Small sample multi-mode electronic voltage transformer fault diagnosis method and system

Abstract

The invention discloses a fault diagnosis method and system for a small-sample multi-modal electronic voltage transformer, and relates to the field of fault diagnosis, and the method comprises the following steps: obtaining multi-modal data and related data; constructing a fault diagnosis model which comprises a feature extraction unit used for modal feature extraction to obtain corresponding modal features; connecting and converting the secondary voltage signal and all modal features into a two-dimensional image, obtaining a support set, inputting the support set into an embedded network unit, and obtaining an embedded support set and a local descriptor of the embedded support set; obtaining to-be-detected multi-modal data and inputting the to-be-detected multi-modal data into the fault diagnosis model to obtain an embedded query set and a local descriptor of the embedded query set; and calculating a normal similarity score and an abnormal similarity score, and comparing the values of the normal similarity score and the abnormal similarity score to obtain a fault diagnosis result.

Description

Technical Field

[0001] This application relates to the field of fault diagnosis, and mainly relates to a fault diagnosis method and system for small-sample multi-modal electronic voltage transformers. Background Art

[0002] An electronic voltage transformer is a device used to replace traditional voltage transformers and is widely used in the measurement, protection, and monitoring of power systems. Different from traditional voltage transformers, electronic voltage transformers use digital and intelligent technologies for voltage sampling, processing, and transmission. Their output is usually a digital signal, with higher accuracy, stability, and anti-interference ability. Due to the complexity of power systems and the high integration of electronic devices, the fault diagnosis of electronic voltage transformers poses high challenges.

[0003] Currently, the fault diagnosis methods for electronic voltage transformers mainly include fault diagnosis based on signal analysis. However, this method has low accuracy for complex non-linear faults and the situation of multiple faults superimposed; while fault diagnosis based on artificial intelligence has high requirements for the quality and quantity of training data, and the adaptability of the model is poor when the system environment changes greatly; traditional methods based on signal analysis have a decline in diagnostic accuracy due to limited feature extraction ability in non-linear fault and multi-fault superimposed scenarios, and insufficient detection sensitivity for weak faults.

[0004] The Chinese patent application with the publication number CN119249341A discloses "A Fault Diagnosis Method Based on Multi-modal Multi-scale Feature Fusion", specifically discloses "collecting signal sources when faults occur at different positions of an aero-engine bearing through sensors, and the signal sources include but are not limited to: vibration signals, acoustic signals, temperature, humidity, current; performing unified denoising processing on the vibration signals to obtain an initial one-dimensional signal X0; performing one-dimensional feature extraction on the initial one-dimensional signal X0 through an MCNN module; performing image conversion on the initial one-dimensional signal X0 to obtain an initial two-dimensional image T1, and performing two-dimensional feature extraction on the initial two-dimensional image T1; fusing the extracted one-dimensional features and two-dimensional features to generate multi-modal feature H1; inputting the fused multi-modal feature H1 into an LSTM network for fault diagnosis and classification". Although this method collects multi-source signals, subsequent processing only takes the vibration signal as the core, and other modal data are not explicitly fused and utilized, resulting in missing feature dimensions and being difficult to comprehensively reflect the multi-physical field coupling characteristics of bearing faults, affecting diagnostic accuracy; in addition, in this method, the one-dimensional and two-dimensional features are only fused through simple splicing, lacking cross-modal correlation analysis, resulting in the complementary nature of different modal information not being fully exploited, and the discriminant ability of the fused features being limited; at the same time, although the LSTM network of this method is good at processing time-series data, its classification ability for the fused high-dimensional features is weak, and overfitting or underfitting problems are likely to occur. Summary of the Invention

[0005] To solve the above problems existing in the prior art, the present application provides a small-sample multi-modal electronic voltage transformer fault diagnosis method and system.

[0006] The technical solution of the present application is as follows: On the one hand, the present invention proposes a small-sample multi-modal electronic voltage transformer fault diagnosis method, and the method includes: Obtain multi-modal data and related data of the electronic voltage transformer, where the multi-modal data includes grid frequency, secondary voltage signal and primary current signal; the related data includes power data; Construct a fault diagnosis model, including a feature extraction unit and an embedding network unit; the feature extraction unit is used to extract modal features from the multi-modal data to obtain corresponding modal features; connect the secondary voltage signal and the modal features and convert them into a two-dimensional image to obtain a support set; input the support set into the embedding network unit to obtain an embedded support set and a local descriptor of the embedded support set; divide the local descriptor into a local descriptor of the normal support set and a local descriptor of the abnormal support set based on the embedded support set; wherein, the modal feature extraction is specifically to input the grid frequency into an improved long short-term memory network (LSTM network), perform fast Fourier frequency domain analysis on the secondary voltage signal, and perform overshoot calculation on the primary current signal; Obtain the multi-modal data to be detected of the target electronic voltage transformer, and input the multi-modal data to be detected into the fault diagnosis model to obtain an embedded query set and a local descriptor of the embedded query set; Calculate the similarity between the local descriptor of the embedded query set and the local descriptor of the normal support set to obtain a normal similarity score; calculate the similarity between the local descriptor of the embedded query set and the local descriptor of the abnormal support set to obtain an abnormal similarity score; compare the values of the normal similarity score and the abnormal similarity score to obtain the fault diagnosis result of the target detected electronic voltage transformer.

[0007] Preferably, the method further includes data cleaning of the multi-modal data and related data, and the data cleaning includes processing missing values, outliers and data format standardization.

[0008] Preferably, input the grid frequency into the improved LSTM network, which is expressed by the formula: ; ; ; ; ; ; In the formula, represents the output of the forget gate at time represents a logical function; represents the weight of the forget gate; represents the hidden state at time represents the grid frequency at time represents the primary current signal at time represents the bias of the forget gate; represents the output of the input gate at time represents the weight of the input gate; represents the bias of the input gate; represents the candidate cell state at time represents the weight of the candidate cell state; represents the bias of the candidate cell state; represents the first activation function; represents the cell state at time represents the cell state at time represents the output gate at time represents the weight of the output gate; represents the bias of the output gate; represents the power at time represents the modal feature of the grid frequency; represents time.

[0009] Preferably, perform fast Fourier transform frequency domain analysis on the secondary voltage signal, which is expressed by the formula: ; ; In the formula, represents the fast Fourier function; represents a preset frequency; represents the imaginary part symbol; represents the secondary voltage signal at time represents the real part function; represents the imaginary part function; Select the largest as the modal feature of the secondary voltage signal, expressed as 。

[0010] Preferably, perform overshoot calculation on the primary current signal, which is expressed by the formula: , ; And ; ; ; In the formula, represents the average value of the primary current signal; represents the maximum value of the local primary current signal; represents the total number of primary current signals; represents the index value of the th primary current signal; represents the th primary current signal; represents the index value of the maximum value of the th local primary current signal; represents the maximum value of the th local primary current signal; represents the index value of the overshoot value of the maximum value of the th local primary current signal;

[0011] Preferably, connect all modal features and the secondary voltage signal and transform them into a two-dimensional image to obtain a support set, which is expressed by the formula: , ; ; In the formula, represents the two-dimensional image matrix, that is, the support set; represents the pixel at the th row and the th column of the two-dimensional image matrix; represents the dimension of the two-dimensional image matrix; represents the connection sequence; represents the th connection sequence value; represents the th index value of the connection sequence; represents the th secondary voltage signal value; Indicates the number of secondary voltage signals; Indicates the row index value of the two-dimensional image matrix; Indicates the row index value of the two-dimensional image matrix; Indicates the index value of the

[0012] Preferably, the support set is input into the embedding network unit, and the embedding network unit includes a Gaussian noise layer, a layer normalization layer, a convolutional layer, and a Dropout layer. Specifically: The Gaussian noise layer receives the support set and adds noise, which is expressed by the formula: ; In the formula, represents the noise support set; represents the Gaussian distribution for selecting noise values in ; represents the variance of the Gaussian distribution; The layer normalization layer receives the noise support set and performs layer normalization to obtain the embedded support set, which is expressed by the formula: ; ; ; In the formula, represents the mean of the noise support set; represents the variance of the noise support set; represents the number of two-dimensional images in the noise support set; represents the index value of the th two-dimensional image in the noise support set; represents the th two-dimensional image in the noise support set; represents the embedded support set; represents a preset constant; represents a preset scaling parameter; The convolutional layer convolves the embedded support set and further inputs it into the Dropout layer for regularization, which is expressed by the formula: ; ⊙ ; In the formula, represents the th feature map, that is, the output of the convolutional layer; represents the index value of the Represents the total number of neurons; Represents the second activation function; Represents the weight of the connection of the neuron in the Represents the bias of the neuron in the Represents the th two-dimensional image in the embedded support set; Represents the index value of the convolutional layer in the Represents the number of convolutional layers; Represents the th regularized feature map; Represents the preset Bernoulli mask matrix; ⊙ represents element-wise multiplication; The local descriptor of the embedded support set is obtained based on the regularized feature map, which is expressed by the formula: ; ; In the formula, represents the local descriptor of the embedded support set; represents the th local descriptor of the embedded support set; represents the weight of the fully connected layer; represents the bias of the fully connected layer; represents the th eigenvector corresponding to the regularized feature map; The local descriptor of the embedded support set is divided into the local descriptor of the normal support set and the local descriptor of the abnormal support set .

[0013] Preferably, the normal similarity score and the abnormal similarity score are calculated, specifically: The embedded query set is represented as , where represents the embedded query set; represents the local descriptor of the embedded query set; represents the th local descriptor of the embedded query set; represents the number of local descriptors of the embedded query set; Traverse the local descriptors in the embedded query set, search for the nearest neighbors in the local descriptors of the normal support set and the local descriptors of the abnormal support set, and calculate the normal similarity score and the abnormal similarity score, which are expressed by the formula: ; ; In the formula, represents the normal similarity score; represents that the weight is similarity metric function; represents the value of the local descriptor of the th embedded query set in the th dimension; represents the value of the local descriptor of the th normal support set in the th dimension; represents the abnormal similarity score; represents the th local descriptor of the abnormal support set in the th dimension; represents the dimension of the embedded query set; represents the index value of the th dimension of the embedded query set; represents a preset constant; If , then the target electronic voltage transformer is normal; otherwise, the target electronic voltage transformer is abnormal.

[0014] Preferably, the method further includes continuously monitoring modal features, collecting diagnostic experiences of each modality, and establishing a modal fault association knowledge base; evaluating the diagnostic result using a preset performance evaluation index to obtain an evaluation result; and optimizing the fault diagnosis process by combining the modal fault association knowledge base and the evaluation result.

[0015] On the other hand, the present invention also proposes a small-sample multi-modal electronic voltage transformer fault diagnosis system, which includes a data acquisition module, a local descriptor generation module, a fault diagnosis module, and a result output module, where: The data acquisition module is used to acquire multi-modal data and related data of the electronic voltage transformer. The multi-modal data includes grid frequency, secondary voltage signal, and primary current signal; the related data includes power data; and the multi-modal data and related data are transmitted to the local descriptor generation module; The local descriptor generation module is used to construct a fault diagnosis model, including a feature extraction unit and an embedding network unit; the feature extraction unit is used to extract modal features from multi-modal data to obtain corresponding modal features; connect the secondary voltage signal and the modal features and convert them into a two-dimensional image to obtain a support set; input the support set into the embedding network unit to obtain an embedded support set and a local descriptor of the embedded support set; divide the local descriptor into a local descriptor of the normal support set and a local descriptor of the abnormal support set based on the embedded support set; wherein, the modal feature extraction is specifically to input the power grid frequency into an improved long short-term memory network (LSTM network), perform fast Fourier frequency domain analysis on the secondary voltage signal, and perform overshoot calculation on the primary current signal. The fault diagnosis module is used to obtain the multi-modal data to be detected of the target electronic voltage transformer, input the multi-modal data to be detected into the fault diagnosis model, and obtain an embedded query set and a local descriptor of the embedded query set. Calculate the similarity between the local descriptor of the embedded query set and the local descriptor of the normal support set to obtain a normal similarity score; calculate the similarity between the local descriptor of the embedded query set and the local descriptor of the abnormal support set to obtain an abnormal similarity score; compare the values of the normal similarity score and the abnormal similarity score to obtain the fault diagnosis result of the target detection electronic voltage transformer. The result output module is used to display the fault diagnosis result of the target detection electronic voltage transformer.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1) The present invention provides a small-sample multi-modal electronic voltage transformer fault diagnosis method and system. By connecting the power grid frequency, secondary voltage signal and primary current signal, the multi-dimensional characterization ability of fault features is improved, and the recognition accuracy of multi-physical field coupling faults inside the transformer is enhanced. 2) The present invention provides a small-sample multi-modal electronic voltage transformer fault diagnosis method and system. The improved LSTM network improves the ability to capture time-varying features in the power grid frequency signal and enhances the detection sensitivity; the embedding network unit enhances the anti-interference ability of the model through the Gaussian noise layer and Dropout layer, and combines layer normalization technology to improve the diagnostic robustness and the adaptability of the model. 3) The present invention provides a small-sample multi-modal electronic voltage transformer fault diagnosis method and system. Through support set division and local descriptor similarity measurement, the fault diagnosis is transformed into a traceable feature matching process, which improves the transparency of decision-making basis and enhances the generalization ability of the model for rare fault types. Description of the Drawings

[0017] Figure 1 It is the flowchart of the method of the embodiment of the present invention. Specific embodiments

[0018] The following describes the specific embodiments of the present invention to facilitate the understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention defined and determined by the appended claims, these changes are obvious, and all inventions created using the concept of the present invention are within the scope of protection.

[0019] The present invention provides the following technical solutions: a small-sample multi-modal electronic voltage transformer fault diagnosis method and system.

[0020] Embodiment 1: Specifically refer to Figure 1 , this embodiment provides a small-sample multi-modal electronic voltage transformer fault diagnosis method, and the specific steps include: S1. Obtain the multi-modal data and relevant data of the electronic voltage transformer, where the multi-modal data includes grid frequency, secondary voltage signal, and primary current signal; the relevant data includes power data; The method further includes data cleaning of the multi-modal data, and the data cleaning includes processing missing values, outliers, and data format standardization; S2. Construct a fault diagnosis model, including a feature extraction unit and an embedding network unit; S21. The feature extraction unit is used to extract modal features from the multi-modal data to obtain corresponding modal features; among them, the modal feature extraction is specifically to input the grid frequency into an improved long short-term memory network (LSTM) network, perform fast Fourier frequency domain analysis on the secondary voltage signal, and perform overshoot calculation on the primary current signal; S211. Input the grid frequency into the improved LSTM network, and the improved LSTM network is specifically to import the primary current signal and power data, which is expressed by the formula: ; ; ; ; ; ; In the formula, represents the output of the forgetting gate at time represents the logical function; represents the weight of the forget gate; represents the hidden state at time represents the grid frequency at time represents the primary current signal at time represents the bias of the forget gate; represents the output of the input gate at time represents the weight of the input gate; represents the bias of the input gate; represents the candidate cell state at time represents the weight of the candidate cell state; represents the bias of the candidate cell state; represents the first activation function; represents the cell state at time represents the cell state at time represents the output gate at time represents the weight of the output gate; represents the bias of the output gate; represents the power at time represents the modal feature of the grid frequency; represents time S212. Perform fast Fourier transform frequency domain analysis on the secondary voltage signal, which is expressed by the formula: ; ; In the formula, represents the fast Fourier transform function; represents the preset frequency; represents the imaginary part symbol; represents the secondary voltage signal at time represents the real part function; represents the imaginary part function; Select the largest as the modal feature of the secondary voltage signal, which is expressed as ; S213. Perform overshoot calculation on the primary current signal, which is expressed by the formula: , ; and ; ; ; In the formula, represents the average value of the primary current signal; represents the maximum value of the local primary current signal; represents the total number of primary current signals; represents the index value of the primary current signal; represents the acquisition period of the primary current signal; represents the maximum value of the local primary current signal; represents the index value of the maximum value of the local primary current signal; represents the overshoot value of the maximum value of the local primary current signal; represents the modal feature of the primary current signal; S22. Connect the secondary voltage signal and all modal features and convert them into a two-dimensional image to obtain a support set, which is expressed by the formula: , ; ; In the formula, represents the two-dimensional image matrix, that is, the support set; represents the pixel at the th row and th column of the two-dimensional image matrix; represents the dimension of the two-dimensional image matrix; represents the connection sequence; represents the value of the connection sequence; represents the index value of the connection sequence; represents the value of the secondary voltage signal; represents the number of secondary voltage signals; represents the index value of the th row of the two-dimensional image matrix; represents the index value of the th row of the two-dimensional image matrix; represents the index value of the value of the S23. Input the support set into the embedding network unit to obtain an embedded support set and a local descriptor of the embedded support set. The embedding network unit includes a Gaussian noise layer, a layer normalization layer, a convolutional layer, and a Dropout layer. The Gaussian noise layer receives the support set and adds noise, which is expressed by the formula: ; In the formula, represents the noisy support set; represents the Gaussian distribution for selecting noise values in ; represents the variance of the Gaussian distribution; The layer normalization layer receives the noisy support set and performs layer normalization, which is expressed by the formula: ; ; ; In the formula, represents the mean of the noisy support set; represents the variance of the noisy support set; represents the number of two-dimensional images in the noisy support set; represents the index value of the th two-dimensional image; represents the th two-dimensional image of the noisy support set; represents the embedded support set; represents a preset constant; represents a preset scaling parameter; represents a preset translation parameter; The convolutional layer convolves the embedded support set and further inputs it into the Dropout layer for regularization, which is expressed by the formula: ; ⊙ ; In the formula, represents the th feature map, i.e., the output of the convolutional layer; represents the index value of the th neuron; represents the total number of neurons; represents the second activation function; represents the th layer and the th neuron's connection weight; represents the th layer and the th neuron's bias; Denote the th two-dimensional image in the embedded support set; Denote the index value of the th convolutional layer; Denote the th regularized feature map; Denote the preset Bernoulli mask matrix; ⊙ denotes element-wise multiplication; Obtain the local descriptor of the embedded support set based on the regularized feature map, which is expressed by the formula: ; ; In the formula, denotes the local descriptor of the embedded support set; Denote the th local descriptor of the embedded support set; Denote the weight of the fully connected layer; Denote the bias of the fully connected layer; Denote the th eigenvector corresponding to the regularized feature map; Divide the local descriptor of the embedded support set into the local descriptor of the normal support set and the local descriptor of the abnormal support set ; S3. Obtain the multi-modal data to be detected of the target electronic voltage transformer, and input the multi-modal data to be detected into the fault diagnosis model to obtain the embedded query set and the local descriptor of the embedded query set; S4. Calculate the similarity between the local descriptor of the embedded query set and the local descriptor of the normal support set to obtain the normal similarity score; calculate the similarity between the local descriptor of the embedded query set and the local descriptor of the abnormal support set to obtain the abnormal similarity score; The embedded query set is expressed as , where denotes the embedded query set; denotes the local descriptor of the embedded query set; Denote the th local descriptor of the embedded query set; Denote the number of local descriptors of the embedded query set; Traverse the local descriptors in the embedded query set, search for the nearest neighbor in the local descriptors of the normal support set and the local descriptors of the abnormal support set, and calculate the normal similarity score and the abnormal similarity score, which are expressed by the formula: ; ; In the formula, represents the normal similarity score; represents that the weight is similarity metric function; represents the value of the local descriptor of the -th embedded query set in the -th dimension; represents the value of the local descriptor of the -th normal support set in the -th dimension; represents the abnormal similarity score; represents the -th local descriptor of the abnormal support set in the -th dimension; represents the dimension of the embedded query set; represents the index value of the -th embedded query set dimension; S5. Compare the scores of the normal similarity score and the abnormal similarity score, and obtain the fault diagnosis result of the target detection electronic voltage transformer; If , then the target electronic voltage transformer is normal; otherwise, the target electronic voltage transformer is abnormal; S6. The method further includes continuously monitoring the modal features, collecting the diagnostic experiences of each modality, and establishing a modal fault association knowledge base; using the preset performance evaluation index to evaluate the diagnostic result to obtain an evaluation result; and optimizing the fault diagnosis process by combining the modal fault association knowledge base and the evaluation result.

[0021] Embodiment 2: This embodiment provides a small-sample multi-modal electronic voltage transformer fault diagnosis system, which includes a data acquisition module, a local descriptor generation module, a fault diagnosis module, and a result output module, where: The data acquisition module is used to acquire the multi-modal data and related data of the electronic voltage transformer. The multi-modal data includes grid frequency, secondary voltage signal, and primary current signal; the related data includes power data; and the multi-modal data and related data are transmitted to the local descriptor generation module; The local descriptor generation module is used to construct a fault diagnosis model, including a feature extraction unit and an embedding network unit; the feature extraction unit is used to extract modal features from multimodal data to obtain corresponding modal features; connect the secondary voltage signal and the modal features and convert them into a two-dimensional image to obtain a support set; input the support set into the embedding network unit to obtain an embedded support set and a local descriptor of the embedded support set; divide the local descriptor into a local descriptor of a normal support set and a local descriptor of an abnormal support set based on the embedded support set; wherein, the modal feature extraction is specifically to input the grid frequency into an improved long short-term memory network (LSTM network), perform fast Fourier frequency domain analysis on the secondary voltage signal, and perform overshoot calculation on the primary current signal. The fault diagnosis module is used to obtain the to-be-detected multimodal data of the target electronic voltage transformer, and input the to-be-detected multimodal data into the fault diagnosis model to obtain an embedded query set and a local descriptor of the embedded query set. Calculate the similarity between the local descriptor of the embedded query set and the local descriptor of the normal support set to obtain a normal similarity score; calculate the similarity between the local descriptor of the embedded query set and the local descriptor of the abnormal support set to obtain an abnormal similarity score; compare the values of the normal similarity score and the abnormal similarity score to obtain the fault diagnosis result of the target detection electronic voltage transformer. The result output module is used to display the fault diagnosis result of the target detection electronic voltage transformer.

[0022] The above are only the embodiments of the present invention, and do not limit the patent scope of the present invention accordingly. Any equivalent structure or equivalent process transformation made by using the content of the specification and drawings of the present invention, or directly or indirectly applied in other related technical fields, shall be equally included in the patent protection scope of the present invention.

Claims

1. A small sample multi-mode electronic voltage transformer fault diagnosis method, characterized in that: The method comprises: Acquire multimodal data and related data of the electronic voltage transformer, wherein the multimodal data includes grid frequency, secondary voltage signal and primary current signal; and the related data includes power data; Construct a fault diagnosis model, including a feature extraction unit and an embedding network unit; the feature extraction unit is used to extract modal features from multimodal data to obtain corresponding modal features; the secondary voltage signal and the modal features are connected and converted into a two-dimensional image to obtain a support set; the support set is input into the embedding network unit to obtain an embedded support set and a local descriptor of the embedded support set; based on the embedded support set, the local descriptor is divided into a local descriptor of a normal support set and a local descriptor of an abnormal support set; wherein the modal feature extraction is specifically to input the power grid frequency into an improved long short-term memory network LSTM network, perform fast Fourier frequency domain analysis on the secondary voltage signal, and perform overshoot calculation on the primary current signal; Acquire multimodal data to be detected of a target electronic voltage transformer, input the multimodal data to be detected into the fault diagnosis model, and obtain an embedded query set and a local descriptor of the embedded query set; The similarity between the local descriptors of the embedded query set and the local descriptors of the normal support set is calculated to obtain the normal similarity score; the similarity between the local descriptors of the embedded query set and the local descriptors of the abnormal support set is calculated to obtain the abnormal similarity score; the scores of the normal similarity score and the abnormal similarity score are compared to obtain the fault diagnosis result of the target detected electronic voltage transformer.

2. A small sample multi-mode electronic voltage transformer fault diagnosis method according to claim 1, characterized in that: The method also includes performing data cleaning on the multimodal data and related data, wherein the data cleaning includes processing missing values, outliers and unifying data formats.

3. A small sample multi-mode electronic voltage transformer fault diagnosis method according to claim 1, characterized in that: The grid frequency is input into the improved LSTM network and is expressed as follows: ； ； ； ； ； ； In the formula, express Always forget the output of the gate; represents a logical function; Represents the weight of the forget gate; express The hidden state of the moment; express The grid frequency at the moment; express The primary current signal at the moment; Represents the bias of the forget gate; express The output of the input gate at any time; Represents the weight of the input gate; represents the bias of the input gate; express Candidate cell states at time instant; represents the weight of the candidate cell state; represents the bias of the candidate cell state; represents the first activation function; express The cell state at a given moment; express The cell state at a given moment; express Output gate at the moment; Represents the weight of the output gate; represents the bias of the output gate; express Power at the moment; Represents the modal characteristics of the power grid frequency; Indicates time.

4. A small sample multi-mode electronic voltage transformer fault diagnosis method according to claim 3, characterized in that: The secondary voltage signal is subjected to fast Fourier frequency domain analysis, which is expressed as follows: ； ； In the formula, represents the fast Fourier function; Indicates the preset frequency; Indicates the sign of the imaginary part; express Secondary voltage signal at time ; represents the real function; represents the imaginary part function; Choose the largest As the modal characteristic of the secondary voltage signal, it is expressed as .

5. A small sample multi-mode electronic voltage transformer fault diagnosis method according to claim 4, characterized in that: The overshoot calculation is performed on the primary current signal, which is expressed as follows: ； and ; ； ； In the formula, Indicates the average value of the primary current signal; Indicates the maximum value of the local primary current signal; Indicates the total number of primary current signals; Indicates The index value of a primary current signal; Indicates A primary current signal; Indicates The maximum value of a local primary current signal; Indicates The index value of the maximum value of the local primary current signal; Indicates the total number of maximum values ​​of local primary current signals; Indicates The overshoot value of the maximum value of the local primary current signal; Represents the modal characteristics of the primary current signal.

6. A small sample multi-mode electronic voltage transformer fault diagnosis method according to claim 5, characterized in that: The secondary voltage signal and all modal features are connected and converted into a two-dimensional image to obtain the support set, which is expressed as: ， ； ； In the formula, represents the two-dimensional image matrix, i.e., the support set; Represents a two-dimensional image matrix Line Column pixels; Represents the dimension of the two-dimensional image matrix; Represents a connection sequence; Indicates A connection sequence value; Indicates The index value of the connection sequence; Indicates Secondary voltage signal value; Indicates the number of secondary voltage signals; Represents a two-dimensional image matrix The index value of the row; Represents a two-dimensional image matrix The index value of the row; Indicates The index value of the secondary voltage signal value.

7. A small sample multi-mode electronic voltage transformer fault diagnosis method according to claim 6, characterized in that: The support set is input into the embedding network unit, which includes a Gaussian noise layer, a layer normalization layer, a convolution layer and a Dropout layer, specifically: The Gaussian noise layer receives the support set and adds noise, which is expressed as: ； In the formula, represents the noise support set; Indicated in Select a Gaussian distribution of noise values ​​in ; represents the variance of the Gaussian distribution; The layer normalization layer receives the noise support set and performs layer normalization to obtain an embedded support set, which is expressed as: ； ； ； In the formula, represents the mean of the noise support set; represents the variance of the noise support set; The number of two-dimensional images representing the noise support set; Indicates The index value of a two-dimensional image; represents the noise support set A two-dimensional image of represents the embedding support set; Indicates a preset constant; Indicates the preset scaling parameters; Indicates the preset translation parameters; The convolution layer convolves the embedded support set and further inputs the Dropout layer for regularization, which can be expressed as: ； ⊙ ； In the formula, Indicates feature maps, which are the outputs of the convolutional layers; Indicates The index value of the neuron; represents the total number of neurons; represents the second activation function; Indicates Tier The connection weights of neurons; Indicates Tier The bias of each neuron; Represents the embedding support set A two-dimensional image; Indicates The index value of the convolutional layer; Indicates the number of convolutional layers; Indicates A regularized feature map; represents the preset Bernoulli mask matrix; ⊙ represents element-by-element multiplication; Based on the regularized feature map, the local descriptor of the embedded support set is obtained, which is expressed as: ； ； In the formula, local descriptors representing the embedded support set; Indicates local descriptors embedded in the support set; represents the weight of the fully connected layer; represents the bias of the fully connected layer; Indicates The feature vector corresponding to the regularized feature map; Divide the local descriptors of the embedded support set into local descriptors of the normal support set and local descriptors of the exception support set .

8. A small sample multi-mode electronic voltage transformer fault diagnosis method according to claim 7, characterized in that: Calculate the normal similarity score and abnormal similarity score, specifically: The embedded query set is represented as ,in Represents an embedded query set; A local descriptor representing the embedded query set; Indicates A local descriptor that embeds the query set; Represents the number of local descriptors embedded in the query set; Traverse the local descriptors in the embedded query set, search for the nearest neighbors in the local descriptors of the normal support set and the local descriptors of the abnormal support set, calculate the normal similarity score and the abnormal similarity score, which can be expressed as: ； ； In the formula, represents the normal similarity score; Indicates the weight is Similarity measurement function of ; Indicates The local descriptor of the embedded query set is The value of the dimension; Indicates The local descriptor of the normal support set is The value of the dimension; represents the anomaly similarity score; Indicates The local descriptor of the abnormal support set is The value of the dimension; Represents the dimension of the embedded query set; Indicates The index value of the embedded query set dimension; Indicates a preset constant; like , the target electronic voltage transformer is normal; otherwise, the target electronic voltage transformer is abnormal.

9. A small sample multi-mode electronic voltage transformer fault diagnosis method according to claim 1, characterized in that: The method also includes continuously monitoring the modal characteristics, collecting diagnostic experience of each modality, and establishing a modal fault association knowledge base; evaluating the diagnostic results using a preset performance evaluation index to obtain an evaluation result; The fault diagnosis process is optimized by combining the modal fault association knowledge base and the evaluation results.

10. A small sample multi-mode electronic voltage transformer fault diagnosis system, characterized in that: The system includes a data acquisition module, a local descriptor generation module, a fault diagnosis module and a result output module, wherein: The data acquisition module is used to acquire multimodal data and related data of the electronic voltage transformer, wherein the multimodal data includes grid frequency, secondary voltage signal and primary current signal; the related data includes power data; and transmit the multimodal data and related data to the local descriptor generation module; The local descriptor generation module is used to construct a fault diagnosis model, including a feature extraction unit and an embedding network unit; the feature extraction unit is used to extract modal features from multimodal data to obtain corresponding modal features; the secondary voltage signal and the modal features are connected and converted into a two-dimensional image to obtain a support set; the support set is input into the embedding network unit to obtain an embedded support set and a local descriptor of the embedded support set; based on the embedded support set, the local descriptor is divided into a local descriptor of a normal support set and a local descriptor of an abnormal support set; wherein the modal feature extraction is specifically to input the power grid frequency into an improved long short-term memory network LSTM network, perform fast Fourier frequency domain analysis on the secondary voltage signal, and perform overshoot calculation on the primary current signal; The fault diagnosis module is used to obtain multi-modal data to be detected of the target electronic voltage transformer, input the multi-modal data to be detected into the fault diagnosis model, and obtain an embedded query set and a local descriptor of the embedded query set; Calculate the similarity between the local descriptors of the embedded query set and the local descriptors of the normal support set to obtain a normal similarity score; calculate the similarity between the local descriptors of the embedded query set and the local descriptors of the abnormal support set to obtain an abnormal similarity score; compare the scores of the normal similarity score and the abnormal similarity score to obtain the fault diagnosis result of the target detection electronic voltage transformer; The result output module is used to display the fault diagnosis result of the target detection electronic voltage transformer.

Images