-trained model based on the multi-task loss function.

[0142] The device for detecting equipment failure provided in the embodiment of the present invention can execute the device failure detection method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0143] Example 4

[0144] Figure 9 1 is a structural diagram of an electronic device provided in Embodiment 4 of the present invention. The electronic device 10 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices (such as helmets, glasses, watches, etc.) and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0145] like Figure 9As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., which is communicatively connected to the at least one processor 11. The memory stores a computer program that can be executed by the at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 into the random access memory (RAM) 13. Various programs and data required for the operation of the electronic device 10 can also be stored in the random access memory (RAM) 13. The processor 11, the read-only memory (ROM) 12, and the random access memory (RAM) 13 are connected to each other via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0146] Various components in the electronic device 10 are connected to an input / output (I / O) interface 15, including an input unit 16, such as a keyboard, a mouse, etc.; an output unit 17, such as various types of displays, speakers, etc.; a storage unit 18, such as a magnetic disk, an optical disk, etc.; and a communication unit 19, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 19 allows the electronic device 10 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0147] The processor 11 can be any general-purpose and / or specialized processing component with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various specialized artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any other suitable processor, controller, microcontroller, etc. The processor 11 executes the various methods and processes described above, such as the device failure detection method.

[0148] In some embodiments, the device failure detection method can be implemented as a computer program, which is tangibly contained in a computer-readable storage medium, such as the storage unit 18. In some embodiments, part or all of the computer program can be loaded and / or installed on the electronic device 10 via the read-only memory (ROM) 12 and / or the communication unit 19. When the computer program is loaded into the random access memory (RAM) 13 and executed by the processor 11, one or more steps of the device failure detection method described above can be performed. Alternatively, in other embodiments, the processor 11 can be configured to execute the device failure detection method in any other appropriate manner (for example, by means of firmware).

[0149] Various embodiments of the systems and techniques described above can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on a chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include being implemented in one or more computer programs that are executable and / or interpreted on a programmable system that includes at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0150] Computer programs for implementing the device failure detection methods of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when executed by the processor, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The computer programs can be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0151] Example 5

[0152] Embodiment 5 of the present invention further provides a computer-readable storage medium storing computer instructions, the computer instructions being used to cause a processor to execute a device failure detection method, the method comprising:

[0153] Acquire equipment operation data; perform fault detection on the equipment operation data based on a pre-trained fault detection model to obtain a target fault detection result; wherein the fault detection model includes a first processing model, a second processing model and a decision fusion module; the first processing model is obtained by training the initial training model through a transfer learning method, and performs fault detection on the equipment operation data to obtain a first fault detection result; the second processing model is obtained by training the initial training model through a meta-learning method, and performs fault detection on the equipment operation data to obtain a second fault detection result; the decision fusion module is connected to the first processing model and the second processing model respectively, and fuses the first fault detection result and the second fault detection result to obtain a target fault detection result.

[0154] In the context of the present invention, computer-readable storage media can be tangible media that can contain or store a computer program for use with an instruction execution system, device or equipment or used in combination with an instruction execution system, device or equipment. Computer-readable storage media can include but are not limited to electronic, magnetic, optical, electromagnetic, infrared or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, computer-readable storage media can be machine-readable signal media. More specific examples of machine-readable storage media can include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0155] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0156] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), a blockchain network, and the Internet.

[0157] The computing system can include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system, to solve the defects of large management difficulty and weak business scalability in traditional physical host and VPS service.

[0158] It should be understood that the various forms of flow shown above can be used to reorder, add or delete steps. For example, each step described in the present application can be executed in parallel, sequentially or in a different order, as long as the desired results of the technical solutions of the present application can be achieved, which is not limited herein.

[0159] The above detailed description does not constitute a limitation on the protection scope of the present application. Those skilled in the art should understand that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modifications, equivalent replacements and improvements made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for detecting equipment failure, characterized in that: include: Obtain equipment operation data; Performing fault detection on the equipment operation data based on a pre-trained fault detection model to obtain a target fault detection result; wherein the fault detection model includes a first processing model, a second processing model and a decision fusion module; The first processing model is obtained by training the initial training model through transfer learning, and fault detection is performed on the equipment operation data to obtain a first fault detection result; The second processing model is obtained by training the initial training model through a meta-learning method, and performs fault detection on the equipment operation data to obtain a second fault detection result; The decision fusion module is connected to the first processing model and the second processing model respectively, and fuses the first fault detection result and the second fault detection result to obtain the target fault detection result.

2. The method according to claim 1, characterized in that The first processing model or the second processing model includes the following structure: a quantum entanglement core mapping unit, configured to perform feature mapping on the device operation data to obtain mapping features in a nonlinear space, wherein the dimension of the mapping features is higher than the dimension of the device operation data; A dynamic kernel dimension compression unit, configured to compress the mapping features to obtain compressed features; An activation function unit, configured to perform nonlinear mapping on the compression feature; A feature fusion unit, configured to perform weighted fusion on the features output by the activation function unit to obtain a fused feature; A classification decision unit is used to perform classification processing on the fusion feature to obtain the first fault detection result or the second fault detection result.

3. The method according to claim 2, characterized in that The quantum entanglement core mapping unit is specifically used for: Performing power mapping on the device operation data to obtain high-dimensional features corresponding to the device operation data; Perform quantum entanglement processing on the high-dimensional features of each order to obtain the entangled features corresponding to each order; The entanglement features corresponding to the respective orders are weighted based on an entanglement weight coefficient to obtain the mapping feature, wherein the entanglement weight coefficient is a learning parameter in the first processing model or the second processing model.

4. The method according to claim 2, characterized in that The dynamic kernel dimension compression unit is specifically used for: performing compression processing on the mapping features based on a dynamic compression matrix, wherein the dynamic compression matrix is ​​a learning parameter in the first processing model or the second processing model; During the training process of the first processing model or the second processing model, the compression matrix corresponding to the t-th iterative training is obtained based on the compression matrix corresponding to the t-1-th iterative training, the learning rate, and the gradient of the first loss function corresponding to the t-1-th iterative training.

5. The method according to claim 2, characterized in that The activation function unit is specifically used for: Performing nonlinear mapping on the compressed features through an adaptive activation function, wherein the adaptive activation function includes an activation adjustment coefficient; the activation adjustment coefficient is a learning parameter in the first processing model or the second processing model; During the training process of the first processing model or the second processing model, the activation adjustment coefficient corresponding to the t-th iterative training is obtained based on the activation adjustment coefficient corresponding to the t-1-th iterative training, the learning rate, and the gradient of the second loss function corresponding to the t-1-th iterative training.

6. The method according to claim 1, characterized in that The training process of the first processing model and the second processing model respectively includes a pre-training stage and a fine-tuning stage; In the pre-training phase, transfer learning is performed on the initial training model based on source domain data with batches as training units to obtain a first pre-training model; in the fine-tuning phase, transfer learning is performed on the first pre-training model based on target domain data with batches as training units to obtain the first processing model; In the pre-training stage, the initial training model is meta-learned based on the source domain data with the fault detection task as the training unit to obtain the second pre-training model; in the fine-tuning stage, the second pre-training model is meta-learned based on the target domain data with the fault detection task as the training unit to obtain the second processing model.

7. The method according to claim 6, characterized in that The meta-learning process in the pre-training stage or the fine-tuning stage includes: Obtain sample data corresponding to each fault detection task, where each fault type corresponds to a fault detection task; Training the initial training model or the second pre-training model based on sample data corresponding to each fault detection task to obtain model parameters corresponding to each fault detection task; A multi-task loss function is determined based on model parameters corresponding to each fault detection task, and parameters of the initial training model or the second pre-training model are adjusted based on the multi-task loss function.

8. A device for detecting equipment failure, characterized in that: include: Equipment operation data acquisition module, used to obtain equipment operation data; a target fault detection result determination module, configured to perform fault detection on the equipment operation data based on a pre-trained fault detection model to obtain a target fault detection result; wherein the fault detection model includes a first processing model, a second processing model, and a decision fusion module; The first processing model is obtained by training the initial training model through transfer learning, and fault detection is performed on the equipment operation data to obtain a first fault detection result; The second processing model is obtained by training the initial training model through a meta-learning method, and performs fault detection on the equipment operation data to obtain a second fault detection result; The decision fusion module is connected to the first processing model and the second processing model respectively, and fuses the first fault detection result and the second fault detection result to obtain the target fault detection result.

9. An electronic device, characterized in that: The electronic device comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor. The computer program is executed by the at least one processor so that the at least one processor can perform the device failure detection method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement the device failure detection method according to any one of claims 1 to 7 when executed.

Images