Transformer excitation inrush current identification method and device based on Gram summation angle field

Abstract

The present invention discloses a transformer excitation inrush current identification method and device based on Gram summation angle field, obtains the three-phase differential current of the power transformer, and obtains a three-dimensional color map; takes the three-dimensional color map as input, and uses the excitation inrush current single classification model based on SVDD to identify whether the transformer is in the excitation inrush current working condition; wherein the excitation inrush current single classification model based on SVDD is based on the three-dimensional color map under various working conditions of the transformer, and is constructed by learning and training using the LeNet‑5 deep convolutional neural network. This technical solution can realize the identification of excitation inrush current by artificial intelligence methods, and solve the problem of scarce transformer fault samples.

Description

Technical Field

[0001] The present invention belongs to the technical field of transformer relay protection, and more specifically refers to a transformer excitation inrush current identification method and device based on Gramian angular summation field (GASF). Background Art

[0002] Transformers usually use differential protection based on Kirchhoff's current law as the main protection, which can achieve good results in most cases. When the transformer is closed under no-load or the voltage is restored after a fault, the transformer voltage suddenly rises from zero or a very small value to the operating voltage. During this transient process of voltage rise, due to the influence of residual magnetism, the transformer may enter the saturation zone, thereby generating a large excitation current, which can reach 8 to 10 times the rated current. The excitation circuit is equivalent to the fault branch of the internal fault of the transformer. When the excitation surge current is generated, the current flows into the differential relay, forming an unbalanced current, causing the transformer to erroneously operate in the absence of faults. It can be seen that accurately identifying the transformer excitation surge current and improving the performance of differential protection to avoid the excitation surge current are important means to reduce the false operation rate and improve the reliability of the transformer.

[0003] The most commonly used methods for identifying inrush current are mainly based on waveform features. Traditional harmonic braking and discontinuity angle detection belong to this type of method. However, due to the influence of factors such as the second harmonic content and CT saturation, there are still cases of false operation. Many scholars also use various signal processing techniques to extract time and frequency domain information from waveforms, such as harmonic analysis, wavelet transform, empirical Fourier transform, Kalman filter, etc. In recent years, artificial intelligence technology has flourished and is one of the most disruptive scientific and technological fields. It has strong processing capabilities in computational intelligence, perceptual intelligence, and cognitive intelligence. Artificial intelligence has also been studied in transformer inrush current identification. The main idea of ​​its application is to mine features in large amounts of data for classification, including support vector machines, artificial neural networks, probabilistic neural networks, etc.

[0004] Most traditional methods for identifying magnetizing inrush current based on waveform characteristics require setting threshold values, which is a difficulty in differential protection. Improper threshold setting will lead to malfunction of transformer protection. At present, artificial intelligence methods still have certain deficiencies in processing time series signals such as differential current, resulting in low generalization ability of the magnetizing inrush current identification model based on artificial intelligence. Summary of the invention

[0005] The purpose of the present invention is to provide a method and device for identifying transformer excitation inrush current based on Gram's summation angle field, which can realize the identification of excitation inrush current by artificial intelligence method and solve the problem of scarce transformer fault samples.

[0006] In order to achieve the above object, the solution of the present invention is:

[0007] A transformer excitation inrush current identification method based on Gram summation angle field, comprising:

[0008] Obtain the three-phase differential current of the power transformer and obtain a three-dimensional color map;

[0009] Taking the three-dimensional color map as input, using the SVDD-based excitation inrush current single classification model to identify whether the transformer is in an excitation inrush current condition;

[0010] Among them, the SVDD-based excitation inrush current single classification model is constructed based on the three-dimensional color map of the transformer under various working conditions and is trained using the LeNet-5 deep convolutional neural network.

[0011] In particular, the three-phase differential current of the power transformer is obtained to obtain a three-dimensional color map, including,

[0012] A segment is intercepted from the three-phase differential current time series of the power transformer to form a vector, which is then normalized and expressed in polar coordinates, and then the three-phase differential current transformation matrix is ​​obtained through Gram summation angular field transformation.

[0013] The three-phase differential current transformation matrix is ​​visualized to obtain the three-dimensional color map.

[0014] In particular, after normalization and polar coordinate expression, the three-phase differential current transformation matrix is ​​obtained through Gram summation angular field transformation, including,

[0015] Using the maximum and minimum normalization method, mapping the three-phase differential current vector to the [0,1] interval;

[0016] Encode the mapped vector value as a cosine angle, which is expressed as a polar angle and a polar diameter in polar coordinates;

[0017] The polar angles are transformed using the Gram summation angle field to obtain the three-phase differential current transformation matrix.

[0018] In particular, the three-phase differential current transformation matrix is ​​visualized to obtain the three-dimensional color map, including:

[0019] Map the values ​​of the three-phase differential current transformation matrix to a pixel value interval of [0,255];

[0020] Assign the matrix to the RGB three channels of the image respectively;

[0021] Generate a three-dimensional color image based on RGB three-channel data;

[0022] Data enhancement is performed on the three-dimensional color image to obtain the three-dimensional color map.

[0023] In particular, the SVDD-based single-classification model of excitation inrush current is constructed based on three-dimensional color images of transformers under various working conditions and trained using the LeNet-5 deep convolutional neural network, including:

[0024] The three-dimensional color image under the condition of excitation inrush current is taken as the positive sample, and the three-dimensional color image under the condition of non-excitation inrush current is taken as the negative sample. The LeNet-5 deep convolutional neural network is used to extract the feature vector of the positive sample.

[0025] All positive sample feature vectors constitute a sample set X, and the SVDD model is trained to obtain the sphere center a and radius R of the single classification model of excitation inrush current based on SVDD;

[0026] Calculate the feature vector of the three-dimensional image to be identified and calculate the generalized distance D from the sample z to the center of the sphere a 2 ;

[0027] If D 2 ≤R 2 , indicating that the transformer is in the excitation inrush current condition; otherwise, it means that no excitation inrush current occurs.

[0028] In particular, the three-phase differential current of the power transformer is obtained, including, in order to eliminate the phase difference between the current on the high-voltage side and the low-voltage side of the power transformer, using the change of the CT secondary connection to correct the phase, and adjusting the CT ratio on both sides to achieve amplitude adjustment; all CTs on both sides of the transformer are connected in star connection.

[0029] A transformer excitation inrush current identification device based on Gram summation angle field, comprising:

[0030] A three-dimensional color image acquisition module is configured to obtain a three-dimensional color image by acquiring the three-phase differential current of the power transformer; and

[0031] An excitation inrush current working condition identification module is configured to use the three-dimensional color map as input and use an excitation inrush current single classification model based on SVDD to identify whether the transformer is in an excitation inrush current working condition;

[0032] Among them, the SVDD-based excitation inrush current single classification model is constructed based on the three-dimensional color map of the transformer under various working conditions and is trained using the LeNet-5 deep convolutional neural network.

[0033] A computer device comprises a memory, a processor, and a computer program stored in the memory and executable on the processor; when the processor executes the computer program, the steps of the transformer excitation inrush current identification method based on the Gram summation angle field as described above are implemented.

[0034] A computer-readable storage medium stores a computer program; when the computer program is executed by a processor, the steps of the transformer excitation inrush current identification method based on the Gram summation angle field as described above are implemented.

[0035] After adopting the above scheme, the present invention has the following advantages:

[0036] (1) The present invention transforms the transformer three-phase differential current time series signal into a matrix through the Gram sum angular field, and then generates a three-dimensional color image, which enhances the characterization capability of the transformer working condition and is suitable for deep learning-based excitation inrush current identification;

[0037] (2) The present invention uses the LeNet-5 deep convolutional neural network to establish a discrimination model for the three-dimensional image of the excitation inrush current and extract sample feature vectors, making full use of the advantages of deep learning in image processing;

[0038] (3) The present invention adopts the idea of ​​semi-supervised learning and uses the extracted feature vector as input to establish a transformer excitation inrush current SVDD single classification recognition model. The inrush current recognition can be realized by learning only the excitation inrush current samples, thus solving the problem of scarce transformer fault samples. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is the overall workflow diagram of the present invention;

[0040] Figure 2 This is an example of the CT connection of a three-phase transformer targeted by the present invention;

[0041] Figure 3(a)-Figure 3(o) It is the characteristic of three-phase transformer under magnetizing inrush current, normal operating conditions and internal fault conditions;

[0042] Among them, Figure 3(a) is the three-phase differential current under the excitation inrush condition, Figure 3(b) is the polar coordinates of the differential current of phase A under the excitation inrush condition, Figure 3(c) is the polar coordinates of the differential current of phase B under the excitation inrush condition, Figure 3(d) is the polar coordinates of the differential current of phase C under the excitation inrush condition, Figure 3(e) is the three-dimensional diagram under the excitation inrush condition, Figure 3(f) is the three-phase differential current under normal conditions, Figure 3(g) is the polar coordinates of the differential current of phase A under normal conditions, and Figure 3(h) is the normal Figure 3(i) is the polar coordinates of the differential current of phase B under normal working conditions, Figure 3(j) is the three-dimensional diagram under normal working conditions, Figure 3(k) is the three-phase differential current under internal fault conditions, Figure 3(l) is the polar coordinates of the differential current of phase A under internal fault conditions, Figure 3(m) is the polar coordinates of the differential current of phase B under internal fault conditions, Figure 3(n) is the polar coordinates of the differential current of phase C under internal fault conditions, and Figure 3(o) is the three-dimensional diagram under internal fault conditions;

[0043] Figure 4This is a feature extraction network structure diagram based on LeNet-5. DETAILED DESCRIPTION

[0044] The technical solutions and beneficial effects of the present invention will be described in detail below with reference to the accompanying drawings.

[0045] The present invention provides a transformer excitation inrush current identification method and device using Gram summation angle field, the basic route of which is as follows: Figure 1 First, obtain the three-phase differential current from the power transformer The three-phase current time series are converted into GASF matrices using the Gram summation angle field, denoted as G ASF.A , G ASF.B and G ASF.C , assign the three matrices to the RGB channels of the three-dimensional image to generate a three-dimensional color image; then use a convolutional neural network to extract features from the image to obtain a multi-dimensional feature vector, and input the feature vector into the support vector data description (SVDD) algorithm to build an excitation inrush current recognition model. During model training, the transformer excitation inrush current, the three-phase differential current under normal working conditions and internal faults are obtained, and the three-dimensional color image is output to CNN after transformation and conversion, and the SVDD single-classification excitation inrush current recognition model is trained with the feature vector extracted by CNN. During online operation, the three-dimensional image and feature vector are obtained from the three-phase differential current in turn according to the same method, and the feature vector is input into the SVDD single-classification model to identify whether an excitation inrush current occurs.

[0046] The method for identifying transformer excitation inrush current based on Gram summation angle field provided by the present invention comprises the following steps:

[0047] Step 1, obtaining the three-phase differential current of the power transformer and obtaining a three-dimensional color map;

[0048] Step 2, using the three-dimensional color map as input, using the SVDD-based excitation inrush current single classification model to identify whether the transformer is in an excitation inrush current condition;

[0049] Among them, the SVDD-based excitation inrush current single classification model is constructed based on the three-dimensional color map of the transformer under various working conditions and is trained using the LeNet-5 deep convolutional neural network.

[0050] The specific contents of step one are:

[0051] Firstly, a segment is intercepted from the three-phase differential current time series of the power transformer to form a vector, which is then normalized and expressed in polar coordinates, and then the three-phase differential current transformation matrix is ​​obtained through Gram summation angular field transformation.

[0052] Then the three-phase differential current transformation matrix is ​​visualized to obtain the three-dimensional color map.

[0053] After the above normalization and polar coordinate expression, the specific process of obtaining the three-phase differential current transformation matrix through Gram summation angular field transformation is:

[0054] Firstly, the three-phase differential current vector is mapped to the interval [0,1] using the maximum and minimum normalization method;

[0055] Then, the mapped vector value is encoded as a cosine angle, which is represented as a polar angle and a polar diameter in polar coordinates;

[0056] Finally, the polar angles are transformed using the Gram summation angle field to obtain the three-phase differential current transformation matrix.

[0057] The specific process of visualizing the three-phase differential current transformation matrix to obtain the three-dimensional color map is as follows:

[0058] Firstly, the values ​​of the three-phase differential current transformation matrix are mapped to the pixel value interval of [0,255]; then the matrix is ​​assigned to the RGB three channels of the image respectively; a three-dimensional color image is generated according to the RGB three-channel data; finally, the three-dimensional color image is data enhanced to obtain the three-dimensional color image.

[0059] The construction process of the SVDD-based excitation inrush current single classification model includes the following steps:

[0060] The three-dimensional color image under the condition of excitation inrush current is taken as the positive sample, and the three-dimensional color image under the condition of non-excitation inrush current is taken as the negative sample. The LeNet-5 deep convolutional neural network is used to extract the feature vector of the positive sample.

[0061] All positive sample feature vectors constitute a sample set X, and the SVDD model is trained to obtain the sphere center a and radius R of the single classification model of excitation inrush current based on SVDD;

[0062] Calculate the feature vector of the three-dimensional image to be identified and calculate the generalized distance D from the sample z to the center of the sphere a 2 ;

[0063] According to D 2 Make a judgment, if D 2 ≤R 2 , indicating that the transformer is in the excitation inrush current condition; otherwise, it means that no excitation inrush current occurs.

[0064] Among them, before obtaining the three-phase differential current of the power transformer, in order to eliminate the phase difference between the current on the high-voltage side and the low-voltage side of the power transformer, the phase is corrected by using the change of the CT secondary connection, and the CT ratio on both sides is adjusted to achieve amplitude adjustment; all CTs on both sides of the transformer are connected in star connection.

[0065] The present invention also provides a transformer excitation inrush current identification device based on Gram summation angle field, comprising:

[0066] A three-dimensional color image acquisition module is configured to obtain a three-dimensional color image by acquiring the three-phase differential current of the power transformer; and

[0067] An excitation inrush current working condition identification module is configured to use the three-dimensional color map as input and use an excitation inrush current single classification model based on SVDD to identify whether the transformer is in an excitation inrush current working condition;

[0068] Among them, the SVDD-based excitation inrush current single classification model is constructed based on the three-dimensional color map of the transformer under various working conditions and is trained using the LeNet-5 deep convolutional neural network.

[0069] A specific embodiment is provided below to explain the technical solution and technical effects of the present invention in detail, including the following steps:

[0070] Step 1, three-phase differential current acquisition: The present invention samples the three-phase differential current of the power transformer, and then performs subsequent processing to establish an excitation inrush current identification model. In order to suppress the third harmonic and improve the waveform, the power transformer generally adopts the Yd-11 wiring method, which leads to a phase difference between the high-voltage side and the low-voltage side current. Therefore, it is necessary to use phase shifting technology and balancing technology to adjust the phase and amplitude of the current on both sides of the transformer differential protection.

[0071] In the above-mentioned transformer excitation inrush current identification method using Gram summation angular field transformation, in step 1, the specific operation method of differential current adjustment is as follows:

[0072] Step 1.1, use analog method to adjust phase and amplitude. For the transformer with Yd-11 connection, use the change of CT secondary connection to correct the phase. The specific method is to connect the Y side CT into triangle and the triangle side into Y shape, such as Figure 2 As shown. Then the three-phase differential current is expressed as:

[0073]

[0074] And the CT ratio on both sides must meet:

[0075]

[0076] Step 1.2, use digital method to adjust phase and amplitude, and connect all CTs on both sides of the transformer into star connection. The phase shift and amplitude compensation are realized by software calculation internally, and the shift formula is as follows:

[0077]

[0078] The differential current is calculated as follows

[0079]

[0080] Where K is the balance coefficient, which is derived from formula (2).

[0081] Step 2: Preprocessing of three-phase differential current. In step 1, the three-phase differential current is obtained by analog or digital means. A segment is cut from the A phase current time series and represented as x A =(x1,x2,…,x N ), phase B and phase C are also represented by x B =(x1,x2,…,x N ) and x C =(x1,x2,…,x N ), N is the vector dimension, which depends on the sampling period Δt, and the sampling period is preferably set to 0.002s. A 、x B and x C Perform preprocessing.

[0082] In the above-mentioned transformer excitation inrush current identification method using Gram summation angular field transformation, in step 2, the specific operation method of data preprocessing is as follows:

[0083] Step 2.1, normalization. Use the maximum and minimum normalization method to normalize the vector x A 、x B and x C Mapped to the [0,1] interval, we get the normalized one-dimensional structured data. Taking phase A as an example, the maximum and minimum normalized expressions are as follows:

[0084]

[0085] x B and x C The same method is used for processing, and the normalized vectors are expressed as and

[0086] Step 2.2, polar coordinate expression. and The value of is encoded as the cosine angle, which is expressed in polar coordinates as the polar angle Φ i and the polar diameter r i , taking phase A as an example, it is processed according to the following formula:

[0087]

[0088] In the formula, Φ i is the polar angle of point i, ri is the polar diameter of point i. and The polar coordinate expression of is also in the same way.

[0089] Step 2.3, Gram sum angular field transformation. and After conversion to polar coordinates, the Gram summation angle field is used for transformation. Taking phase A as an example, the GASF transformation is expressed as:

[0090] G ASF,A (i,j)=[cos(Φ i +Φ j )] (7)

[0091] In the formula, Φ j is the polar angle of point j (j=1,…,N), and j can be equal to i. Using the relationship between the sum angles, equation (7) can be further expressed as:

[0092]

[0093] Formula (8) is more suitable for calculation. The matrix G can be obtained by formula (8): ASF,A Phase B and phase C are also processed in the same way, and the obtained matrices are represented as G ASF,B and G ASF,C .

[0094] Step 3: Generate a three-dimensional color map. Step 2: Obtain the GASF transformation matrix G of the three-phase differential current of A, B, and C ASF,A , G ASF,B and G ASF,C ,The following three matrices are visualized to generate a three-dimensional color image to enhance the data features for subsequent CNN processing.

[0095] In the above-mentioned transformer excitation inrush current identification method using Gram summation angular field transformation, in step 3, the specific operation method of generating the three-dimensional color map is as follows:

[0096] Step 3.1, GASF transformation matrix data mapping. Matrix G ASF,A , G ASF,B and G ASF,C The value range is between [-1, 1], and it needs to be mapped to the pixel value interval of [0, 255] first:

[0097]

[0098]

[0099]

[0100] Step 3.2, RGB channel assignment. In order to draw a three-dimensional image, the matrix G R , G G and G B As the RGB channels of the image:

[0101] R(i,j)=G R (i,j) (12)

[0102] G(i,j)=G G (i,j) (13)

[0103] B(i,j)=G B (i,j) (14)

[0104] Where R(i,j), G(i,j), and B(i,j) represent the values ​​of the three primary color channels of the image at (i,j).

[0105] Step 3.3, three-dimensional image generation. After obtaining the RGB three-channel data, generate a standard three-dimensional color image (width × height × number of three primary color channels). Figure 3 shows the results of the transformer excitation inrush current, normal operating conditions and internal faults. Figure 3(a) is the three-phase differential current under the transformer excitation inrush current condition. The sampling period is set to 0.002s, and N=200 is set. The differential currents of phases A, B, and C are processed according to formula (6) to obtain polar coordinate expressions, and they are plotted in polar coordinates, as shown in Figures 3(b), 3(c), and 3(d). Then, the Gram sum angular field is transformed according to formula (8) to obtain the GASF transformation matrix G ASF,A , G ASF,B and G ASF,C And according to equations (12) to (14), the matrix G ASF,A , G ASF,B and G ASF,C Assign values ​​to the RGB channels of the three-dimensional image respectively. After obtaining the matrix of the three-dimensional image, draw it out, and the three-dimensional image under the excitation inrush current condition is shown in Figure 3(e). Then, obtain the waveform of the transformer when it starts normally, as shown in Figure 3(f), and perform the same processing as above to obtain the polar coordinate expression of the three-phase differential current, as shown in Figure 3(g), Figure 3(h) and Figure 3(i), respectively, and finally obtain the three-dimensional image under normal working conditions, as shown in Figure 3(j). The three-phase differential current under the internal fault condition of the power transformer is shown in Figure 3(k), and the corresponding polar coordinates are shown in Figure 3(l), Figure 3(m) and Figure 3(n), respectively. Finally, the three-dimensional color image under the internal fault condition is obtained, as shown in Figure 3(o).

[0106] FIG3 shows three typical 3D images under three working conditions. It can be seen that the colors and textures of the 3D color images under different working conditions are quite different. In other words, the 3D color images enhance the working characteristics of the transformer, which is the basic basis for identifying the excitation inrush current through the 3D color images.

[0107] Step 3.4, image sample data enhancement. Since the probability of excitation inrush current and internal fault in the transformer is low and samples are scarce, this embodiment proposes a data augmentation (DA) method to reasonably expand the samples to improve the performance of the deep learning model. The following methods are mainly used for processing: (1) Exposure adjustment. Adjust the image brightness to make it dim or bright; (2) Image noise. Add salt and pepper noise and Gaussian noise to the image, where salt and pepper noise is the change of the disturbed image pixel value to the maximum value 255 or the minimum value 0, and Gaussian noise refers to the noise distributed on each pixel with a Gaussian distribution. (3) Image blur. Gaussian blur differentiates the image according to the center, so that the weight of each pixel changes with its distance from the center. By enhancing the image sample data, richer learning samples can be obtained.

[0108] Step 4: Image feature extraction. In step 3, the three-dimensional color image of the transformer under various working conditions was obtained through the Gram sum angular field transform, and the samples were expanded through data enhancement to form a three-dimensional color image sample set. In this step, CNN will be used to extract the features of the three-dimensional image to facilitate the subsequent SVDD-based inrush current recognition modeling.

[0109] In the above-mentioned transformer excitation inrush current identification method using the Gram summation angle field, in step 4, the specific operation method of extracting based on CNN features is as follows:

[0110] Step 4.1, image adjustment. Step 3 shows that the three-dimensional images obtained under normal operating conditions, magnetizing inrush current conditions and internal faults of the transformer have large differences in color and texture. These features can be used to distinguish different operating conditions and identify magnetizing inrush currents. The present invention uses the LeNet-5 deep convolutional neural network to extract three-dimensional image features. The LeNet-5 input is a 32×32 image, so the image obtained in step 3 needs to be resized to meet the requirements of LeNet-5 for the input image. Preferably, the graphics transformation function in the PyTorch graphics library torchvision is used to complete it. Its core function transforms.Resize() is an image processing function provided by the transforms module, which can complete the scaling operation of the image. Specifically, this function can adjust the size of the input image to a given target size.

[0111] Step 4.2, feature extraction based on LeNet-5. The LeNet-5 network consists of 7 layers, the input image size is 32×32, and each layer includes trainable parameters. This embodiment designs the LeNet-5 network to adapt to the use scenario of the excitation inrush current identification. The network structure is as follows Figure 4 This embodiment has only two outputs when identifying transformer inrush current, so the number of output layer neurons is 2; the number of feature maps has a certain impact on the overall recognition performance of the network. This embodiment compares and verifies the number of convolution kernels of multiple F5 layers, and finally adopts an output layer network with 20 feature maps to output and save the features of the F5 layer as the input of the next SVDD classification model. The smaller number of feature maps also reduces the amount of computation for the image features of subsequent training.

[0112] Step 4.3, LeNet-5 network training. Step 3 obtains three-dimensional images of the transformer under normal working conditions, magnetizing inrush current working conditions, and internal faults. The three-dimensional images under the magnetizing inrush current working conditions are used as positive samples, and the three-dimensional images under normal working conditions and internal faults are synthesized as negative samples. The LeNet-5 network is trained with the positive and negative sample sets to generate model Model1, and the training features of the positive sample data can be extracted based on Model1.

[0113] Step 5, excitation inrush current identification. In step 4, the model Model1 is generated by learning the positive samples and negative samples through the LeNet-5 network. The positive sample data training features can be extracted from Model1. This step will use the extracted positive sample features as input to establish an excitation inrush current single classification model based on SVDD.

[0114] In step 5, the specific operation method based on SVDD surge current identification is as follows:

[0115] Step 5.1, establish the inrush current SVDD model. SVDD is a single classification algorithm. Its main idea is to construct a hypersphere of the normal domain by training positive samples, so as to determine whether the new sample belongs to the normal domain of normal data, and then determine whether the new sample is abnormal. In this embodiment, the three-dimensional image features under the excitation inrush current condition are used as positive samples, so the obtained SVDD model can effectively determine whether the transformer has an inrush current condition. Assume x = (x1, x2, ..., x 20 ) is the feature vector extracted from a positive sample, and all positive sample feature vectors constitute the sample set X. When constructing the hypersphere, in order to reduce the impact of abnormal data points being included in the normal domain, the penalty factor C and the slack variable ξ are introduced i Assuming that the center of the hypersphere when the positive sample is completely surrounded is a and the radius is R, the corresponding optimization equation is:

[0116]

[0117] By introducing the Lagrange multiplier and solving the corresponding function, equation (15) can be written as

[0118]

[0119] In formula (16), non-zero α i The corresponding samples are the support vectors. Solving the radius and center of the hypersphere shows that the radius and center are only determined by the support vectors, expressed as x sv , which is independent of other samples. Using the feature vector sample set X for training, the magnetizing inrush current SVDD discrimination model can be obtained, and its sphere center a and radius R are expressed as follows:

[0120]

[0121] Step 5.2, magnetizing inrush current identification. The transformer magnetizing inrush current identification model is established above. When the feature vector z=(z1,z2,…,z 20 ), find the generalized distance D from sample z to the center of the sphere a 2 , the calculation formula is as follows:

[0122]

[0123] According to the generalized distance D 2 To identify the excitation inrush current. The linear separable model is used above to describe the excitation inrush current discrimination model. In fact, there is a large nonlinearity between the 20-dimensional feature vector of the three-dimensional image and the transformer excitation inrush current state. When establishing the SVDD inrush current discrimination model, this embodiment performs a nonlinear mapping on the data, that is, the original sample space is mapped to a higher-dimensional feature space through a nonlinear function φ(x) to perform linear separability in the high-dimensional feature space. The inner product operation in the high-dimensional feature space is often very complicated, so the kernel function is introduced to simplify the inner product operation. The kernel function is defined as K(x i ,x j )=φ(x i ).φ(x j ). Accordingly, R and D in the magnetizing inrush current identification model will be rewritten as:

[0124]

[0125] and

[0126]

[0127] This embodiment uses a Gaussian kernel function, which has excellent performance in most cases and is suitable for samples presenting multiple feature dimensions. The Gaussian kernel function has two hyperparameters that need to be tuned, the penalty coefficient C and the bandwidth σ. This embodiment uses a grid method for parameter optimization.

[0128] The above established a vertical SVDD inrush current discrimination model based on kernel function. When running online, D is obtained according to formula (20): 2 , if D 2 ≤R 2 , it indicates that the transformer is in the excitation inrush current condition; otherwise, it means that there is no excitation inrush current. Thus, the transformer excitation inrush current identification is completed.

[0129] In this embodiment, a device applying the above transformer excitation inrush current identification method using the Gram summation angle field includes: a transformer three-phase differential current sensor, a signal acquisition unit and a calculation unit.

[0130] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

[0131] Although the present embodiment uses more terms such as three-phase transformer, three-phase differential current, Gram sum angular field transform, three-dimensional image, LeNet-5, etc., it does not exclude the possibility of using other terms. These terms are used only to more conveniently describe and explain the essence of the present invention; interpreting them as any additional limitation is contrary to the spirit of the present invention.

[0132] An embodiment of the present invention further provides another computer device, including a processor and a memory configured to store a computer program that can be run on the processor; wherein, when the processor is configured to run the computer program, the method steps in the aforementioned embodiment are executed.

[0133] In practical applications, the processor includes a field programmable gate array (FPGA), and the processor may be a central processing unit (CPU) or a digital signal processor (DSP). It is understandable that for different devices, the electronic device used to implement the function of the processor may be other, and the embodiment of the present invention does not specifically limit it.

[0134] The above-mentioned memory can be a volatile memory (volatile memory), such as a random access memory (RAM); or a non-volatile memory (non-volatile memory), such as a read-only memory (ROM), a flash memory, a hard disk (HDD) or a solid-state drive (SSD); or a combination of the above-mentioned types of memory, and provide instructions and data to the processor.

[0135] In an exemplary embodiment, an embodiment of the present invention further provides a computer-readable storage medium for storing a computer program.

[0136] Optionally, the computer-readable storage medium can be applied to any one of the methods in the embodiments of the present invention, and the computer program enables the computer to execute the corresponding processes implemented by the processor in each method in the embodiments of the present invention. For the sake of brevity, they are not described here.

[0137] In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the components shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0138] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes. The schemes in the embodiments of the present invention may be implemented in various computer languages, for example, object-oriented programming language Java and literal scripting language JavaScript, etc.

[0139] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0140] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0141] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0142] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0143] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for identifying transformer excitation inrush current based on Gram summation angle field, characterized in that: Obtain the three-phase differential current of the power transformer and obtain a three-dimensional color map; Taking the three-dimensional color map as input, using the SVDD-based excitation inrush current single classification model to identify whether the transformer is in an excitation inrush current condition; The SVDD-based single-classification model of magnetizing inrush current is constructed based on three-dimensional color images of transformers under various working conditions and is trained using a LeNet-5 deep convolutional neural network. Among them, the three-phase differential current of the power transformer is obtained to obtain a three-dimensional color map, including, A segment is intercepted from the three-phase differential current time series of the power transformer to form a vector, which is then normalized and expressed in polar coordinates, and then the three-phase differential current transformation matrix is ​​obtained through Gram summation angular field transformation. Visualizing the three-phase differential current transformation matrix to obtain the three-dimensional color map; The three-phase differential current transformation matrix is ​​visualized to obtain the three-dimensional color map, including: Map the values ​​of the three-phase differential current transformation matrix to a pixel value interval of [0,255]; Assign the matrix to the RGB three channels of the image respectively; Generate a three-dimensional color image based on RGB three-channel data; Data enhancement is performed on the three-dimensional color image to obtain the three-dimensional color map.

2. The method according to claim 1, characterized in that: After normalization and polar coordinate expression, the three-phase differential current transformation matrix is ​​obtained through Gram summation angular field transformation, including: Using the maximum and minimum normalization method, mapping the three-phase differential current vector to the [0,1] interval; Encode the mapped vector value as a cosine angle, which is expressed as a polar angle and a polar diameter in polar coordinates; The polar angles are transformed using the Gram summation angle field to obtain the three-phase differential current transformation matrix.

3. The method according to claim 1, characterized in that: The SVDD-based excitation inrush current single classification model is constructed based on three-dimensional color images of transformers under various working conditions and trained using the LeNet-5 deep convolutional neural network, including: The three-dimensional color image under the condition of excitation inrush current is taken as the positive sample, and the three-dimensional color image under the condition of non-excitation inrush current is taken as the negative sample. The LeNet-5 deep convolutional neural network is used to extract the feature vector of the positive sample. All positive sample feature vectors constitute a sample set X, and the SVDD model is trained to obtain the sphere center a and radius R of the single classification model of excitation inrush current based on SVDD; Calculate the feature vector of the three-dimensional image to be identified and calculate the generalized distance D from the sample z to the center of the sphere a 2 ; If D 2 ≤R 2 , indicating that the transformer is in the excitation inrush current condition; otherwise, it means that no excitation inrush current occurs.

4. The method according to claim 1, characterized in that: The three-phase differential current of the power transformer is obtained, including, in order to eliminate the phase difference between the current on the high-voltage side and the low-voltage side of the power transformer, using the change of the CT secondary connection to correct the phase, and adjusting the CT ratio on both sides to achieve amplitude adjustment; all CTs on both sides of the transformer are connected in star connection.

5. A transformer excitation inrush current identification device based on Gram summation angle field, characterized in that: include, A three-dimensional color image acquisition module is configured to obtain a three-dimensional color image by acquiring the three-phase differential current of the power transformer; and An excitation inrush current working condition identification module is configured to use the three-dimensional color map as input and use an excitation inrush current single classification model based on SVDD to identify whether the transformer is in an excitation inrush current working condition; The SVDD-based single-classification model of magnetizing inrush current is constructed based on three-dimensional color images of transformers under various working conditions and is trained using a LeNet-5 deep convolutional neural network. The three-dimensional color map acquisition module acquires the three-phase differential current of the power transformer to obtain a three-dimensional color map, including: A segment is intercepted from the three-phase differential current time series of the power transformer to form a vector, which is then normalized and expressed in polar coordinates, and then the three-phase differential current transformation matrix is ​​obtained through Gram summation angular field transformation. Visualizing the three-phase differential current transformation matrix to obtain the three-dimensional color map; The three-phase differential current transformation matrix is ​​visualized to obtain the three-dimensional color map, including: Map the values ​​of the three-phase differential current transformation matrix to a pixel value interval of [0,255]; Assign the matrix to the RGB three channels of the image respectively; Generate a three-dimensional color image based on RGB three-channel data; Data enhancement is performed on the three-dimensional color image to obtain the three-dimensional color map.

6. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor; characterized in that: When the processor executes the computer program, the steps of the transformer excitation inrush current identification method based on the Gram summation angle field as described in any one of claims 1 to 4 are implemented.

7. A computer-readable storage medium storing a computer program; characterized in that: When the computer program is executed by a processor, the steps of the transformer excitation inrush current identification method based on the Gram summation angle field as claimed in any one of claims 1 to 4 are implemented.

Images