Multi-source domain adaptive fusion fault prediction method and device

Abstract

The invention relates to a multi-source-domain adaptive fusion fault prediction method and device, and belongs to the technical field of data processing.The multi-source-domain adaptive fusion fault prediction method comprises the steps that target domain data are input into a first prediction model, and first fault category probability distribution is obtained; inputting the first fault category probability distribution into a second prediction model to obtain second fault category probability distribution; wherein the first prediction model comprises K sub-prediction models, the first prediction model is obtained through training based on sample source domain data which are from K source domains and carry real labels and sample target domain data, and the second prediction model is obtained through training based on the sample target domain data. And inputting the sample target domain data into a pseudo tag obtained by the first prediction model for training. According to the method, the fault prediction efficiency of migration from the multi-source domain to the target domain is improved, and the fault prediction accuracy is further improved.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a multi-source domain adaptive fusion fault prediction method and device. Background Art

[0002] Deep learning methods can adaptively extract features from a large amount of data, meet the requirements of mechanical fault diagnosis for adaptive feature extraction, effectively overcome the shortcomings of poor generalization ability and poor robustness of traditional manually extracted features, and reduce the uncertainty of traditional fault diagnosis methods in the process of manual design and extraction. Therefore, deep learning technology can be used to establish an end-to-end diagnostic model, that is, to directly establish a diagnostic model based on the collected original data, omitting the complicated manual feature extraction process. However, the performance of deep learning fault diagnosis methods will decline when facing cross-domain diagnosis problems. This is because there are differences in the data distribution of source domain data and target domain data. Domain adaptive methods are an effective way to solve this problem. However, most domain adaptive methods are committed to extracting information more efficiently from a single source domain, ignoring the possibility of multiple source domains in reality. Due to the differences between different source domains, how to effectively fuse these source domains to predict the target domain remains a challenge.

[0003] Existing multi-source domain adaptive fault diagnosis methods focus on the source domain weighting strategy, without considering the alignment effect of different fault types in each source domain to the corresponding fault types in the target domain, which will significantly reduce the diagnostic efficiency of migrating from multiple source domains to the target domain. Summary of the invention

[0004] In view of this, it is necessary to provide a multi-source domain adaptive fusion fault prediction method and device to solve the problem of low diagnostic efficiency of multi-source domain to target domain migration in the existing multi-source domain adaptive fault diagnosis method.

[0005] In order to solve the above problems, the present invention provides a multi-source domain adaptive fusion fault prediction method, comprising: Inputting the target domain data into a first prediction model to obtain a first fault category probability distribution; Inputting the first fault category probability distribution into a second prediction model to obtain a second fault category probability distribution; Wherein, the first prediction model includes K sub-prediction models, wherein the first prediction model is based on K The second prediction model is trained based on sample source domain data carrying real labels in a source domain and sample target domain data, and the second prediction model is trained based on the sample target domain data and pseudo labels obtained by inputting the sample target domain data into the first prediction model.

[0006] In a possible implementation manner, each sub-prediction model in the first prediction model corresponds to a source domain.

[0007] In a possible implementation, inputting the target domain data into the first prediction model to obtain the first fault category probability distribution includes: Based on each sub-prediction model in the first prediction model, feature extraction and fault category probability distribution prediction are performed on the target domain data; A first fault category probability distribution is determined based on the fault category probability distribution prediction results output by each sub-prediction model in the first prediction model.

[0008] In a possible implementation, the training of the first prediction model includes: Based on the difference between the result of fault category probability distribution prediction of sample source domain data by each sub-prediction model in the first prediction model and the true label, each sub-prediction model in the first prediction model is trained.

[0009] In a possible implementation, inputting the first fault category probability distribution into a second prediction model to obtain a second fault category probability distribution includes: The first fault category probability distribution is input into the second prediction model, and the second fault category probability distribution is obtained based on the first weight matrix, the second weight matrix and the third weight matrix in the second prediction model and the first fault category probability distribution.

[0010] In a possible implementation, obtaining the second fault category probability distribution based on the first weight matrix, the second weight matrix and the third weight matrix in the second prediction model and the first fault category probability distribution includes: The second fault category probability distribution is determined based on the following formula:

[0011] in, represents the second fault category probability distribution, represents the normalization function, represents the first weight matrix, represents the second weight matrix, represents the third weight matrix, represents the probability distribution of the first fault category, The dimension representing the probability distribution of the first fault category.

[0012] In a possible implementation, the training of the second prediction model includes: Based on the result of the second prediction model predicting the fault category probability distribution of the sample target domain data and the difference between the pseudo labels corresponding to the sample target domain data, the first weight matrix, the second weight matrix and the third weight matrix in the second prediction model are optimized.

[0013] The present invention also provides a multi-source domain adaptive fusion fault prediction device, comprising: A first prediction module, used for inputting target domain data into a first prediction model to obtain a first fault category probability distribution; A second prediction module, configured to input the first fault category probability distribution into a second prediction model to obtain a second fault category probability distribution; Among them, the first prediction model includes K sub-prediction models, the first prediction model is trained based on sample source domain data carrying real labels from K source domains, and sample target domain data, and the second prediction model is trained based on the sample target domain data and pseudo labels obtained by inputting the sample target domain data into the first prediction model.

[0014] The present invention also provides an electronic device, including a memory and a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the multi-source domain adaptive fusion fault prediction method as described above is implemented.

[0015] The present invention also provides a computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the multi-source domain adaptive fusion fault prediction method as described above is implemented.

[0016] The beneficial effect of the present invention is as follows: the fault prediction method and device for adaptive fusion of multiple source domains provided by the present invention first migrate the multi-source domain data to the target domain data through a first prediction model, thereby realizing the initial fault prediction of the target domain data, and then realize the adaptive fusion of multiple source domains through a second prediction model with an attention mechanism, and further determine the final fault prediction result of the target domain data according to the initial fault prediction result of the target domain data, thereby improving the fault prediction efficiency of migrating from multiple source domains to the target domain, and thereby improving the accuracy of fault prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A schematic diagram of a flow chart of an embodiment of a multi-source domain adaptive fusion fault prediction method provided by the present invention; Figure 2 A schematic diagram of an embodiment of a domain weighting strategy and a fault category weighting strategy provided by the present invention; Figure 3 A schematic diagram of the structure of an embodiment of a multi-source domain adaptive fusion fault prediction device provided by the present invention; Figure 4 A schematic structural diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The preferred embodiments of the present invention are described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not used to limit the scope of the present invention.

[0019] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In addition, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0020] In the description of the present invention, reference to an "embodiment" means that a particular feature, structure, or characteristic described in conjunction with the embodiment may be included in at least one embodiment of the present invention. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive of other embodiments. It is explicitly and implicitly understood by those skilled in the art that the described embodiments may be combined with other embodiments.

[0021] Deep learning methods can adaptively extract features from a large amount of data, meet the requirements of mechanical fault diagnosis for adaptive feature extraction, effectively overcome the shortcomings of poor generalization ability and poor robustness of traditional manually extracted features, and reduce the uncertainty of traditional fault diagnosis methods in the process of manual design and extraction. Therefore, deep learning technology can be used to establish an end-to-end diagnostic model, that is, to directly establish a diagnostic model based on the collected original data, omitting the complicated manual feature extraction process. However, the performance of deep learning fault diagnosis methods will decline when facing cross-domain diagnosis problems. This is because there are differences in the data distribution of source domain data and target domain data. Domain adaptive methods are an effective way to solve this problem. However, most domain adaptive methods are committed to extracting information more efficiently from a single source domain, ignoring the possibility of multiple source domains in reality. Due to the differences between different source domains, how to effectively fuse these source domains to predict the target domain remains a challenge.

[0022] Existing multi-source domain adaptive fault diagnosis methods focus on the source domain weighting strategy, without considering the alignment effect of different fault types in each source domain to the corresponding fault types in the target domain, which will significantly reduce the diagnostic efficiency of migrating from multiple source domains to the target domain.

[0023] In order to solve the above problems, the present invention provides a fault prediction method of multi-source domain adaptive fusion.

[0024] The specific embodiments are described in detail below: A specific embodiment of the present invention discloses a fault prediction method for multi-source domain adaptive fusion, combining Figure 1 Come and see, Figure 1 The flowchart of an embodiment of a multi-source domain adaptive fusion fault prediction method provided by the present invention includes step S101 and step S102, wherein: In step S101, the target domain data is input into a first prediction model to obtain a first fault category probability distribution; In step S102, the first fault category probability distribution is input into a second prediction model to obtain a second fault category probability distribution; Wherein, the first prediction model includes K sub-prediction models, wherein the first prediction model is based on K The second prediction model is trained based on sample source domain data carrying real labels in a source domain and sample target domain data, and the second prediction model is trained based on the sample target domain data and pseudo labels obtained by inputting the sample target domain data into the first prediction model.

[0025] When implementing, we can firstly K The first prediction model is trained using sample source domain data carrying true labels from source domains and sample target domain data.

[0026] The first prediction model includes K This is a sub-prediction model. K The network structures of the sub-prediction models are the same, and are all built based on the single-source domain adaptive adversarial learning network. Each sub-prediction model includes a feature extractor, a task classifier, and a domain classifier. The function of the first prediction model is to migrate multi-source domain data to target domain data.

[0027] After the training of the first prediction model is completed, the sample target domain data can be input into the first prediction model to obtain the pseudo labels corresponding to the sample target domain data, and then the second prediction model is trained by the sample target domain data and its corresponding pseudo labels. The role of the second prediction model is to achieve adaptive fusion of multiple source domains through the attention mechanism.

[0028] After completing the training of the first prediction model and the second prediction model, the target domain data can be input into the first prediction model to obtain the first fault category probability distribution, and then the first fault category probability distribution can be input into the second prediction model to obtain the second fault category probability distribution. The second fault category probability distribution is the fault prediction result corresponding to the target domain data.

[0029] The multi-source domain adaptive fusion fault prediction method provided by the present invention can be applied to mechanical component fault prediction scenarios, and can also be applied to other types of fault prediction scenarios, and the present invention does not make specific limitations on this.

[0030] Compared with the prior art, the fault prediction method of multi-source domain adaptive fusion provided in this embodiment first migrates multi-source domain data to target domain data through a first prediction model, thereby realizing initial fault prediction of the target domain data, and then uses a second prediction model to realize adaptive fusion of multi-source domains with an attention mechanism, and further determines the final fault prediction result of the target domain data based on the initial fault prediction result of the target domain data, thereby improving the fault prediction efficiency of migrating from multi-source domains to the target domain, and thereby improving the accuracy of fault prediction.

[0031] Exemplarily, each sub-prediction model in the first prediction model corresponds to a source domain.

[0032] Specifically, each sub-prediction model in the first prediction model corresponds to a source domain, thereby better realizing the migration of multi-source domain data to target domain data.

[0033] Exemplarily, inputting the target domain data into the first prediction model to obtain the first fault category probability distribution includes: Based on each sub-prediction model in the first prediction model, feature extraction and fault category probability distribution prediction are performed on the target domain data; A first fault category probability distribution is determined based on the fault category probability distribution prediction results output by each sub-prediction model in the first prediction model.

[0034] Specifically, when the target domain data is input into the first prediction model to obtain the first fault category probability distribution, the target domain data can first be feature extracted according to each sub-prediction model in the first prediction model, the purpose of which is to map the target domain data to the same feature space as the source domain corresponding to the sub-prediction model. Then each sub-prediction model can output the corresponding fault category probability distribution prediction according to the extracted features.

[0035] Finally, the first fault category probability distribution may be determined according to the fault category probability distribution prediction results output by each sub-prediction model in the first prediction model.

[0036] Exemplarily, the training of the first prediction model includes: Based on the difference between the result of fault category probability distribution prediction of sample source domain data by each sub-prediction model in the first prediction model and the true label, each sub-prediction model in the first prediction model is trained.

[0037] Specifically, when training the first prediction model, each sub-prediction model in the first prediction model can be trained based on the difference between the result of each sub-prediction model in the first prediction model predicting the probability distribution of fault categories of sample source domain data and the true label.

[0038] In addition, the differences in domain classification results of each sub-prediction model can be combined during training.

[0039] For example, the loss function of the task classifier of each sub-prediction model can be expressed as:

[0040] The loss function of the domain classifier of each sub-prediction model can be expressed as:

[0041] The loss function of each sub-prediction model can be expressed as:

[0042] in, Indicates the data range, Represents source domain data, represents the target domain data, represents the binary cross entropy loss operation, represents the task classifier, represents the domain classifier, represents the features extracted by the feature extractor, represents the real fault category label, represents the real domain label, is an adaptive factor that changes with the number of iterations. During the model training process, the model can be optimized using the Stochastic Gradient Descent (SGD) optimization algorithm.

[0043] Exemplarily, inputting the first fault category probability distribution into a second prediction model to obtain a second fault category probability distribution includes: The first fault category probability distribution is input into the second prediction model, and the second fault category probability distribution is obtained based on the first weight matrix, the second weight matrix and the third weight matrix in the second prediction model and the first fault category probability distribution.

[0044] Specifically, when inputting the first fault category probability distribution into the second prediction model to obtain the second fault category probability distribution, the first fault category probability distribution can be first input into the second prediction model, and then the first weight matrix, second weight matrix and third weight matrix corresponding to the attention module in the second prediction model are combined with the first fault category probability distribution to obtain the second fault category probability distribution.

[0045] Exemplarily, obtaining the second fault category probability distribution based on the first weight matrix, the second weight matrix and the third weight matrix in the second prediction model and the first fault category probability distribution includes: The second fault category probability distribution is determined based on the following formula:

[0046] in, represents the second fault category probability distribution, represents the normalization function, represents the first weight matrix, represents the second weight matrix, represents the third weight matrix, represents the probability distribution of the first fault category, The dimension representing the probability distribution of the first fault category.

[0047] Specifically, the second fault category probability distribution can be calculated according to the above formula.

[0048] Exemplarily, the training of the second prediction model includes: Based on the result of the second prediction model predicting the fault category probability distribution of the sample target domain data and the difference between the pseudo labels corresponding to the sample target domain data, the first weight matrix, the second weight matrix and the third weight matrix in the second prediction model are optimized.

[0049] Specifically, when training the second prediction model, the first weight matrix, the second weight matrix and the third weight matrix in the second prediction model can be optimized based on the result of the fault category probability distribution prediction of the sample target domain data by the second prediction model and the difference between the pseudo labels corresponding to the sample target domain data.

[0050] For example, the loss function of the second prediction model can be expressed as:

[0051] in, Indicates the pseudo label corresponding to the sample target domain data. The model can be optimized using the SGD optimization algorithm during model training.

[0052] The technical solution of the present invention is better described below with reference to a specific embodiment: Assume that Source domain data and a single unlabeled target domain data. The target domain shares the state label space with all source domains, that is, the failure modes are the same, with a total of m types of failures, but there are different marginal distributions between different domains. From the distribution extracted from Indicates that it is taken from the source domain A sample of is the actual label corresponding to the sample. Target domain dataset From the distribution extracted from The purpose of multi-source domain adaptive learning is to make full use of data from multiple source domains to better capture shared features and knowledge, thereby improving the generalization performance of the model in the target domain.

[0053] This paper proposes a multi-source domain adaptive fault diagnosis network (SAMSAAN) based on self-attention mechanism. The whole model can be roughly divided into two parts: multi-branch network domain adaptation and multi-source domain adaptive fusion.

[0054] 1. Multi-branch network domain adaptation.

[0055] Multi-branch network domain adaptation is The structure of the branch network is based on a single-source domain adaptive adversarial learning network and consists of three parts: feature extractor, task classifier, and domain classifier. branch networks in parallel, thus achieving Migrate source domain data to target domain data.

[0056] (1) Feature extractor.

[0057] The feature extractor transforms the source domain data signal and target domain data signal Mapped to the same feature space, the branch network can extract and Domain invariant features.

[0058] (2) Task classifier.

[0059] The source domain data features extracted by the feature extractor Input to the task classifier, where Corresponding branch network The feature extractor is used to output the probability distribution of the sample prediction fault category of the source domain data belonging to different fault types. Accordingly, for each source domain task classifier, the cross entropy loss function is selected to optimize the model parameters, and its loss can be expressed as follows:

[0060] in is the fault type prediction label calculated by the task classifier based on the extracted source domain features, where Corresponding branch network The task classifier Represents the corresponding true sample label.

[0061] (3) Domain classifier.

[0062] The domain classifier identifies and distinguishes the source domain data features input from the feature extractor and target domain features , the loss can be expressed as follows:

[0063] in is the domain prediction label calculated by the domain classifier based on the extracted features, where Corresponding branch network The domain classifier Indicates the true domain label to which the corresponding sample belongs. Calculate their binary cross entropy loss.

[0064] The final loss function of the branch network is expressed as:

[0065] A gradient reversal layer (GRL) is used between the feature extractor and the domain classifier. During the back-propagation process, GRL takes the gradient from the domain classifier and multiplies it by , and pass it to the feature extractor. The adaptive factor It is a parameter that changes with the number of iterations, and the SGD optimization algorithm is used to optimize the model.

[0066] 2. Adaptive fusion of multiple source domains.

[0067] The multi-source domain adaptive fusion consists of two stages. In stage 1, pseudo labels are added to the unlabeled target domain samples based on the trained multi-branch network. In stage 2, the multi-source domain fusion module based on the self-attention mechanism is trained using the target domain data with pseudo labels.

[0068] (1) Pseudo-label screening.

[0069] Based on the idea of ​​ensemble learning, the target domain data is input into the trained branch network to obtain the probability distribution of sample prediction fault categories belonging to different fault types. The formula is as follows:

[0070] in, For the sample , The predicted fault category probability distribution set given by the branch network. Represents a branch network The predicted fault category probability distribution is given. Represents a branch network Given a sample belongs to the fault category probability.

[0071] Combine the probability distribution of fault categories predicted by branch networks together to obtain the joint prediction results of multi-branch networks for samples. This method uses features extracted from multiple models through multiple data projections, which may be complementary, to obtain excellent noise resilience and pseudo labels that are more strongly inclined to actual category labels.

[0072]

[0073] in Indicates that The predicted fault category probability distributions given by the branch networks are added together to obtain the joint predicted probability distribution. For the sample The joint prediction fault category label is selected. In the joint prediction result, the one with higher confidence is selected as the label of the target domain data, and the pseudo label Target domain data Composing labeled target domain data , as training data for the self-attention mechanism module.

[0074] (2) Self-attention mechanism module Combination Figure 2 Come and see, Figure 2This is a schematic diagram of an embodiment of the domain weighting strategy and fault category weighting strategy provided by the present invention. The self-attention mechanism can model different fault type information in multiple source domains. The output of each branch network can be regarded as a different position in the sequence. Therefore, the self-attention mechanism can dynamically adjust the degree of attention to different branch outputs, so that the specific information of each source domain can be better captured. This is very useful in multi-source domain adaptation tasks because each source domain may have different characteristics and distributions. The model pays more attention to branches with better performance or more contributions to the task. Secondly, when dealing with different fault category probability distributions, some categories may contribute less to the task, while others may play a key role in the success of the task. By dynamically adjusting the degree of attention to each probability, the attention mechanism can make the model more flexibly adapt to different task requirements, thereby achieving adaptive fusion of multiple source domains.

[0075] The target domain data enter After the trained branch network, the input of the self-attention mechanism network is obtained ,in It means the The probability distribution of predicted fault categories given by the branch network. The calculation formula of the attention mechanism is as follows:

[0076]

[0077] in , , They are the three different weight matrices required for the linear mapping of X, and They are the information mapping of the current prediction results and other prediction results, The feature map representing the input vector itself, and The dot product can be used to characterize the correlation between data. is the dimension of the input vector, is the normalization function, Indicates the final calculation result.

[0078] The final loss function of the attention module can be expressed as:

[0079] in Indicates that All rows of are accumulated to obtain a row vector, that is, the joint prediction fault category probability distribution, The cross entropy loss between the pseudo-label and the trained model is calculated, and the SGD optimization algorithm is used to optimize the model.

[0080] The network model optimization uses mini-batch SGD with momentum 0.9, and the batch size for each domain is set to 100. , the annealing strategy is adopted. The formula is as follows:

[0081] in It is a training progress that changes linearly from 0 to 1.

[0082] In order to reduce the noise interference from the domain classifier in the early stage of the training process, the adaptive factor Use the following formula to gradually change from 0 to 1:

[0083] In order to suppress the influence of pseudo-label uncertainty on multi-source domain adaptation fusion, the confidence threshold parameter is set to 0.85. Other parameters are set after several preliminary experiments.

[0084] In order to verify the effectiveness of the intelligent fault detection algorithm proposed in the present invention, two common single-source domain methods and two multi-source domain methods were used for comparison. For the convenience of comparison, all models adopted the same network architecture and hyperparameters as the present invention. The comparison methods are: CNN, MMD, DANN, DANNMB and MBWAAN. The final verification results confirm that the present invention can obtain good fault diagnosis results in most fault diagnosis scenarios, showing the effectiveness and superiority of the method.

[0085] The present invention adopts a multi-branch adversarial domain adaptive structure to reduce the data distribution difference between different source domains and target domains, thereby improving the fault diagnosis accuracy of the network model in the target domain data. Secondly, the weight coefficients of the fault types in different source domains are obtained through a pseudo-label training strategy. The weight coefficients are combined with the corresponding domain fault classifier results to obtain the target fault diagnosis results, thereby greatly improving the generalization ability and accuracy of the model.

[0086] The embodiment of the present invention also provides a multi-source domain adaptive fusion fault prediction device, combined with Figure 3 Come and see, Figure 3 This is a schematic diagram of the structure of an embodiment of a multi-source domain adaptive fusion fault prediction device provided by the present invention. The multi-source domain adaptive fusion fault prediction device 300 includes: The first prediction module 301 is used to input the target domain data into the first prediction model to obtain a first fault category probability distribution; A second prediction module 302, configured to input the first fault category probability distribution into a second prediction model to obtain a second fault category probability distribution; Among them, the first prediction model includes K sub-prediction models, the first prediction model is trained based on sample source domain data carrying real labels from K source domains, and sample target domain data, and the second prediction model is trained based on the sample target domain data and pseudo labels obtained by inputting the sample target domain data into the first prediction model.

[0087] The specific implementation methods of the various modules of the multi-source domain adaptive fusion fault prediction device can refer to the description of the multi-source domain adaptive fusion fault prediction method mentioned above, and have similar beneficial effects, which will not be repeated here.

[0088] It should be noted that the multi-source domain adaptive fusion fault prediction device can be set in a mechanical component processing factory or in a mechanical component design laboratory, and the present invention does not make specific limitations on this.

[0089] The embodiment of the present invention further provides an electronic device, Figure 4 Come and see, Figure 4 This is a structural diagram of an embodiment of an electronic device provided by the present invention. The electronic device 400 includes a processor 401, a memory 402, and a computer program stored in the memory 402 and executable on the processor 401. When the processor 401 executes the program, the fault prediction method of multi-source domain adaptive fusion as described above is implemented.

[0090] As a preferred embodiment, the electronic device 400 further includes a display 403 for displaying that the processor 401 executes the multi-source domain adaptive fusion fault prediction method as described above.

[0091] Exemplarily, the computer program may be divided into one or more modules / units, one or more modules / units are stored in the memory 402, and executed by the processor 401 to complete the present invention. One or more modules / units may be a series of computer program instruction segments capable of completing specific functions, and the instruction segments are used to describe the execution process of the computer program in the electronic device 400. For example, the computer program may be divided into the first prediction module 301 and the second prediction module 302 in the above embodiment, and the specific functions of each module are as described above, and are not described one by one here.

[0092] The electronic device 400 may be a desktop computer, a notebook, a PDA or a smart phone with an adjustable camera module.

[0093] Among them, the processor 401 may be an integrated circuit chip with signal processing capabilities. The above-mentioned processor 401 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field-programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components. The methods, steps and logic block diagrams disclosed in the embodiments of the present invention can be implemented or executed. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0094] The memory 402 may be, but is not limited to, a random access memory (RAM), a read only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electric erasable programmable read-only memory (EEPROM), etc. The memory 402 is used to store programs, and the processor 401 executes the program after receiving the execution instruction. The process definition method disclosed in any of the embodiments of the present invention may be applied to the processor 401 or implemented by the processor 401.

[0095] The display 403 may be an LCD display screen or an LED display screen, for example, a display screen on a mobile phone.

[0096] Understandably, Figure 4 The structure shown is only a schematic diagram of the structure of the electronic device 400. The electronic device 400 may also include Figure 4 More or fewer components as shown. Figure 4 Each component shown in the figure can be implemented by hardware, software or a combination thereof.

[0097] The electronic device provided according to the above-mentioned embodiment of the present invention can be implemented with reference to the specific description of the fault prediction method for adaptive fusion of multiple source domains as described above according to the present invention, and has similar beneficial effects as the fault prediction method for adaptive fusion of multiple source domains as described above, which will not be described in detail here.

[0098] An embodiment of the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the fault prediction method of multi-source domain adaptive fusion as described above is implemented.

[0099] Generally speaking, the computer instructions for implementing the method of the present invention can be carried by any combination of one or more computer-readable storage media. Non-transitory computer-readable storage media can include any computer-readable media except for the signal itself that is temporarily propagating.

[0100] Computer-readable storage media may be, for example, but not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatuses, or devices, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium may be any tangible medium that contains or stores a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0101] Computer program codes for performing the operations of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages, such as Java, Smalltalk, C++, and conventional procedural programming languages, such as "C" or similar programming languages, in particular, Python suitable for neural network computing and platform frameworks based on TensorFlow, PyTorch, etc. may be used. The program code may be executed entirely on the user's computer, partially on the user's computer, as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0102] Those skilled in the art will appreciate that all or part of the processes of the above-mentioned embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, wherein the computer-readable storage medium is a disk, an optical disk, a read-only storage memory, or a random access memory, etc.

[0103] The present invention discloses a fault prediction method and device for adaptive fusion of multiple source domains. Firstly, a first prediction model is used to migrate multi-source domain data to target domain data, thereby realizing initial fault prediction of the target domain data. Then, a second prediction model is used to realize adaptive fusion of multiple source domains with an attention mechanism. The final fault prediction result of the target domain data is further determined according to the initial fault prediction result of the target domain data, thereby improving the fault prediction efficiency of migrating from multiple source domains to the target domain, thereby improving the accuracy of fault prediction.

[0104] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by any technician familiar with the technical field within the technical scope disclosed by the present invention should be covered within the protection scope of the present invention.

Claims

1. A multi-source domain adaptive fusion fault prediction method, characterized in that: include: Inputting the target domain data into a first prediction model to obtain a first fault category probability distribution; Inputting the first fault category probability distribution into a second prediction model to obtain a second fault category probability distribution; Wherein, the first prediction model includes K sub-prediction models, wherein the first prediction model is based on K The second prediction model is trained based on sample source domain data carrying real labels in a source domain and sample target domain data, and the second prediction model is trained based on the sample target domain data and pseudo labels obtained by inputting the sample target domain data into the first prediction model.

2. The fault prediction method of multi-source domain adaptive fusion according to claim 1 is characterized in that: Each sub-prediction model in the first prediction model corresponds to a source domain.

3. The fault prediction method of multi-source domain adaptive fusion according to claim 2 is characterized in that: The step of inputting the target domain data into the first prediction model to obtain a first fault category probability distribution includes: Based on each sub-prediction model in the first prediction model, feature extraction and fault category probability distribution prediction are performed on the target domain data; A first fault category probability distribution is determined based on the fault category probability distribution prediction results output by each sub-prediction model in the first prediction model.

4. The fault prediction method of multi-source domain adaptive fusion according to claim 2 is characterized in that: The training of the first prediction model includes: Based on the difference between the result of fault category probability distribution prediction of sample source domain data by each sub-prediction model in the first prediction model and the true label, each sub-prediction model in the first prediction model is trained.

5. The fault prediction method of multi-source domain adaptive fusion according to claim 1, characterized in that: The step of inputting the first fault category probability distribution into a second prediction model to obtain a second fault category probability distribution includes: The first fault category probability distribution is input into the second prediction model, and the second fault category probability distribution is obtained based on the first weight matrix, the second weight matrix and the third weight matrix in the second prediction model and the first fault category probability distribution.

6. The multi-source domain adaptive fusion fault prediction method according to claim 5 is characterized in that: The obtaining the second fault category probability distribution based on the first weight matrix, the second weight matrix and the third weight matrix in the second prediction model and the first fault category probability distribution includes: The second fault category probability distribution is determined based on the following formula: in, represents the second fault category probability distribution, represents the normalization function, represents the first weight matrix, represents the second weight matrix, represents the third weight matrix, represents the probability distribution of the first fault category, The dimension representing the probability distribution of the first fault category.

7. The fault prediction method of multi-source domain adaptive fusion according to claim 5 is characterized in that: The training of the second prediction model comprises: Based on the result of the second prediction model predicting the fault category probability distribution of the sample target domain data and the difference between the pseudo labels corresponding to the sample target domain data, the first weight matrix, the second weight matrix and the third weight matrix in the second prediction model are optimized.

8. A multi-source domain adaptive fusion fault prediction device, characterized in that: include: A first prediction module, used for inputting target domain data into a first prediction model to obtain a first fault category probability distribution; A second prediction module, configured to input the first fault category probability distribution into a second prediction model to obtain a second fault category probability distribution; Among them, the first prediction model includes K sub-prediction models, the first prediction model is trained based on sample source domain data carrying real labels from K source domains, and sample target domain data, and the second prediction model is trained based on the sample target domain data and pseudo labels obtained by inputting the sample target domain data into the first prediction model.

9. An electronic device, characterized in that: The invention comprises a memory and a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, the fault prediction method of multi-source domain adaptive fusion according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the fault prediction method of multi-source domain adaptive fusion as claimed in any one of claims 1 to 7 is implemented.

Images