CN115219845A

CN115219845A - Power grid fault diagnosis system and method

Info

Publication number: CN115219845A
Application number: CN202210817584.6A
Authority: CN
Inventors: 喻晨
Original assignee: Wenxi Hangzhou Technology Co ltd
Current assignee: Wenxi Hangzhou Technology Co ltd
Priority date: 2022-07-12
Filing date: 2022-07-12
Publication date: 2022-10-21

Abstract

The application relates to the field of intelligent diagnosis, and particularly discloses a power grid fault diagnosis system and a method thereof, which comprehensively utilize electric quantity information and switching value information from the perspective of multi-sensor information fusion to improve the precision of power grid fault diagnosis. In particular, in order to enable the switching value characteristic and the electrical value characteristic to be adaptive in a high-dimensional target domain, the electrical value characteristic is subjected to re-probabilistic rendering, so that in the characteristic extraction process of the neural network, parameters of the neural network are subjected to self-adaptation of mapping to characteristic probability distribution along with iteration, and therefore the adaptability of the neural network is improved in terms of a characteristic extraction task. Therefore, the accuracy of grid fault diagnosis is improved.

Description

Power grid fault diagnosis system and method thereof

Technical Field

The invention relates to the field of intelligent diagnosis, in particular to a power grid fault diagnosis system and method based on deep learning.

Background

Under the background of current electric power big data, the scale of a power grid is continuously enlarged, the interconnection among areas is increasingly tight, uncertain factors influencing the stable operation of a power system are increased, the risk of large-scale power failure accidents caused by the fault of the power grid is increased, and higher requirements are provided for the accuracy, the instantaneity and the like of power grid fault detection. The traditional power grid fault detection method is mostly based on switching value information such as protection and breaker action. However, due to uncertainty factors such as failure, malfunction, and information loss of the protection and the breaker, it is difficult to obtain an accurate result in the fault detection based on the switching amount information alone.

At present, deep learning and neural networks have been widely applied in the fields of computer vision, natural language processing, speech signal processing, and the like. In addition, deep learning and neural networks also exhibit a level close to or even exceeding that of humans in the fields of image classification, object detection, semantic segmentation, text translation and the like.

Deep learning and the development of a neural network provide a new solution and scheme for detecting the power grid fault.

Disclosure of Invention

The present application is proposed to solve the above-mentioned technical problems. The embodiment of the application provides a power grid fault diagnosis system and a method thereof, which comprehensively utilize electric quantity information and switching value information from the perspective of multi-sensor information fusion to improve the precision of power grid fault diagnosis. In particular, in order to enable the switching value characteristic and the electrical value characteristic to be adaptive in a high-dimensional target domain, the electrical value characteristic is subjected to re-probabilistic rendering, so that in the characteristic extraction process of the neural network, parameters of the neural network are subjected to self-adaptation of mapping to characteristic probability distribution along with iteration, and therefore the adaptability of the neural network is improved in terms of a characteristic extraction task. Therefore, the accuracy of grid fault diagnosis is improved.

According to one aspect of the application, an electrical quantity data acquisition unit is provided for acquiring electrical quantity data of a to-be-detected power grid, wherein the electrical quantity data comprises voltage signals and current signals of each line of the to-be-detected power grid;

the current characteristic extraction unit is used for extracting a plurality of current fault characteristics from current signals of each line of the power grid to be detected by utilizing multi-resolution wavelet change, wherein the current fault characteristics comprise current energy fault degrees and current energy distortion degrees;

the current feature coding unit is used for enabling a plurality of current fault features of the current signals of each line of the power grid to be detected to pass through a context encoder comprising an embedded layer so as to obtain a first feature vector corresponding to the current signals of each line of the power grid to be detected, and arranging the first feature vectors of the current signals corresponding to each line of the power grid to be detected according to line sample dimensions so as to obtain a first feature matrix;

the voltage characteristic extraction unit is used for extracting a plurality of voltage fault characteristics from voltage signals of each line of the power grid to be detected by utilizing multi-resolution wavelet change, wherein the plurality of circuit fault characteristics comprise voltage energy fault degrees and voltage energy distortion degrees;

the voltage characteristic encoding unit is used for enabling a plurality of voltage fault characteristics of the voltage signals of all lines of the power grid to be detected to pass through the context encoder comprising the embedded layer so as to obtain second characteristic vectors corresponding to the voltage signals of all lines of the power grid to be detected, and arranging the second characteristic vectors corresponding to the voltage signals of all lines of the power grid to be detected along the dimension of the circuit board sample so as to obtain a second characteristic matrix;

the electric quantity characteristic extraction unit is used for calculating a transfer matrix between the first characteristic matrix and the second characteristic matrix to serve as an electric quantity characteristic matrix, and the transfer matrix is used for representing the responsiveness information of the voltage characteristic of the power grid to be detected and the current characteristic of the power grid to be detected;

the switching value data acquisition unit is used for acquiring a switching value matrix of the power grid to be detected, wherein values of all positions in the switching value matrix are used for representing the on-off states of corresponding switches of corresponding lines;

the switching value coding unit is used for enabling the switching value matrix to pass through a deep convolutional neural network so as to obtain a switching value characteristic matrix;

a repritization unit configured to repritize the electrical quantity feature matrix to obtain a repritized electrical quantity feature matrix, the repritization being performed based on a ratio between a natural exponent function value raised to a power of a feature value at each position in the electrical quantity feature matrix and a weighted sum of natural exponent function values raised to a power of a feature value at each position in the electrical quantity feature matrix;

the fusion unit is used for fusing the re-probability electric quantity characteristic matrix and the switching value characteristic matrix to obtain a classification characteristic matrix; and

and the detection result generation unit is used for enabling the classification characteristic matrix to pass through a classifier to obtain a classification result, and the classification result is used for indicating whether the power grid to be detected has faults or not.

According to another aspect of the present application, there is also provided a grid fault diagnosis method, including:

acquiring electrical quantity data of a to-be-detected power grid, wherein the electrical quantity data comprises voltage signals and current signals of all lines of the to-be-detected power grid;

extracting a plurality of current fault characteristics from current signals of each line of the power grid to be detected by utilizing multi-resolution wavelet change, wherein the current fault characteristics comprise current energy fault degree and current energy distortion degree;

enabling a plurality of current fault characteristics of the current signals of each line of the power grid to be detected to pass through a context encoder comprising an embedded layer to obtain first eigenvectors of the current signals corresponding to each line of the power grid to be detected, and arranging the first eigenvectors of the current signals corresponding to each line of the power grid to be detected according to line sample dimensions to obtain a first characteristic matrix;

extracting a plurality of voltage fault characteristics from voltage signals of each line of the power grid to be detected by utilizing multi-resolution wavelet change, wherein the plurality of circuit fault characteristics comprise voltage energy fault degree and voltage energy distortion degree;

enabling a plurality of voltage fault characteristics of the voltage signals of each line of the power grid to be detected to pass through a context encoder comprising an embedded layer to obtain second characteristic vectors corresponding to the voltage signals of each line of the power grid to be detected, and arranging the second characteristic vectors corresponding to the voltage signals of each line of the power grid to be detected along the dimension of a circuit board sample to obtain a second characteristic matrix;

calculating a transfer matrix between the first characteristic matrix and the second characteristic matrix as an electrical quantity characteristic matrix, wherein the transfer matrix is used for representing responsiveness information of voltage characteristics of the power grid to be detected and current characteristics of the power grid to be detected;

acquiring a switching value matrix of the power grid to be detected, wherein values of all positions in the switching value matrix are used for representing the on-off state of corresponding switches of corresponding lines;

passing the switching value matrix through a deep convolution neural network to obtain a switching value characteristic matrix;

performing a re-probability on the electrical quantity feature matrix to obtain a re-probability electrical quantity feature matrix, wherein the re-probability is performed based on a ratio between a weighted sum of natural exponent function values raised to the power of the feature value of each position in the electrical quantity feature matrix and a weighted sum of natural exponent function values raised to the power of the feature value of each position in the electrical quantity feature matrix;

fusing the re-probability electric quantity characteristic matrix and the switching value characteristic matrix to obtain a classification characteristic matrix; and

and passing the classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the power grid to be detected has faults or not.

Compared with the prior art, the power grid fault diagnosis system and the method thereof comprehensively utilize the electric quantity information and the switching value information from the perspective of multi-sensor information fusion to improve the precision of power grid fault diagnosis. In particular, in order to enable the switching value characteristic and the electrical value characteristic to be adaptive in a high-dimensional target domain, the electrical value characteristic is subjected to re-probability, so that in the characteristic extraction process of the neural network, parameters of the neural network are subjected to self-adaptation of mapping to characteristic probability distribution along with iteration, and the adaptability of the neural network is improved in the aspect of a characteristic extraction task. Therefore, the accuracy of power grid fault diagnosis is improved.

Drawings

The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally indicate like parts or steps.

Fig. 1 illustrates an application scenario diagram of a power grid fault system according to an embodiment of the present application;

FIG. 2 illustrates a block diagram of a grid fault diagnosis system according to an embodiment of the present application;

fig. 3 illustrates a system architecture diagram of a grid fault diagnosis system according to an embodiment of the present application.

Fig. 4 illustrates a block diagram of a current signature encoding unit in a grid fault system according to an embodiment of the application;

fig. 5 illustrates a block diagram of a voltage signature encoding unit in a grid fault system according to an embodiment of the application; and

FIG. 6 illustrates a flow chart of a method of grid fault diagnosis according to an embodiment of the present application;

Detailed Description

Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and that the present application is not limited by the example embodiments described herein.

Overview of a scene

As mentioned above, under the background of current big power data, the scale of the power grid is continuously enlarged, the interconnection between areas is increasingly tight, uncertain factors influencing the stable operation of the power system are increased, the risk of large-scale power failure accidents caused by power grid failure is increased, and higher requirements are provided for the accuracy, real-time performance and the like of power grid failure detection. The traditional power grid fault detection method is mostly based on switching value information such as protection and breaker action. However, due to numerous uncertainty factors such as protection and circuit breaker failure, malfunction, and information loss, it is difficult to obtain accurate results for fault detection based only on switching amount information.

In order to solve the accuracy problem of the current power grid fault detection method, an optimized power grid fault detection scheme is expected.

Through research, the inventor finds that when a power grid fails, firstly, the information of electric quantity such as voltage and current changes, and the information has incomparable advantages in the aspects of accuracy, real-time performance, completeness and the like. Therefore, in the technical scheme of the application, the inventor tries to introduce electrical quantity information in power grid fault diagnosis and combines the switching quantity information to construct a fault diagnosis scheme of multi-source information fusion.

Specifically, in the technical scheme of the application, firstly, electrical quantity data and switching value data are acquired, wherein the electrical quantity data include voltage signals and current signals of each line of a power grid to be detected. Then, using the multi-resolution analysis wavelet variation to extract current fault characteristics of the current signals of each line, wherein the current fault characteristics comprise a current energy fault degree and a current energy distortion degree. Here, the multiresolution analysis of wavelet transform is actually converting the current signal from a time domain signal to a wavelet domain signal based on wavelet transform and extracting the current fault feature in the wavelet domain. It should be understood that there is an association between the current fault signatures, and therefore, in the technical solution of the present application, the current fault signatures are further encoded by a context including an embedded layer, for example, based on a Bert model of a converter, to obtain a first feature vector corresponding to the current fault signature of each line. Then, in a feature space, arranging the first eigenvectors corresponding to the current signals of each line of the to-be-detected power grid according to a line sample dimension to obtain a first feature matrix, wherein the first feature matrix is used for representing an incidence matrix of the overall current fault feature of the to-be-detected power grid. It should be understood that, for the voltage signals of the lines of the power grid to be detected, a second feature matrix for representing the overall voltage fault feature of the power grid to be detected is obtained from the voltage signals of the lines of the power grid to be detected in the same manner.

Further, considering that there is a rich linear correlation between voltage and current, further, a transfer matrix between the first feature matrix and the second feature matrix is calculated as an electrical quantity feature matrix, wherein the transfer matrix is used for representing responsiveness information of the voltage feature of the to-be-detected power grid and the current feature of the to-be-detected power grid.

For the switching value data of the power grid to be detected, firstly, on a data structure, a switching value matrix is constructed to represent the switching value data. Specifically, the switching value matrix may be constructed in such a way that the dimension of a line is taken as a row, and the state data of a plurality of switches on each line is taken as the value of each element in a row vector, where the on state is 1 and the off state is 0; considering that the number of switches of each line is different, the maximum switch value in each line is taken as the length of the row vector, and other elements of other row vectors are supplemented to form a switch value matrix. Then, the convolutional neural network is used to perform explicit spatial coding on the switching value matrix to extract high-dimensional local implicit correlation characteristics of the switching value matrix, that is, high-dimensional implicit correlation characteristics of state correlation between switches of the same line and high-dimensional implicit correlation characteristics of state correlation between switches of different lines, so as to obtain a switching value characteristic matrix.

And then, fusing the electric quantity characteristic matrix and the switching quantity characteristic matrix and using a classifier to diagnose the power grid fault. However, since the switching value feature matrix is a feature target domain mapped to the source domain of the 0 and 1 states of the switch, the electrical value feature matrix is re-probabilistic so as to be adapted to the feature probability distribution of the target domain of the switching value feature matrix, specifically:

m _i,j is the eigenvalue of each position of the electrical quantity eigenvalue matrix.

The re-probability can promote the neural network to map the distribution characteristics of the initial data into a probabilistic characteristic space by performing probabilistic interpretation on the electric quantity characteristic matrix, so that the parameters of the neural network are self-adapted to the mapping of the characteristic probability distribution along with iteration in the characteristic extraction process of the neural network, and the adaptability of the neural network is improved in the aspect of a characteristic extraction task. Therefore, the accuracy of grid fault diagnosis is improved.

Based on this, the present application proposes a power grid fault diagnosis system, which includes: the electric quantity data acquisition unit is used for acquiring electric quantity data of a to-be-detected power grid, and the electric quantity data comprises voltage signals and current signals of all lines of the to-be-detected power grid; the current characteristic extraction unit is used for extracting a plurality of current fault characteristics from current signals of each line of the power grid to be detected by utilizing multi-resolution wavelet change, wherein the current fault characteristics comprise current energy fault degrees and current energy distortion degrees; the current feature coding unit is used for enabling a plurality of current fault features of the current signals of each line of the power grid to be detected to pass through a context coder comprising an embedded layer so as to obtain a first feature vector of the current signals corresponding to each line of the power grid to be detected, and arranging the first feature vectors of the current signals corresponding to each line of the power grid to be detected according to line sample dimensions so as to obtain a first feature matrix; the voltage characteristic extraction unit is used for extracting a plurality of voltage fault characteristics from voltage signals of each line of the power grid to be detected by utilizing multi-resolution wavelet change, wherein the plurality of circuit fault characteristics comprise voltage energy fault degrees and voltage energy distortion degrees; the voltage characteristic encoding unit is used for enabling a plurality of voltage fault characteristics of the voltage signals of all lines of the power grid to be detected to pass through the context encoder comprising the embedded layer so as to obtain second characteristic vectors corresponding to the voltage signals of all lines of the power grid to be detected, and arranging the second characteristic vectors corresponding to the voltage signals of all lines of the power grid to be detected along the dimension of the circuit board sample so as to obtain a second characteristic matrix; the electric quantity characteristic extraction unit is used for calculating a transfer matrix between the first characteristic matrix and the second characteristic matrix as an electric quantity characteristic matrix, and the transfer matrix is used for representing responsiveness information of the voltage characteristic of the power grid to be detected and the current characteristic of the power grid to be detected; the switching value data acquisition unit is used for acquiring a switching value matrix of the power grid to be detected, wherein values of all positions in the switching value matrix are used for representing the on-off state of corresponding switches of corresponding lines; the switching value coding unit is used for enabling the switching value matrix to pass through a deep convolutional neural network so as to obtain a switching value characteristic matrix; a reprioritization unit configured to reprioritize the electrical quantity feature matrix to obtain a reprioritized electrical quantity feature matrix, the reprioritization being performed based on a ratio between a natural exponent function value raised to a power of a feature value at each position in the electrical quantity feature matrix and a weighted sum of natural exponent function values raised to a power of a feature value at each position in the electrical quantity feature matrix; the fusion unit is used for fusing the re-probability electric quantity characteristic matrix and the switching value characteristic matrix to obtain a classification characteristic matrix; and the diagnosis result generating unit is used for enabling the classification characteristic matrix to pass through a classifier to obtain a classification result, and the classification result is used for indicating whether the power grid to be detected has faults or not.

Fig. 1 illustrates an application scenario diagram of a power grid fault diagnosis system according to an embodiment of the present application. As shown in fig. 1, in this application scenario, first, a current meter (e.g., a as illustrated in fig. 1) and a voltage meter (e.g., V as illustrated in fig. 1) are used to obtain a voltage signal and a current signal of each line in the power grid to be detected from the power grid to be detected, and obtain the switch state information of each line in the power grid to be detected from the power grid to be detected. Then, the voltage signals and the current signals of each line in the power grid to be detected and the switch state information of each line in the power grid to be detected are input into a server (for example, S shown in fig. 1) deployed with a power grid fault diagnosis algorithm, where the server can process the input information with the power grid fault diagnosis algorithm to obtain a power grid fault diagnosis result.

Having described the basic principles of the present application, various non-limiting embodiments of the present application will now be described with reference to the accompanying drawings.

Exemplary System

Fig. 2 illustrates a block diagram of a grid fault diagnosis system according to an embodiment of the present application.

As shown in fig. 2, the grid fault diagnosis system 100 according to the embodiment of the present application includes: the electrical quantity data acquisition unit 110, the current feature extraction unit 120, the current feature encoding unit 130, the voltage feature extraction unit 140, the voltage feature encoding unit 150, the electrical quantity feature extraction unit 160, the switching value data acquisition unit 170, the switching value encoding unit 180, the repritization unit 190, the fusion unit 200, and the diagnosis result generation unit 210.

The electrical quantity data acquiring unit 110 is configured to acquire electrical quantity data of a to-be-detected power grid, where the electrical quantity data includes voltage signals and current signals of each line of the to-be-detected power grid; the current feature extraction unit 120 is configured to extract a plurality of current fault features from current signals of each line of the power grid to be detected by using multi-resolution wavelet transformation, where the plurality of current fault features include a current energy fault degree and a current energy distortion degree. The current feature encoding unit 130 is configured to pass multiple current fault features of the current signal of each line of the power grid to be detected through a context encoder including an embedded layer to obtain a first feature vector corresponding to the current signal of each line of the power grid to be detected, and arrange the first feature vector corresponding to the current signal of each line of the power grid to be detected according to a line sample dimension to obtain a first feature matrix. The voltage feature extraction unit 140 is configured to extract a plurality of voltage fault features from the voltage signals of each line of the power grid to be detected by using multi-resolution wavelet transformation, where the plurality of voltage fault features include a voltage energy fault degree and a voltage energy distortion degree. The voltage feature encoding unit 150 is configured to obtain a second feature vector corresponding to the voltage signal of each line of the power grid to be detected by passing the multiple voltage fault features of the voltage signal of each line of the power grid to be detected through the context encoder including the embedded layer, and arrange the second feature vector corresponding to the voltage signal of each line of the power grid to be detected along a circuit board sample dimension to obtain a second feature matrix. The electrical quantity feature extraction unit 160 is configured to calculate a transfer matrix between the first feature matrix and the second feature matrix as an electrical quantity feature matrix, where the transfer matrix is used to represent responsiveness information of a voltage feature of the to-be-detected power grid and a current feature of the to-be-detected power grid. The switching value data obtaining unit 170 is configured to obtain a switching value matrix of the power grid to be detected, where a value at each position in the switching value matrix is used to indicate an on-off state of a corresponding switch of a corresponding line. The switching value encoding unit 180 is configured to pass the switching value matrix through a deep convolutional neural network to obtain a switching value feature matrix. The reptile unit 190 is configured to reptile the electrical quantity feature matrix to obtain a reptile electrical quantity feature matrix, where the reptile is performed based on a ratio between a natural exponent function value raised to a power of a feature value at each position in the electrical quantity feature matrix and a weighted sum of the natural exponent function values raised to a power of the feature value at each position in the electrical quantity feature matrix. The fusion unit 200 is configured to fuse the re-probability electric quantity feature matrix and the switching value feature matrix to obtain a classification feature matrix; the diagnostic result generating unit 210 is configured to pass the classification feature matrix through a classifier to obtain a classification result, where the classification result is used to indicate whether the power grid to be detected has a fault.

Fig. 3 illustrates a system architecture diagram of the grid fault diagnosis system 100 according to an embodiment of the present application. As shown in fig. 3, in the system architecture of the power grid fault diagnosis system 100, first, the electrical quantity data of the power grid to be detected is obtained through the electrical quantity data obtaining unit 110, where the electrical quantity data includes voltage signals and current signals of each line of the power grid to be detected. Then, the electrical quantity data obtaining unit 110 transmits the electrical quantity data of the to-be-detected power grid to the current feature extraction unit 120, wherein the current feature extraction unit 120 extracts a plurality of current fault features from the current signals of each line of the to-be-detected power grid respectively by using multi-resolution wavelet transform, and the plurality of current fault features include a current energy fault degree and a current energy distortion degree. Next, the current signature encoding unit 130 encodes a plurality of current fault signatures of the current signals of the lines of the to-be-detected power grid by using a context encoder including an embedded layer to obtain first eigenvectors of the current signals corresponding to the lines of the to-be-detected power grid, and arranges the first eigenvectors of the current signals corresponding to the lines of the to-be-detected power grid according to a line sample dimension to obtain a first feature matrix. Meanwhile, the electrical quantity data acquisition unit 110 transmits the electrical quantity data of the power grid to be detected to the voltage feature extraction unit 140, wherein the voltage feature extraction unit 140 extracts a plurality of voltage fault features from the voltage signals of each line of the power grid to be detected respectively by using multi-resolution wavelet transformation, and the plurality of circuit fault features include a voltage energy fault degree and a voltage energy distortion degree. Next, the voltage feature encoding unit 150 encodes a plurality of voltage fault features of the voltage signals of each line of the to-be-detected power grid by using the context encoder including the embedded layer to obtain second feature vectors corresponding to the voltage signals of each line of the to-be-detected power grid, and arranges the second feature vectors corresponding to the voltage signals of each line of the to-be-detected power grid along a circuit board sample dimension to obtain a second feature matrix. After receiving the first feature matrix output by the current feature encoding unit 130 and the second feature matrix output by the voltage feature encoding unit 150, the electrical quantity feature extraction unit 160 calculates a transfer matrix between the first feature matrix and the second feature matrix as an electrical quantity feature matrix, where the transfer matrix is used to represent responsiveness information of the voltage feature of the power grid to be detected and the current feature of the power grid to be detected. Then, the switching value data obtaining unit 170 receives a switching value matrix of the power grid to be detected, where values of each position in the switching value matrix are used to represent an on-off state of a corresponding switch of a corresponding line, and the switching value coding unit 180 is used to code the switching value matrix with a deep convolutional neural network to obtain a switching value feature matrix. Further, the reptile unit 190 performs a reptile on the electrical quantity feature matrix to obtain a reptile electrical quantity feature matrix, the reptile being performed based on a ratio between a natural exponent function value raised to a power of a feature value at each position in the electrical quantity feature matrix and a weighted sum of the natural exponent function values raised to a power of the feature value at each position in the electrical quantity feature matrix. Next, the fusion unit 200 fuses the re-probability electric quantity feature matrix and the switching value feature matrix to obtain a classification feature matrix. Finally, the diagnostic result generating unit 210 uses a classifier to obtain a classification result from the classification feature matrix, where the classification result is used to indicate whether the power grid to be detected has a fault.

Specifically, in an operation process of the power grid fault diagnosis system 100, the electrical quantity data obtaining unit 110 obtains electrical quantity data of a power grid to be detected, where the electrical quantity data includes voltage signals and current signals of each line of the power grid to be detected. That is, the electrical quantity data acquisition unit 110 receives electrical quantity data of the power grid to be detected, which is acquired by sensors (including a current sensor and a voltage sensor) deployed in the power grid to be detected, where the electrical quantity data includes voltage signals and current signals of each line of the power grid to be detected.

As described above, the inventor of the present application finds that, when a power grid fails, first, electrical quantity information such as voltage and current changes, and has incomparable advantages in terms of accuracy, real-time performance, completeness and the like. Therefore, in the power grid fault diagnosis system 100 according to the present application, the electrical quantity data of the power grid to be detected is used as the source data for power grid fault diagnosis.

Further, during the operation of the power grid fault diagnosis system 100, the current feature extraction unit 120 extracts a plurality of current fault features from the current signals of the lines of the power grid to be detected respectively by using multi-resolution wavelet transformation, where the plurality of current fault features include a current energy fault degree and a current energy distortion degree. Those of ordinary skill in the art will appreciate that multi-resolution wavelet transform is a commonly used technical authority in the field of current signal processing that is capable of wavelet transforming a current signal to transform the current signal from a time domain signal into a wavelet domain to derive the plurality of current fault signatures.

Further, in the operation process of the power grid fault diagnosis system 100, the current feature encoding unit 130 obtains a first feature vector of the current signal corresponding to each line of the power grid to be detected by passing a plurality of current fault features of the current signal of each line of the power grid to be detected through a context encoder including an embedded layer, and arranges the first feature vector of the current signal corresponding to each line of the power grid to be detected according to a line sample dimension to obtain a first feature matrix. Here, the multiresolution analysis wavelet transform is actually converting the current signal from a time domain signal to a wavelet domain signal based on the wavelet transform and extracting the current fault feature in the wavelet domain. It should be understood that there is an association between the current fault features, and therefore, in the technical solution of the present application, the plurality of current fault features are further encoded through a context including an embedded layer to extract a high-dimensional implicit association feature between the current fault features of the respective lines.

Specifically, in one example of the present application, the context encoder includes an embedded layer and a Transformer (Transformer) -based Bert model. Wherein the embedded layer is configured to vectorially convert the current fault signature to convert a current fault signature into an embedded vector, and in particular embodiments, the vector converter of the embedded layer may be configured based on a knowledge graph, so that the current fault signature may be combined with knowledge graph information of the circuit fault to improve the information richness of the current fault signature. On the other hand, the current fault characteristics can be converted into structured data which is more convenient for a computer to operate through vectorization.

The role of the converter-based Bert model is to globally context-based semantically encode the sequence of embedded vectors (i.e., globally context-based semantically encode each embedded vector in the sequence of embedded vectors based on the sequence of embedded vectors) to obtain a plurality of current fault feature vectors corresponding to the sequence of embedded vectors, wherein one current fault feature vector in the plurality of current fault feature vectors corresponds to one current fault feature. In particular, a global semantic feature representation of each of the plurality of current fault features relative to the plurality of current fault features may be extracted by globally based context semantic coding the sequence of embedded vectors.

And then, cascading the plurality of current fault feature vectors to obtain first feature vectors corresponding to the current signals of the lines of the power grid to be detected, wherein the cascading is used for fully retaining feature representation of the plurality of current fault features in a high-dimensional semantic feature space without causing information loss.

Further, the first eigenvectors of the current signals corresponding to the lines of the power grid to be detected are arranged according to the dimension of the line sample to obtain the first characteristic matrix. That is, the signature representations of the current signals of the individual lines are integrated into a two-dimensional data structure, i.e. the first signature matrix, in accordance with the line sample dimension.

Fig. 4 illustrates a block diagram of the current signature encoding unit 130 in the grid fault system according to an embodiment of the present application. As shown in fig. 4, the current feature encoding unit 130 of the grid fault system according to the embodiment of the present application includes: a first embedded vector conversion unit 131, configured to convert each current fault feature of the multiple current fault features of the current signal of each line of the to-be-detected power grid into an embedded vector to obtain an embedded vector sequence; a first global semantic coding subunit 132, configured to pass the sequence of embedded vectors through a converter-based Bert model to obtain a plurality of current fault feature vectors; the first cascade subunit 133 is configured to cascade the multiple current fault feature vectors to obtain first feature vectors of the current signals corresponding to each line of the to-be-detected power grid; and a first matrix construction subunit 134, configured to arrange the first eigenvectors corresponding to the current signals of each line of the to-be-detected power grid according to a line sample dimension to obtain the first eigenvector.

Accordingly, during the operation of the power grid fault diagnosis system 100, the voltage feature extraction unit 140 and the voltage feature encoding unit 150 encode the voltage signals of the lines in the same manner to obtain a second feature matrix corresponding to the voltage signals of the power grid to be detected. Specifically, the voltage feature extraction unit 140 extracts a plurality of voltage fault features from the voltage signals of each line of the power grid to be detected by using multi-resolution wavelet transform, where the plurality of voltage fault features include a voltage energy fault degree and a voltage energy distortion degree. The voltage feature encoding unit 150 obtains a second feature vector corresponding to the voltage signal of each line of the power grid to be detected by passing through the context encoder including the embedded layer for the multiple voltage fault features of the voltage signal of each line of the power grid to be detected, and arranges the second feature vector corresponding to the voltage signal of each line of the power grid to be detected along the dimension of the circuit board sample to obtain a second feature matrix.

Specifically, in one example of the present application, the context encoder includes an embedded layer and a Transformer (Transformer) -based Bert model. Wherein the embedded layer is configured to vectorially convert the voltage fault signature to convert a voltage fault signature into an embedded vector, and in particular embodiments, the vector converter of the embedded layer may be configured based on a knowledge graph, so that the voltage fault signature may be combined with knowledge graph information of the circuit fault to improve the information richness of the voltage fault signature. On the other hand, the voltage fault characteristics can be converted into structured data which is more convenient for a computer to operate through vectorization.

The role of the converter-based Bert model is to globally context-based semantically encode the sequence of embedded vectors (i.e., globally context-based semantically encode each embedded vector in the sequence of embedded vectors based on the sequence of embedded vectors) to obtain a plurality of voltage fault feature vectors corresponding to the sequence of embedded vectors, wherein one voltage fault feature vector in the plurality of voltage fault feature vectors corresponds to one voltage fault feature. In particular, a global semantic feature representation of each of the plurality of voltage fault features relative to the plurality of voltage fault features may be extracted by globally based context semantic encoding the sequence of embedded vectors.

And then, cascading the voltage fault feature vectors to obtain second feature vectors corresponding to the voltage signals of all lines of the power grid to be detected, wherein the cascading is used for fully retaining feature representation of the voltage fault features in a high-dimensional semantic feature space without information loss.

Further, the second eigenvectors corresponding to the voltage signals of each line of the power grid to be detected are arranged according to the dimension of the line sample to obtain the second eigenvector matrix. That is, the signature representations of the voltage signals of the individual lines are integrated into a two-dimensional data structure, i.e. the second signature matrix, in accordance with the line sample dimension.

Fig. 5 illustrates a block diagram of a voltage signature encoding unit 150 in a grid fault system according to an embodiment of the application. As shown in fig. 5, the voltage characteristic encoding unit 150 of the grid fault system according to the embodiment of the present application includes: a second embedded vector conversion unit 151, configured to convert each voltage fault feature of the multiple voltage fault features of the voltage signal of each line of the power grid to be detected into an embedded vector to obtain a sequence of embedded vectors; a second global semantic encoding subunit 152 for passing the sequence of embedded vectors through the converter-based Bert model to obtain a plurality of voltage fault feature vectors; the second cascade subunit 153 is configured to cascade the multiple voltage fault feature vectors to obtain second feature vectors of the voltage signals corresponding to the lines of the power grid to be detected; and a second matrix constructing subunit 154, configured to arrange the second eigenvectors corresponding to the voltage signals of each line of the to-be-detected power grid according to a line sample dimension to obtain the second eigenvector.

Further, in the operation process of the power grid fault diagnosis system 100, considering that there is a rich linear correlation between voltage and current, a transfer matrix between the first feature matrix and the second feature matrix is calculated as an electrical quantity feature matrix, where the transfer matrix is used to represent responsiveness information of the voltage feature of the power grid to be detected and the current feature of the power grid to be detected.

Correspondingly, the electrical quantity feature extraction unit 160 calculates a transfer matrix between the first feature matrix and the second feature matrix as an electrical quantity feature matrix, where the transfer matrix is used to represent responsiveness information of the voltage feature of the to-be-detected power grid and the current feature of the to-be-detected power grid. Specifically, the electrical quantity characteristic extraction unit 160 is further configured to calculate a transfer matrix between the first characteristic matrix and the second characteristic matrix as the electrical quantity characteristic matrix according to the following formula; wherein the formula is: s = T × M, wherein T denotes the transition matrix, T denotes the first feature matrix, and S denotes the second feature matrix.

By calculating a transfer matrix between the first feature matrix and the second feature matrix, the feature value of each position in the transfer matrix is the quotient of the feature values of the corresponding two positions in the first feature matrix and the second feature matrix, so that on one hand, the responsiveness relationship between voltage and current is fully utilized, and on the other hand, the transfer matrix establishes the association between the current feature of each position and the voltage feature of each position. That is, in the high-dimensional feature space, a data representation containing current feature information, voltage feature information, and voltage-current associated feature information is constructed.

Further, during the operation of the power grid fault diagnosis system 100, the switching value data obtaining unit 170 obtains a switching value matrix of the power grid to be detected, where a value of each position in the switching value matrix is used to indicate an open/close state of a corresponding switch of a corresponding line. As described above, in the technical solution of the present application, the inventor of the present application tries to introduce electrical quantity information in power grid fault diagnosis, and combines the switching quantity information to construct a fault diagnosis scheme of multi-source information fusion.

Specifically, for the switching value data of the power grid to be detected, firstly, on the data structure, a switching value matrix is constructed to represent the switching value data. Specifically, the switching value matrix may be constructed in such a way that the dimension of a line is taken as a row, and the state data of a plurality of switches on each line is taken as the value of each element in a row vector, where open is 1 and close is 0; considering that the number of switches of each line is different, the maximum switch value in each line is taken as the length of a row vector, and other elements of other row vectors are supplemented to form a switch value matrix.

Then, the convolutional neural network is used to perform explicit spatial coding on the switching value matrix to extract high-dimensional local implicit correlation characteristics of the switching value matrix, that is, high-dimensional implicit correlation characteristics of state correlation between switches of the same line and high-dimensional implicit correlation characteristics of state correlation between switches of different lines, so as to obtain a switching value characteristic matrix. That is, during the operation of the grid fault diagnosis system 100, the switching value encoding unit 180 passes the switching value matrix through a deep convolutional neural network to obtain a switching value feature matrix.

In this embodiment of the present application, the encoding process of the deep convolutional neural network is as follows: each layer of the deep convolutional neural network respectively carries out the following operations on input data in the forward transmission of the layer: firstly, carrying out convolution processing on the input data to obtain a convolution characteristic diagram; then, performing global mean pooling along channel dimensions on the convolution feature map to obtain a pooled feature map; and carrying out nonlinear activation processing on the pooled feature map to obtain an activated feature map. Thus, the output of the last layer of the deep convolutional neural network is the switching value feature matrix.

And then, fusing the switching quantity characteristic matrix and the electrical quantity characteristic matrix to diagnose the power grid fault. However, since the switching value characteristic matrix is a characteristic target domain to which the 0 and 1 states of the switch are mapped as the source domain, the electrical value characteristic matrix is re-probabilistic so as to be adapted to the characteristic probability distribution of the target domain of the switching value characteristic matrix, specifically:

The re-probability can promote the neural network to map the distribution characteristics of the initial data into the probabilistic feature space by performing probabilistic interpretation on the electrical quantity feature matrix, so that the parameters of the neural network are self-adapted to the mapping of the feature probability distribution along with iteration in the feature extraction process of the neural network, and the adaptability of the neural network is improved in the aspect of a feature extraction task. Therefore, the accuracy of grid fault diagnosis is improved.

Accordingly, during the operation of the grid fault diagnosis system 100, the reptile unit 190 performs a reptile on the electrical quantity feature matrix to obtain a reptile electrical quantity feature matrix, wherein the reptile is performed based on a ratio between a natural exponent function value raised to a power of a feature value at each position in the electrical quantity feature matrix and a weighted sum of the natural exponent function values raised to a power of a feature value at each position in the electrical quantity feature matrix.

Specifically, the repritization unit 190 performs repritization on the electrical quantity feature matrix according to the following formula to obtain the repritized electrical quantity feature matrix;

wherein the formula is:

wherein m is _i,j Is a characteristic value, m ', of each position of the electrical quantity characteristic matrix' _i,j A characteristic value, exp (m), representing each position of the re-probabilistic electric quantity characteristic matrix _i,j ) Representing calculating a natural exponent function value raised to the eigenvalue of each position of the electrical quantity eigenvalue matrix.

Then, in the operation process of the power grid fault diagnosis system 100, the fusion unit 200 fuses the probabilistic electric quantity feature matrix and the switching value feature matrix to obtain a classification feature matrix, and it should be understood that the classification feature matrix includes the electric quantity features and the switching value features of the power grid to be detected.

Specifically, the fusion unit 200 fuses the re-probability electric quantity feature matrix and the switching value feature matrix according to the following formula to obtain the classification feature matrix;

wherein the formula is:

M _c ＝αM _a +βM _g

wherein M is _c For the classification feature matrix, M _a For the probabilistic electric quantity feature matrix, M _g For the switching value feature matrix, "+" indicates the addition of elements at corresponding positions of the re-probabilistic electric quantity feature matrix and the switching value feature matrix, and α and β are weighting parameters for controlling the balance between the re-probabilistic electric quantity feature matrix and the switching value feature matrix in the classification feature matrix.

Further, in an operation process of the power grid fault diagnosis system 100, the diagnosis result generating unit 210 obtains a classification result by passing the classification feature matrix through a classifier, where the classification result is used to indicate whether the power grid to be detected has a fault.

Specifically, in the embodiment of the present application, the classifier processes the classification feature matrix according to the following formula to obtain the classification result;

wherein the formula is: softmax { (W) _n ,B _n ):…:(W ₁ ,B ₁ ) L Project (F) }, where Project (F) denotes the projection of the classification feature matrix as a vector, W ₁ To W _n As a weight matrix for each fully connected layer, B ₁ To B _n A bias matrix representing the layers of the fully connected layer.

That is to say, in the embodiment of the present application, the classifier first performs full-concatenation coding on the classification feature matrix through a plurality of full-concatenation layers to obtain a classification feature vector, where the full-concatenation coding can fully utilize information of each position in the classification feature matrix and reduce the dimension of the classification feature matrix into a one-dimensional classification feature vector. And then, passing the classification feature vector through a Softmax classification function to obtain probability values of the classification feature vector belonging to each classification label respectively. And finally, determining the classification label corresponding to the person with the maximum probability value as the classification result.

In summary, the grid fault diagnosis system 100 according to the embodiment of the present application is illustrated, which improves the accuracy of grid fault diagnosis by comprehensively using the electric quantity information and the switching value information from the perspective of multi-sensor information fusion. In particular, in order to enable the switching value characteristic and the electrical value characteristic to be adaptive in a high-dimensional target domain, the electrical value characteristic is subjected to re-probability, so that in the characteristic extraction process of the neural network, parameters of the neural network are subjected to self-adaptation of mapping to characteristic probability distribution along with iteration, and the adaptability of the neural network is improved in the aspect of a characteristic extraction task. Therefore, the accuracy of grid fault diagnosis is improved.

As described above, the grid fault diagnosis system 100 according to the embodiment of the present application may be implemented in various terminal devices, such as a server having a circuit fault automatic diagnosis function. In one example, the grid fault diagnosis system 100 according to the embodiment of the present application may be integrated into a terminal device as one software module and/or hardware module. For example, the grid fault diagnosis system 100 may be a software module in the operating system of the terminal device, or may be an application developed for the terminal device; of course, the grid fault diagnosis system 100 may also be one of many hardware modules of the terminal device.

Alternatively, in another example, the grid fault diagnosis system 100 and the terminal device may also be separate devices, and the grid fault diagnosis system 100 may be connected to the terminal device through a wired and/or wireless network and transmit the interactive information according to an agreed data format.

Exemplary method

According to another aspect of the application, a power grid fault diagnosis method is further provided.

As shown in fig. 6, the grid fault diagnosis method according to the embodiment of the present application includes the steps of: s110, acquiring electrical quantity data of a to-be-detected power grid, wherein the electrical quantity data comprises voltage signals and current signals of each line of the to-be-detected power grid; s120, extracting a plurality of current fault characteristics from current signals of each line of the power grid to be detected by utilizing multi-resolution wavelet change, wherein the current fault characteristics comprise current energy fault degree and current energy distortion degree; s130, enabling a plurality of current fault characteristics of the current signals of each line of the power grid to be detected to pass through a context encoder comprising an embedded layer to obtain first eigenvectors of the current signals corresponding to each line of the power grid to be detected, and arranging the first eigenvectors of the current signals corresponding to each line of the power grid to be detected according to line sample dimensions to obtain a first characteristic matrix; s140, extracting a plurality of voltage fault characteristics from voltage signals of each line of the power grid to be detected by utilizing multi-resolution wavelet change, wherein the plurality of circuit fault characteristics comprise voltage energy fault degree and voltage energy distortion degree; s150, enabling a plurality of voltage fault characteristics of the voltage signals of each line of the power grid to be detected to pass through the context encoder comprising the embedded layer to obtain second characteristic vectors corresponding to the voltage signals of each line of the power grid to be detected, and arranging the second characteristic vectors corresponding to the voltage signals of each line of the power grid to be detected along the sample dimension of the circuit board to obtain a second characteristic matrix; s160, calculating a transfer matrix between the first characteristic matrix and the second characteristic matrix as an electrical quantity characteristic matrix, wherein the transfer matrix is used for representing responsiveness information of the voltage characteristic of the power grid to be detected and the current characteristic of the power grid to be detected; s170, acquiring a switching value matrix of the power grid to be detected, wherein values of all positions in the switching value matrix are used for representing the on-off state of corresponding switches of corresponding lines; s180, passing the switching value matrix through a deep convolution neural network to obtain a switching value characteristic matrix; s190, performing a repritization on the electrical quantity feature matrix to obtain a repritization electrical quantity feature matrix, where the repritization is performed based on a ratio between a natural exponent function value raised by a power of a feature value of each position in the electrical quantity feature matrix and a weighted sum of the natural exponent function values raised by the power of the feature value of each position in the electrical quantity feature matrix; s200, fusing the re-probability electric quantity characteristic matrix and the switching value characteristic matrix to obtain a classification characteristic matrix; and S210, passing the classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the power grid to be detected has faults or not.

In a specific example, in the power grid fault diagnosis method, the step S190 includes: performing re-probability on the electrical quantity characteristic matrix according to the following formula to obtain a re-probability electrical quantity characteristic matrix;

wherein the formula is:

In a specific example, in the power grid fault diagnosis method, the step S210 includes: the classifier processes the classification characteristic matrix according to the following formula to obtain the classification result;

In summary, the power grid fault diagnosis method according to the embodiment of the present application is elucidated, which improves the accuracy of power grid fault diagnosis by comprehensively using electric quantity information and switching value information from the perspective of multi-sensor information fusion. In particular, in order to enable the switching value characteristic and the electrical value characteristic to be adaptive in a high-dimensional target domain, the electrical value characteristic is subjected to re-probabilistic rendering, so that in the characteristic extraction process of the neural network, parameters of the neural network are subjected to self-adaptation of mapping to characteristic probability distribution along with iteration, and therefore the adaptability of the neural network is improved in terms of a characteristic extraction task. Therefore, the accuracy of power grid fault diagnosis is improved.

Claims

1. A grid fault detection system, comprising:

the electric quantity data acquisition unit is used for acquiring electric quantity data of a to-be-detected power grid, and the electric quantity data comprises voltage signals and current signals of all lines of the to-be-detected power grid;

the current feature coding unit is used for enabling a plurality of current fault features of the current signals of each line of the power grid to be detected to pass through a context coder comprising an embedded layer so as to obtain a first feature vector of the current signals corresponding to each line of the power grid to be detected, and arranging the first feature vectors of the current signals corresponding to each line of the power grid to be detected according to line sample dimensions so as to obtain a first feature matrix;

the voltage characteristic coding unit is used for enabling a plurality of voltage fault characteristics of the voltage signals of each line of the power grid to be detected to pass through the context encoder comprising the embedded layer so as to obtain second characteristic vectors corresponding to the voltage signals of each line of the power grid to be detected, and arranging the second characteristic vectors corresponding to the voltage signals of each line of the power grid to be detected along the dimension of the circuit board sample so as to obtain a second characteristic matrix;

the electric quantity characteristic extraction unit is used for calculating a transfer matrix between the first characteristic matrix and the second characteristic matrix as an electric quantity characteristic matrix, and the transfer matrix is used for representing responsiveness information of the voltage characteristic of the power grid to be detected and the current characteristic of the power grid to be detected;

the switching value data acquisition unit is used for acquiring a switching value matrix of the power grid to be detected, wherein values of all positions in the switching value matrix are used for representing the on-off state of corresponding switches of corresponding lines;

the switching value coding unit is used for enabling the switching value matrix to pass through a deep convolution neural network so as to obtain a switching value characteristic matrix;

2. The grid fault detection system of claim 1, wherein the current signature encoding unit comprises:

the first embedded vector conversion unit is used for converting each current fault characteristic in a plurality of current fault characteristics of the current signal of each line of the power grid to be detected into an embedded vector so as to obtain a sequence of the embedded vector;

a first global semantic coding subunit, configured to pass the sequence of embedded vectors through a Bert model based on a converter to obtain a plurality of current fault feature vectors;

the first cascade subunit is used for cascading the current fault characteristic vectors to obtain first characteristic vectors of the current signals corresponding to the lines of the power grid to be detected; and

and the first matrix construction subunit is used for arranging the first eigenvectors of the current signals corresponding to each line of the power grid to be detected according to the line sample dimension to obtain the first eigenvector.

3. The grid fault detection system of claim 2, wherein the voltage signature encoding unit comprises:

the second embedded vector conversion unit is used for converting each voltage fault characteristic in a plurality of voltage fault characteristics of the voltage signal of each line of the power grid to be detected into an embedded vector so as to obtain a sequence of the embedded vector;

a second global semantic coding subunit, configured to pass the sequence of embedded vectors through the converter-based Bert model to obtain a plurality of voltage fault feature vectors;

the second cascade subunit is used for cascading the voltage fault characteristic vectors to obtain second characteristic vectors of the voltage signals corresponding to the lines of the power grid to be detected; and

and the second matrix construction subunit is used for arranging the second eigenvectors corresponding to the voltage signals of each line of the power grid to be detected according to the line sample dimension to obtain the second eigenvector matrix.

4. The grid fault detection system according to claim 3, wherein the electrical quantity feature extraction unit is further configured to calculate a transition matrix between the first feature matrix and the second feature matrix as the electrical quantity feature matrix in the following formula;

wherein the formula is: s = T × M, wherein T denotes the transition matrix, T denotes the first feature matrix, and S denotes the second feature matrix.

5. The grid fault detection system according to claim 4, wherein in the switch coding unit, each layer of the deep convolutional neural network performs, in layer forward pass, input data respectively:

performing convolution processing on the input data to obtain a convolution characteristic diagram;

performing global mean pooling along channel dimensions on the convolution feature map to obtain a pooled feature map; and

carrying out nonlinear activation processing on the pooled feature map to obtain an activated feature map;

and the output of the last layer of the deep convolutional neural network is the switching value characteristic matrix.

6. The grid fault diagnosis system according to claim 5, wherein the re-probability unit is further configured to re-probability the electrical quantity feature matrix to obtain the re-probability electrical quantity feature matrix according to the following formula;

wherein the formula is:

wherein m is _i,j Is a characteristic value, m ', of each position of the electrical quantity characteristic matrix' _i,j Characteristic value, exp (m), representing each position of the re-probabilistic electric quantity characteristic matrix _i,j ) Representing calculation with characteristic matrix of said electrical quantityThe characteristic value of each position is a natural exponential function value of a power.

7. The grid fault diagnosis system according to claim 6, wherein the fusion unit is further configured to fuse the re-probabilistic electric quantity feature matrix and the switching quantity feature matrix to obtain the classification feature matrix according to the following formula;

wherein the formula is:

M _c ＝αM _a +βM _g

wherein M is _c For the classification feature matrix, M _a For the probabilistic electric quantity feature matrix, M _g The sign + represents the addition of the elements at the corresponding positions of the re-probability electric quantity feature matrix and the switching quantity feature matrix, and α and β are weighting parameters for controlling the balance between the re-probability electric quantity feature matrix and the switching quantity feature matrix in the classification feature matrix.

8. The power grid fault diagnosis system according to claim 7, wherein in the diagnosis result generation unit, the classifier processes the classification feature matrix to obtain the classification result according to the following formula;

9. A power grid fault diagnosis method is characterized by comprising the following steps:

extracting a plurality of current fault characteristics from current signals of each line of the power grid to be detected by using multi-resolution wavelet change, wherein the current fault characteristics comprise current energy fault degree and current energy distortion degree;

enabling a plurality of voltage fault characteristics of the voltage signals of each line of the power grid to be detected to pass through a context encoder comprising an embedded layer so as to obtain second characteristic vectors corresponding to the voltage signals of each line of the power grid to be detected, and arranging the second characteristic vectors corresponding to the voltage signals of each line of the power grid to be detected along the dimension of a circuit board sample so as to obtain a second characteristic matrix;

performing a re-probability on the electrical quantity feature matrix to obtain a re-probability electrical quantity feature matrix, the re-probability being performed based on a ratio between a natural exponent function value raised to a power of a feature value of each position in the electrical quantity feature matrix and a weighted sum of the natural exponent function values raised to a power of the feature value of each position in the electrical quantity feature matrix;

and enabling the classification characteristic matrix to pass through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the power grid to be detected has faults or not.

10. The power grid fault diagnosis method according to claim 9, wherein passing a plurality of current fault signatures of the current signal of each line of the power grid to be detected through a context encoder comprising an embedded layer to obtain a first signature vector corresponding to the current signal of each line of the power grid to be detected comprises:

converting each current fault feature in a plurality of current fault features of the current signal of each line of the power grid to be detected into an embedded vector to obtain a sequence of the embedded vector;

passing the sequence of embedded vectors through a Bert model based converter to obtain a plurality of current fault feature vectors; and

and cascading the plurality of current fault characteristic vectors to obtain first characteristic vectors corresponding to the current signals of all lines of the power grid to be detected.