Power equipment fault diagnosis method and system based on category alignment

Abstract

The invention discloses a power equipment fault diagnosis method and system based on category alignment, and belongs to the technical field of fault prediction. The method comprises the following steps: acquiring sound data of to-be-diagnosed equipment; the method comprises the following steps: preprocessing sound data of to-be-diagnosed equipment, and then converting the sound data into a Mel spectrogram; and inputting the Mel spectrum into a pre-trained power equipment fault diagnosis domain adaptation model constructed based on category alignment to obtain a fault diagnosis result of the to-be-diagnosed equipment. Based on an optimal transportation theory and a domain adaptation technology, the power equipment fault diagnosis domain self-adaptation method based on category alignment is constructed.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a method and system for diagnosing faults of electric power equipment based on category alignment. Background Art

[0002] In recent years, although the fault monitoring technology for transformers has developed rapidly, with the continuous deepening of research on deep learning algorithms, the powerful fitting ability of deep algorithms has also been increasingly valued. In fault diagnosis tasks, due to factors such as limited fault samples and insufficient cross-scenario flexibility, the fault diagnosis model based on deep learning still has some shortcomings. In terms of data, due to the scarcity of single-scenario fault samples, it is difficult to effectively support the training of fault diagnosis models. At the same time, the existing collected samples need to be fully labeled data obtained manually, which brings huge cost problems; in terms of models, due to the large differences in working conditions of power equipment in different operating environments, there is an obvious distribution deviation between the data collected by different signal sources, which makes it difficult for the existing fault diagnosis model for a certain power equipment to be directly reused in the diagnosis tasks of the corresponding equipment in other working environments. Summary of the invention

[0003] In view of the above problems, the present invention proposes a method and system for power equipment fault diagnosis based on category alignment. The method is based on optimal transportation theory and domain adaptation technology to construct a domain adaptation method for power equipment fault diagnosis based on category alignment.

[0004] In order to solve the above technical problems, the technical solution adopted by the present invention is: In a first aspect, the present invention provides a method for diagnosing faults of electric power equipment based on category alignment, comprising: Acquire sound data of the device to be diagnosed; The sound data of the device to be diagnosed is pre-processed and then converted into a mel-spectrogram; The mel-spectrogram is input into the pre-trained domain adaptation model of power equipment fault diagnosis based on category alignment to obtain the fault diagnosis results of the equipment to be diagnosed.

[0005] In the above scheme, the training process of the pre-trained power equipment fault diagnosis domain adaptation model constructed based on category alignment includes: Acquire actually collected sound data in different domains; the sound data in different domains includes a source domain and a target domain; After preprocessing the sound data in different domains, they are converted into mel-spectrograms; The Mel-spectrogram is used for feature extraction using a feature extraction network, and the feature information is fused with the label information to obtain fused features of different domains; Based on a neural network, a domain adaptation model for fault diagnosis of electric power equipment based on category alignment is constructed, and the fusion features combined with weights of the different domains are used as input, and the classification results of the target data are used as output, so as to train the pre-trained domain adaptation model for fault diagnosis of electric power equipment based on category alignment.

[0006] As a further improvement of the present invention, the training process of the pre-trained power equipment fault diagnosis domain adaptation model constructed based on category alignment includes: Acquire actually collected sound data in different domains; the sound data in different domains includes a source domain and a target domain; After preprocessing the sound data in different domains, they are converted into mel-spectrograms; The Mel-spectrogram is used for feature extraction using a feature extraction network, and the feature information is fused with the label information to obtain fused features of different domains; Based on a neural network, a domain adaptation model for fault diagnosis of electric power equipment based on category alignment is constructed, and the fusion features combined with weights of the different domains are used as input, and the classification results of the target data are used as output, so as to train the pre-trained domain adaptation model for fault diagnosis of electric power equipment based on category alignment.

[0007] As a further improvement of the present invention, the preprocessing includes: first performing discrete sampling, and performing normalization processing on the data after the discrete sampling; uniformly cropping the normalized data to achieve signal scale alignment to obtain processed sound data.

[0008] As a further improvement of the present invention, the conversion into a mel-spectrogram comprises: For the processed sound data, firstly, it is represented as a spectrum graph through fast Fourier transform; Convert the obtained spectrum into Mel spectrum; The obtained Mel spectrum is passed through the Mel filter bank to perform logarithmic operations on all filter outputs; The logarithmic operation result is subjected to discrete cosine transform to obtain the Mel-spectrogram.

[0009] As a further improvement of the present invention, the Mel-spectrogram is subjected to feature extraction using a feature extraction network, and the feature information is fused with the label information, including: The processed Mel-atlas data is used as input, and the feature extractor is used to project the data of the source domain and the target domain into a shared space; Input the projected source domain features and target domain features into the classifier to obtain the classification results; The classification results are further fused with the projection features to obtain fused features in different domains.

[0010] As a further improvement of the present invention, the power equipment fault diagnosis domain adaptation model based on category alignment is constructed based on a neural network, and the fusion features of different domains combined with weights are used as input, and the classification results of the target data are used as output, and the pre-trained power equipment fault diagnosis domain adaptation model based on category alignment is trained, including: The fused features of different domains are sent to the category alignment module to calculate the loss function corresponding to each loss. Based on the neural network, the gradient between the output and input of the overall neural network is calculated. The feedforward neural network is back-propagated according to the loss function according to the gradient descent method, and the weights and biases in the network are iteratively updated. After iteration, the pre-trained domain adaptation model for fault diagnosis of electric equipment based on category alignment is obtained.

[0011] As a further improvement of the present invention, the loss function of the category alignment module is:

[0012] in, is the category alignment loss proposed based on the optimal transportation theory. For The additional constraint is used to ensure the category alignment module Can be optimized normally, is a weight parameter used to balance the model training and The proportion of attention; ,

[0013] Where D represents the category alignment module, sup represents the supremum, It is the fusion feature of different domains The backpropagated gradients of the class alignment module when passing through the class alignment module.

[0014] As a further improvement of the present invention, the overall loss of the power equipment fault diagnosis domain adaptation model based on category alignment is as follows:

[0015] in, and is a parameter that controls the balance between losses, The selection range of the traditional weight parameter is between 0.001 and 10. is set to , is the number of iterations of the current training.

[0016] As a further improvement of the present invention, the iterative updating of the weights and biases in the network further includes: The optimization method is used to further optimize the training process.

[0017] In a second aspect, the present invention provides a method for diagnosing faults of electric power equipment based on category alignment, comprising: An acquisition module, used for acquiring sound data of the device to be diagnosed; A preprocessing module is used to preprocess the sound data of the device to be diagnosed and then convert it into a mel-spectrogram; The diagnosis module is used to input the Mel-spectrogram into a pre-trained power equipment fault diagnosis domain adaptation model built based on category alignment to obtain the fault diagnosis result of the equipment to be diagnosed.

[0018] Preferably, a pre-training module is also included, and the pre-training module includes: The acquisition submodule is used to obtain the actually collected sound data in different domains; The preprocessing submodule is used to obtain the actually collected sound data in different domains; the sound data in different domains includes the source domain and the target domain; The conversion submodule is used to pre-process the sound data in different domains and then convert them into mel-spectrograms; The extraction submodule is used to extract features from the Mel-spectrogram using a feature extraction network, and fuse the feature information with the label information to obtain fused features of different domains; The training submodule is used to construct a domain adaptation model for fault diagnosis of electric equipment based on category alignment based on a neural network, and uses the fusion features of different domains combined with weights as input and the classification results of target data as output to train the pre-trained domain adaptation model for fault diagnosis of electric equipment based on category alignment.

[0019] In a third aspect, the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the class alignment-based power equipment fault diagnosis method when executing the computer program.

[0020] In a fourth aspect, the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for diagnosing faults of power equipment based on category alignment is implemented.

[0021] In a fifth aspect, the present invention provides a computer program product, wherein the computer program product comprises computer instructions, wherein the computer instructions instruct a computer to execute the method for diagnosing faults of power equipment based on category alignment.

[0022] The beneficial effects of the present invention compared with the prior art are as follows: The present invention proposes a method for fault diagnosis of electric equipment based on category alignment. By utilizing the fusion and alignment of feature information and label information, a fault diagnosis model for electric equipment is constructed to solve the problem of inaccurate fault diagnosis caused by traditional deep learning models only considering feature information training; based on the optimal transportation theory, a domain adaptation model for fault diagnosis of electric equipment based on category alignment is constructed to solve the problem of poor reliability of fault diagnosis when traditional deep learning models process multi-source data, and enhance the versatility of the model. Compared with traditional deep learning fault diagnosis methods, the present invention can make the distribution and category alignment of data closer through category alignment, and combine feature information and label information. It alleviates the problem of insufficient accuracy of the fault diagnosis model caused by data distribution deviation between different signal sources, and improves the versatility and diagnostic accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the embodiments of the present invention or the drawings of related technical solutions in the prior art are introduced below. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 A flow chart of a method for diagnosing faults of power equipment based on category alignment; Figure 2 Flowchart of the training process for the pre-trained model; Figure 3 Flowchart of the power equipment fault diagnosis model based on category alignment; Figure 4 A diagram of a power equipment fault diagnosis system based on category alignment; Figure 5 Schematic diagram of the structure of a pre-training module according to an embodiment of the present invention; Figure 6 The figure is a schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limitations of the present invention. For the step numbers in the following embodiments, they are only provided for the convenience of explanation, and the order between the steps is not limited in any way, and the execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.

[0026] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.

[0027] Terminology explanation: Adaptive Moment Estimation (ADAM), commonly referred to as Adam, is a stochastic function optimization algorithm based on gradient optimization.

[0028] Momentum: Based on gradient descent, the momentum term (i.e., the accumulated gradient) is introduced to update the parameters. The momentum term can accelerate the convergence speed of the model in the gradient direction and reduce oscillation.

[0029] RMSProp algorithm: uses the moving average of the squared gradient to dynamically adjust the learning rate to better adapt to changes in different parameters. This helps to improve the stability and convergence speed of model training.

[0030] AdaGrad algorithm (Adaptive Gradient Algorithm): As the training progresses, the learning rate is gradually decayed. It adjusts the learning rate according to the cumulative sum of the squared gradients of the parameters, so that the learning rate of each parameter is different.

[0031] AdaDelta algorithm: An improvement on the AdaGrad algorithm, it uses a more complex formula to calculate the decay of the learning rate. The AdaDelta algorithm does not require manual setting of the global learning rate.

[0032] In view of the problems existing in the prior art, the present invention constructs a domain adaptation method for fault diagnosis of power equipment based on category alignment based on optimal transportation theory and domain adaptation technology.

[0033] like Figure 1 As shown, the present invention proposes a method for diagnosing faults of electric power equipment based on category alignment, comprising: S1, obtaining sound data of the device to be diagnosed; S2, pre-processing the sound data of the device to be diagnosed and then converting it into a mel-spectrogram; S3, input the mel-spectrogram into the pre-trained domain adaptation model of power equipment fault diagnosis based on category alignment to obtain the fault diagnosis result of the equipment to be diagnosed.

[0034] like Figure 2 As shown, the training process of the pre-trained power equipment fault diagnosis domain adaptation model constructed based on category alignment includes the following steps: S101, obtaining actually collected sound data in different domains; the sound data in different domains includes a source domain and a target domain; S102, pre-processing the sound data in different domains and then converting them into mel-spectrograms; S103, extracting features from the mel-spectrogram using a feature extraction network, and fusing the feature information with the label information to obtain fused features of different domains; S104, based on the neural network, construct a domain adaptation model for fault diagnosis of electric equipment based on category alignment, and use the fusion features of different domains combined with weights as input and the classification results of target data as output to train and obtain the pre-trained domain adaptation model for fault diagnosis of electric equipment based on category alignment.

[0035] Based on the above scheme, the power equipment fault diagnosis method based on category alignment proposed in the present invention uses deep learning technology and domain adaptation principle to improve the accuracy of fault diagnosis. The method is mainly divided into two main stages: data acquisition and model application, and model pre-training.

[0036] The principles of data acquisition and model application are: Data is collected from the transformer device to be diagnosed. The data may be various types of data such as sound, vibration, temperature, etc., but the application of sound data is particularly emphasized in this invention.

[0037] The collected data is input into a pre-trained model, which is a domain adaptation model for power equipment fault diagnosis based on category alignment. The model is able to process sound data from different domains (e.g., data collected from different equipment and under different environmental conditions) and output fault diagnosis results.

[0038] The principle of pre-training of the model is: Stages S101 to S103 are the data preparation and feature extraction process before model training. First, sound data is collected from different domains, preprocessed (such as denoising, standardization, etc.), and then converted into Mel spectrograms. Mel spectrograms are a commonly used feature representation method in audio signal processing that can reflect the frequency characteristics of audio signals. Next, the feature extraction network is used to extract features from the Mel spectrograms and fused with label information to obtain fused features from different domains.

[0039] In the S104 stage, a domain adaptation model for power equipment fault diagnosis based on category alignment is constructed. The key to this model is that it can learn the common features between different domains, namely category alignment, so as to achieve accurate classification of target data. Through training, the model can learn how to extract useful information for fault diagnosis from fused features and output classification results.

[0040] Therefore, the present invention can utilize domain adaptation technology and category alignment principles. The method can process data from different domains, reduce diagnostic errors caused by data differences, and thus improve the accuracy of diagnosis. Since the model can learn the common features between different domains, it also has a certain generalization ability for data in new domains, and can perform accurate fault diagnosis under different equipment and environmental conditions. By utilizing domain adaptation technology, the method can reduce the dependence on target domain data to a certain extent, reducing the cost and difficulty of data acquisition. The method can be extended to other types of power equipment fault diagnosis, and only the model needs to be adjusted and optimized accordingly. At the same time, since the method mainly relies on sound data, it has high feasibility and practicality in practical applications. It can improve the accuracy of fault diagnosis, enhance the generalization ability of the model, and reduce the cost of data acquisition.

[0041] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0042] The present invention proposes a method for fault diagnosis of electric power equipment based on category alignment, and in particular relates to a transformer fault diagnosis technology based on multi-domain signals. The specific steps are as follows: First, the actual collected sound data is preprocessed, including signal denoising, normalization, signal cropping and alignment, etc.

[0043] Step 1: Data preprocessing of sound data.

[0044] Discrete sampling (sampling frequency 16000Hz) is performed on the collected device sound training signals and the sound data to be tested, and normalization is performed on the data after discrete sampling:

[0045] in, is the normalized sound data, is the original data value of the sound data, , They are the maximum and minimum signals in the original sound data respectively. The normalized data is uniformly cropped to achieve signal scale alignment and obtain the source domain sound training set , target domain sound training set And the test set , , where K is the signal length, , , , is the number of data in the corresponding dataset.

[0046] Secondly, the sound data is converted into the data format of Mel spectrogram.

[0047] Step 2: Mel-spectrogram conversion.

[0048] Convert the sound data set into a Mel-spectrogram format. For the processed sound data, first use the fast Fourier transform to represent it as a spectrogram:

[0049] in, is the transformed spectrum, Represents the time domain data sequence (corresponding to the training set , training set And the test set , single data in), K is the signal length.

[0050] For the obtained spectrum ,pass Convert it into a Mel spectrum representation, where f is the frequency expressed in Hertz, and m is the corresponding Mel frequency. Pass the obtained Mel spectrum through the following Mel filter bank:

[0051] All filter outputs are logarithmically operated and further discrete cosine transform (DTC) is performed to obtain the Mel-spectrogram.

[0052]

[0053] Again, the processed data is used to extract features using a feature extraction network, and the feature information is fused with the label information.

[0054] Step 3: Construct a category alignment module that fuses feature information and label information.

[0055] Mel-level atlas data obtained after future processing As input (where * Indicates whether the data comes from the source domain or the target domain), uses a feature extractor to project the data of the source domain and the target domain into a shared space; then the projected source domain features and target domain features Input into the classifier. , Further fusion with the projection features, we get and As a fusion feature of different domains, this feature contains both data distribution information and category information. The specific fusion method is as follows:

[0056] in is the fusion feature of different domains, * Indicates whether the data originates from the source domain or the target domain. is the dimension of fusion features of different domains. and The dimensions of are m and n respectively, then the random matrix The size is , The size is ,here and is a random matrix that is sampled only once and fixed during training. represents the Hadamard product, Represents a multilinear mapping method. is the output of a classifier trained by optimizing the following classification loss.

[0057]

[0058] Finally, a domain adaptation model for power equipment fault diagnosis based on category alignment is constructed to solve the problems of insufficient accuracy and poor versatility of traditional deep learning diagnosis models.

[0059] Step 4: Domain adaptation model for power equipment fault diagnosis based on category alignment like Figure 3 As shown in the figure, it is a flow chart of the power equipment fault diagnosis model based on category alignment; the fusion features of different domains are Send to category alignment module , to reduce the difference between the fusion feature distributions of different domains. When the model is well trained, It can be regarded as the classification result of the target data. Specifically, the loss function of the category alignment module is as follows:

[0060] in, is the category alignment loss proposed based on the optimal transportation theory. For The additional constraint is used to ensure the category alignment module Can be optimized normally, is a weight parameter used to balance the model training and proportion of attention.

[0061] ,

[0062] Where D represents the category alignment module, sup represents the supremum, It is the fusion feature of different domains When passing through the category alignment module, the gradient of the return of the category alignment module is constrained to be around 1 as the function is minimized. Therefore, this penalty term can avoid the problem of gradient disappearance or explosion.

[0063] Step 5: Overall model training. Based on the above modules, the overall loss of the model is as follows:

[0064] in and is a weight parameter that controls the balance between losses. The selection range of the traditional weight parameter is between 0.001 and 10. is set to (in is the number of iterations of the current training). Under this setting, the model can pay more attention to the classification loss at the beginning of model training.

[0065] When the model starts training, the preprocessed data information from different domains is sent to the model together, features are extracted through the corresponding feature extraction module, and then fused with the label information through the feature fusion module. The fused features of different domains are then sent to the category alignment module to calculate the loss function corresponding to each loss, and the gradient between the output and input of the overall network model is calculated. The feedforward neural network is backpropagated according to the loss function set by the signature based on the gradient descent method, and the weights and biases in the network are iteratively updated.

[0066] As a preferred solution, an optimization method such as adaptive moment estimation method can also be used to further optimize the model training process. For example, momentum method, RMSProp algorithm, AdaGrad algorithm, AdaDelta algorithm, etc. can be used.

[0067] Step 6: Implement fault diagnosis on sound data from different domains based on the network model.

[0068] After the fault diagnosis model is trained, the sound data collected by the device to be diagnosed can be input according to the model to realize the fault diagnosis of the device corresponding to the data.

[0069] It can be seen that the method for power equipment fault diagnosis based on category alignment provided by the present invention can be used for transformer fault diagnosis technology based on multi-domain signals, including a data fusion module construction strategy based on feature information and label information, and a fault diagnosis domain adaptation model construction strategy based on category alignment. It mainly includes: 1) A data fusion module construction strategy based on feature information and label information is proposed. By organically integrating the feature information of the data with the label information, the traditional data fusion method based on only considering the data information is improved, and a feature fusion model is constructed. This further improves the ability of fusion features in different domains to focus on data category information, solves the problem of inaccurate fault diagnosis caused by traditional deep learning models relying only on data sample training, and improves the diagnostic accuracy of the fault diagnosis model.

[0070] 2) A fault diagnosis domain adaptation model construction strategy based on category alignment. Based on the optimal transportation theory, a fault diagnosis domain adaptation model based on category alignment is constructed to make the distribution alignment of fusion features from different domains from different sources in the projection space more matched, thereby improving the diagnostic accuracy and versatility of the fault diagnosis model, solving the problem that traditional deep learning diagnosis models are difficult to perform fault diagnosis when data sources are different, and enhancing the applicability of the model.

[0071] like Figure 4 As shown, the second object of the present invention is to provide a power equipment fault diagnosis system based on category alignment, comprising: An acquisition module 100 is used to acquire sound data of a device to be diagnosed; The preprocessing module 200 is used to preprocess the sound data of the device to be diagnosed and then convert it into a mel-spectrogram; The diagnosis module 300 is used to input the mel-spectrogram into a pre-trained domain adaptation model for fault diagnosis of electric power equipment constructed based on category alignment, so as to obtain the fault diagnosis result of the equipment to be diagnosed.

[0072] It also includes a pre-training module 400, which is used to train the pre-trained power equipment fault diagnosis domain adaptation model constructed based on category alignment, such as Figure 5 The pre-training module 400 includes: The acquisition submodule 401 is used to acquire the actually collected sound data in different domains; The preprocessing submodule 402 is used to obtain the actually collected sound data in different domains; the sound data in different domains includes a source domain and a target domain; The conversion submodule 403 is used to pre-process the sound data of different domains and then convert them into mel-spectrograms; The extraction submodule 404 is used to extract features from the mel-spectrogram using a feature extraction network, and fuse the feature information with the label information to obtain fused features of different domains; The training submodule 405 is used to construct a domain adaptation model for fault diagnosis of electric equipment based on category alignment based on a neural network, and uses the fusion features of different domains combined with weights as input and the classification results of target data as output to train and obtain the pre-trained domain adaptation model for fault diagnosis of electric equipment based on category alignment.

[0073] The power equipment fault diagnosis system based on category alignment of the present invention is based on the above-mentioned power equipment fault diagnosis method based on category alignment.

[0074] like Figure 6 As shown, the third object of the embodiment of the present invention is to provide an electronic device, including a memory 701, a processor 702, and a computer program stored in the memory 701 and executable on the processor, wherein the processor implements the method for diagnosing faults of power equipment based on category alignment when executing the computer program. It also includes a communication interface 703 and a bus 704.

[0075] A fourth objective of an embodiment of the present invention is to provide a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the above-mentioned method for diagnosing faults of power equipment based on category alignment is implemented.

[0076] A fifth objective of an embodiment of the present invention is to provide a computer program product, wherein the computer program product comprises computer instructions, wherein the computer instructions instruct a computer to execute the above-mentioned method for diagnosing faults of electric power equipment based on category alignment.

[0077] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0078] The present invention provides computer program instructions that can also be loaded onto a computer or other programmable data processing device, so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, so that the instructions executed on the computer or other programmable device provide for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0079] The present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, readable storage media, optical storage, etc.) containing computer-usable program codes.

[0080] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0081] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0082] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the relevant field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents, and any modifications or equivalent replacements that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the present invention.

Claims

1. A method for diagnosing faults of electric power equipment based on category alignment, characterized in that: include: Acquire sound data of the device to be diagnosed; The sound data of the device to be diagnosed is pre-processed and then converted into a mel-spectrogram; The mel-spectrogram is input into the pre-trained domain adaptation model of power equipment fault diagnosis based on category alignment to obtain the fault diagnosis results of the equipment to be diagnosed.

2. A method for diagnosing faults of electric power equipment based on category alignment according to claim 1, characterized in that: The training process of the pre-trained power equipment fault diagnosis domain adaptation model constructed based on category alignment includes: Acquire actually collected sound data in different domains; the sound data in different domains includes a source domain and a target domain; After preprocessing the sound data in different domains, they are converted into mel-spectrograms; The Mel-spectrogram is used for feature extraction using a feature extraction network, and the feature information is fused with the label information to obtain fused features of different domains; Based on a neural network, a domain adaptation model for fault diagnosis of electric power equipment based on category alignment is constructed, and the fusion features combined with weights of the different domains are used as input, and the classification results of the target data are used as output, so as to train the pre-trained domain adaptation model for fault diagnosis of electric power equipment based on category alignment.

3. A method for diagnosing faults of electric power equipment based on category alignment according to claim 2, characterized in that: The preprocessing includes: firstly performing discretization sampling, and then performing normalization processing on the data after the discretization sampling; and uniformly clipping the normalized data to achieve signal scale alignment to obtain processed sound data.

4. The method for diagnosing faults of electric power equipment based on category alignment according to claim 2, characterized in that: The conversion into a mel-spectrogram includes: For the processed sound data, firstly, it is represented as a spectrum graph through fast Fourier transform; Convert the obtained spectrum into Mel spectrum; The obtained Mel spectrum is passed through the Mel filter bank to perform logarithmic operations on all filter outputs; The logarithmic operation result is subjected to discrete cosine transform to obtain the Mel-spectrogram.

5. The method for diagnosing faults of electric power equipment based on category alignment according to claim 2, characterized in that: The Mel-spectrogram is subjected to feature extraction using a feature extraction network, and the feature information is fused with the label information, including: The processed Mel-atlas data is used as input, and the feature extractor is used to project the data of the source domain and the target domain into a shared space; Input the projected source domain features and target domain features into the classifier to obtain the classification results; The classification results are further fused with the projection features to obtain fused features in different domains.

6. The method for diagnosing faults of electric power equipment based on category alignment according to claim 1, characterized in that: The method of constructing a domain adaptation model for fault diagnosis of electric equipment based on category alignment based on a neural network, taking the fusion features of different domains combined with weights as input and the classification results of target data as output, and training the pre-trained domain adaptation model for fault diagnosis of electric equipment based on category alignment, includes: The fused features of different domains are sent to the category alignment module to calculate the loss function corresponding to each loss. Based on the neural network, the gradient between the output and input of the overall neural network is calculated. The feedforward neural network is back-propagated according to the loss function according to the gradient descent method, and the weights and biases in the network are iteratively updated. After iteration, the pre-trained domain adaptation model for fault diagnosis of electric equipment based on category alignment is obtained.

7. A method for diagnosing faults of electric power equipment based on category alignment according to claim 6, characterized in that: The loss function of the category alignment module is: in, is the category alignment loss proposed based on the optimal transportation theory. For The additional constraint is used to ensure the category alignment module Can be optimized normally, is a weight parameter used to balance the model training and The proportion of attention; ， Where D represents the category alignment module, sup represents the supremum, It is the fusion feature of different domains The backpropagated gradients of the class alignment module when passing through the class alignment module.

8. The method for diagnosing faults of electric power equipment based on category alignment according to claim 6, characterized in that: The overall loss of the domain adaptation model for power equipment fault diagnosis based on category alignment is as follows: in, and is a parameter that controls the balance between losses, The selection range of the traditional weight parameter is between 0.001 and 10. is set to , is the number of iterations of the current training.

9. The method for diagnosing faults of electric power equipment based on category alignment according to claim 6, characterized in that: The iterative updating of the weights and biases in the network also includes: The optimization method is used to further optimize the training process.

10. A power equipment fault diagnosis system based on category alignment, characterized in that: include: An acquisition module, used for acquiring sound data of the device to be diagnosed; A preprocessing module is used to preprocess the sound data of the device to be diagnosed and then convert it into a mel-spectrogram; The diagnosis module is used to input the Mel-spectrogram into a pre-trained power equipment fault diagnosis domain adaptation model built based on category alignment to obtain the fault diagnosis result of the equipment to be diagnosed.

11. The power equipment fault diagnosis system based on category alignment according to claim 10, characterized in that: Also included is a pre-training module, the pre-training module comprising: The acquisition submodule is used to obtain the actually collected sound data in different domains; The preprocessing submodule is used to obtain the actually collected sound data in different domains; the sound data in different domains includes the source domain and the target domain; The conversion submodule is used to pre-process the sound data in different domains and then convert them into mel-spectrograms; The extraction submodule is used to extract features from the Mel-spectrogram using a feature extraction network, and fuse the feature information with the label information to obtain fused features of different domains; The training submodule is used to construct a domain adaptation model for fault diagnosis of electric equipment based on category alignment based on a neural network, and uses the fusion features of different domains combined with weights as input and the classification results of target data as output to train the pre-trained domain adaptation model for fault diagnosis of electric equipment based on category alignment.

12. An electronic device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor implements the method for diagnosing faults of power equipment based on category alignment as described in any one of claims 1 to 9 when executing the computer program.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for diagnosing faults of electric power equipment based on category alignment as described in any one of claims 1 to 9 is implemented.

14. A computer program product, comprising computer instructions, characterized in that: The computer instructions instruct the computer to execute the power equipment fault diagnosis method based on category alignment as described in any one of claims 1-9.

Images