Power transmission line fault identification method and device based on gram angle field and residual network

Abstract

The present application relates to a kind of power transmission line fault identification method based on gram angle field and residual network, including the steps of constructing fault type and fault cause identification model, it includes: the original identification model based on multi-objective residual network is constructed;Three-phase fault recording data of power transmission line fault sample is obtained, the three-phase fault recording data obtained is converted into two-dimensional color graphics based on gram angle field, as fault sample image;Fault sample graphics is classified according to fault type and fault cause, meanwhile, to the small sample type Smote oversampling processing is carried out, to obtain fault sample set;Original identification model is trained and tested based on fault sample set, and final identification model is obtained.The present application converts one-dimensional fault recording data into two-dimensional color image by GAF method, reduces the influence of voltage and current waveform, fault distance and transition resistance and other factors on fault feature extraction, and also can well deal with the class imbalance problem between data.

Description

Technical Field

[0001] This invention relates to the field of power transmission line fault identification technology, and in particular to a method and apparatus for power transmission line fault identification based on Gram angle field and residual network. Background Technology

[0002] The safe and reliable operation of transmission lines is crucial for ensuring the safe and stable operation of the power system. However, due to the wide geographical distribution and large area spanned by overhead transmission lines, their complex and variable operating environment, and susceptibility to harsh natural conditions, they are prone to various types of faults during operation. Timely and accurate identification of the cause and type of fault is of great significance for guiding protection and automatic safety devices to quickly isolate faults, restore subsequent power supply, reduce line outage time, and ensure the stable operation of the power system.

[0003] Existing research on transmission line fault identification typically involves signal decomposition of collected fault voltage and current data to extract fault waveform features for fault identification. Some literature uses wavelet transform to directly analyze signals of different properties for fault detection and classification. Other literature utilizes Discrete Fourier Transform (DFT) to evaluate the phase of voltage and current signals with rapidly changing frequency characteristics, ultimately distinguishing faults based on phase differences in the fault signals. Furthermore, some literature employs Fourier transform to extract fault signal features and combines it with frequency response methods to propose a fault identification model that integrates Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM). However, the method of decomposing one-dimensional signals is easily affected by factors such as voltage and current waveforms, fault distance and transition resistance when selecting fault features. Moreover, the feature selection process is complicated and the accuracy of classification and identification is not ideal. Furthermore, the fault trip records obtained from the power grid company show that the fault samples of transmission lines often have serious class imbalances, which also leads to poor actual identification results of this method.

[0004] Therefore, it is necessary to design a new method for identifying transmission line faults to solve the above problems. Summary of the Invention

[0005] In view of this, the present invention provides a transmission line fault identification method based on Gram angle field (GAF) and multi-objective residual network. The method uses GAF to convert the collected one-dimensional three-phase fault voltage and current waveform data into two-dimensional color images, which serve as input to the subsequent multi-objective residual network. Furthermore, leveraging the strong feature extraction capability of multi-objective residual networks in image recognition, the method fully mines the fault features in the fault waveform data to train the network model. In addition, addressing the class imbalance problem in transmission line fault samples, the present invention employs Somte oversampling to eliminate class imbalance between data, ultimately achieving effective identification of transmission line fault types and causes.

[0006] The first aspect of this invention discloses a method for identifying transmission line faults based on Gram angle field and residual network, comprising the following steps:

[0007] Acquire three-phase waveform data from the secondary side of the transmission line;

[0008] Based on the Gram angle field, the extracted secondary side three-phase waveform data of the transmission line is converted into a two-dimensional color image, which is used as a fault identification image.

[0009] The fault identification images obtained are identified according to the pre-constructed transmission line fault identification model. The transmission line fault identification model is a multi-output convolutional neural network architecture based on a multi-objective residual network, which can output the fault cause and fault type as prediction results simultaneously.

[0010] In the above implementation process, the three-phase waveform data of the secondary side of the transmission line is converted into a two-dimensional color image using the Gram angle field method. This two-dimensional color image is then input into a pre-constructed transmission line fault identification model for fault identification. Because the transmission line fault identification model of this invention can simultaneously output the fault cause and fault type as prediction results, it enables simultaneous prediction of the fault cause and fault type of the tested transmission line.

[0011] According to the transmission line fault identification method based on Gram angle field and residual network disclosed in the first aspect of the present invention, before acquiring the three-phase waveform data of the transmission line, the method further includes a step of constructing a transmission line fault identification model, comprising:

[0012] An original identification model is constructed, which is a multi-output convolutional neural network architecture based on a multi-objective residual network. With the help of the multi-output convolutional neural network architecture of the multi-objective residual network, it is possible to construct an original identification sub-model for fault cause and an original identification sub-model for fault type in the original identification model.

[0013] Acquire three-phase fault waveform data on the secondary side of the faulty transmission line;

[0014] Based on the Gram angle field, the secondary side three-phase fault recording data of each faulty transmission line is converted into a two-dimensional color image as a fault sample image;

[0015] The converted fault sample images are classified according to fault type and fault cause, and Smote oversampling is performed on the minority fault samples to obtain the fault sample set.

[0016] Using a fault sample set based on two-dimensional color images as input, the original identification model is trained and tested to obtain the trained transmission line fault identification model. In this step, the process of training and testing the original identification model is also the process of training and testing the fault cause identification sub-model and the fault type original identification sub-model.

[0017] According to the transmission line fault identification method based on Gram angle field and residual network disclosed in the first aspect of the present invention, preferably, the three phases in the three-phase fault recording data of the secondary side of the transmission line refer to phase A, phase B and phase C, respectively. The three-phase fault recording data of the secondary side includes transient recording data of fault voltage and transient recording data of fault current of the three phases of the secondary side of the transmission line. Specifically, it includes: transient recording data of phase A voltage and current of the secondary side of the transmission line; transient recording data of phase B voltage and current of the secondary side of the transmission line; and transient recording data of phase C voltage and current of the secondary side of the transmission line.

[0018] In a preferred embodiment, the sampling interval of the fault recording data is from one week before the fault occurred to two weeks after the fault occurred.

[0019] According to the transmission line fault identification method based on Gram angle field and residual network disclosed in the first aspect of the present invention, the step of converting the acquired secondary side three-phase fault recording data of each faulty transmission line into a two-dimensional color image based on the Gram angle field includes:

[0020] The one-dimensional time series arrays of the three-phase voltage and current on the secondary side of the faulty transmission line are concatenated into a two-dimensional tensor matrix.

[0021] The transient recording data of fault voltage and transient recording data of fault current of the three phases on the secondary side of the faulty transmission line are normalized.

[0022] The normalized data is then transformed into polar coordinates to obtain the radius and angle corresponding to each data point.

[0023] By utilizing sum-angle and difference-angle relationships, the two-dimensional tensor matrix is ​​calculated to obtain its corresponding Gram angle field transformation.

[0024] In the above process, because the one-dimensional time series arrays of voltage and current are concatenated into a two-dimensional tensor matrix, and normalization and polar coordinate transformation are performed based on the tensor matrix, the graphical features derived from the Gram angle field transformation are amplified, making the characteristics of the fault waveform more obvious. This is highly beneficial for feature extraction by convolutional networks. Moreover, in the above implementation, performing the Gram angle field transformation in tensor form is simpler and allows for much faster calculation of these parameters compared to calculating only one-dimensional data for Gram angle field changes. This method is faster and involves fewer computational steps.

[0025] According to the transmission line fault identification method based on Gram angle field and residual network disclosed in the first aspect of the present invention, in the step of concatenating a one-dimensional time series array of the secondary three-phase voltage and current of the faulty transmission line into a two-dimensional tensor matrix, the expression of the two-dimensional tensor matrix is:

[0026]

[0027] In the step of normalizing the transient fault voltage and transient fault current data of the three phases on the secondary side of the faulty transmission line, the normalization formula is as follows:

[0028]

[0029] In the formula: The fault data points are the fault voltages and fault currents collected for phases A, B, and C. These are the electrical quantity vectors corresponding to the fault voltages and fault currents of phases A, B, and C.

[0030] In the step of transforming the normalized data into polar coordinates to obtain the radius and angle corresponding to each data point, the formula for polar coordinate transformation is:

[0031]

[0032] In the formula: This represents the phase angle corresponding to each data point in the two-dimensional tensor matrix; The radius corresponding to each data point; The time number corresponding to each data point in the fault waveform data; N is the total number of sampling points in the acquired fault waveform data;

[0033] In the step of calculating the corresponding Gram angle field transformation of a two-dimensional tensor matrix using the sum and difference angle relationships, the calculation formula used is as follows:

[0034] ;

[0035] ;

[0036] In the formula: Let be the phase angle value corresponding to each data point in each electrical quantity, where This refers to one phase of the three-phase voltage and current (A, B, C). n and m refer to the phase angles corresponding to these data points. I is the unit vector. GASF is the Gram angle and field transformation formula. GADF is the Gram angle difference field transformation formula.

[0037] In this invention, after the transmission line fault identification model is constructed, the process of predicting the causes and types of transmission line faults includes the following steps: converting the acquired three-phase waveform data of the secondary side of the transmission line into a two-dimensional color image based on the Gram angle field. This involves: concatenating a one-dimensional time series array of the three-phase voltage and current of the secondary side of the transmission line to be predicted into a two-dimensional tensor matrix; normalizing the transient voltage and current waveform data of the three phases of the secondary side of the transmission line to be predicted based on the two-dimensional tensor matrix; transforming the normalized data into polar coordinates to obtain the radius and angle corresponding to each data point in the two-dimensional tensor matrix; and calculating the corresponding Gram angle field transformation of the two-dimensional tensor matrix using sum and difference angle relationships. The formulas used in these steps are consistent with the corresponding formulas mentioned above.

[0038] According to the transmission line fault identification method based on Gram angle field and residual network disclosed in the first aspect of the present invention, in the step of classifying the converted fault sample images according to fault type and fault cause, the converted two-dimensional images are first classified according to fault type. Preferably, the classification can yield 10 fault types: single-phase ground fault of phase A, single-phase ground fault of phase B, single-phase ground fault of phase C, two-phase faults of phases A and B, two-phase faults of phases A and C, two-phase faults of phases B and C, two-phase ground faults of phases A and B, two-phase ground faults of phases A and C, two-phase ground faults of phases B and C, and three-phase faults of phases A, B, and C. Further, the method also includes classifying the converted two-dimensional color images according to fault cause, wherein preferably, the fault cause includes i types such as lightning strike fault, foreign object fault, wildfire fault, wind deflection fault, icing fault, and tree flashover fault.

[0039] In a preferred embodiment, the step of classifying the converted fault sample images according to fault type and fault cause further includes a step of performing Shote oversampling on minority class fault samples. This step includes: converting the classified fault transient waveform image into a pixel matrix; using the Shote oversampling algorithm to calculate the Euclidean distance from each minority class fault sample in the minority class fault sample set to all samples in the minority class fault sample set; then calculating the k nearest neighbors of the minority class fault sample based on the Euclidean distance; randomly selecting a neighbor from the k nearest neighbors; selecting any position on the straight line between the neighbor and the minority class fault sample point; and generating a new fault sample using linear interpolation.

[0040] After the above classification is completed, the generated samples are divided into two main categories: training set and test set, in order to train and test the original recognition model.

[0041] According to the first aspect of the present invention, the transmission line fault identification method based on Gram angle field and residual network is disclosed, wherein the convolutional neural network of the original identification model adopts a multi-output convolutional neural network architecture based on multiple residual network modules, so that the original identification model can have a fault type original identification sub-model and a fault cause original identification sub-model.

[0042] A feature fusion layer is built between the original fault type identification sub-model and the original fault cause identification sub-model to fuse the feature parameters obtained from the training of the two sub-models; with the help of the fusion parameters implemented based on the feature fusion layer, the original fault type identification sub-model and the original fault cause identification sub-model achieve their respective outputs.

[0043] In the above implementation process, since the convolutional neural network of the original identification model adopts a multi-output convolutional neural network architecture based on multiple residual network modules, the original identification model has a fault type original identification sub-model and a fault cause original identification sub-model. As a result, the transmission line fault identification model after final training has a transmission line fault type identification sub-model and a transmission line fault cause identification sub-model, so as to realize the synchronous prediction and output of fault cause and fault type.

[0044] Furthermore, in the above implementation process, since a feature fusion layer is built between the two sub-models, the feature parameters obtained from the training of the two sub-models can be fused together using the feature fusion layer, thereby eliminating the impact of fault feature differences on the final model performance.

[0045] According to the first aspect of the present invention, a method for identifying transmission line faults based on Gram angle field and residual network is disclosed, wherein in the multi-output convolutional neural network architecture of the multi-residual network module, each residual neural network module contains two identical basic computational units, and a jump connection layer that can prevent model overfitting is connected between the two basic computational units.

[0046] In a preferred embodiment, the optimizer of the original identification model is optimized using the SGD algorithm, and the loss function is the cross-entropy loss function combined with the Softmax classifier. The training stride of the model is set to 64, and the initial learning rate of the model is set to 0.001. Furthermore, during the training process of the convolutional neural network, 70% of the fault type and fault cause identification sample set is randomly selected as the training sample set, and the remainder is used as the test sample set. This is used as the target domain data input to the multi-output convolutional neural network model, and training is performed with fault type and fault cause as the output.

[0047] Furthermore, according to the transmission line fault identification method based on Gram angle field and residual network disclosed in the first aspect of the present invention, in the step of training and testing the original identification model, a maximum number of training iterations is set and it is determined whether the model's loss function converges. If the loss function converges, the training of the model can be completed after reaching the maximum number of training iterations. If the loss function does not converge, the maximum number of training iterations is increased until convergence is achieved, and finally the transmission line fault identification model is obtained.

[0048] A second aspect of the present invention also discloses a power transmission line fault identification device based on Gram angle field and residual network, comprising:

[0049] The acquisition unit is used to acquire the three-phase waveform recording data of the secondary side of the transmission line;

[0050] The Gram angle field conversion unit is used to convert the extracted secondary three-phase waveform data of the transmission line into a two-dimensional color image based on the Gram angle field, which can be used as a fault identification image.

[0051] The transmission line fault identification unit is used to identify the obtained fault identification graphics according to the pre-built transmission line fault identification model. The transmission line fault identification model is a multi-output convolutional neural network architecture based on a multi-objective residual network, which can output the fault cause and fault type as prediction results simultaneously.

[0052] This also includes:

[0053] The model building unit is used to build the original identification model, which is a multi-output convolutional neural network architecture based on a multi-objective residual network.

[0054] The second acquisition unit is used to acquire the secondary side three-phase fault recording data of the faulty transmission line;

[0055] The Gram angle field conversion unit is also used to convert the acquired secondary side three-phase fault recording data of each faulty transmission line into a two-dimensional color image based on the Gram angle field, as a fault sample image.

[0056] The fault sample classification unit is used to classify the converted fault sample images according to fault type and fault cause, and to perform Smote oversampling processing on the minority class fault samples to obtain the fault sample set.

[0057] The training unit is used to train and test the original identification model by taking a fault sample set based on two-dimensional color images as input, so as to obtain the trained transmission line fault identification model.

[0058] A third aspect of the present invention discloses an electronic device, the electronic device including a memory and a processor, the memory being used to store a computer program, the processor running the computer program to cause the electronic device to perform the transmission line fault identification method based on Gram angle field and residual network as described above.

[0059] Compared with the prior art, the present invention has the following beneficial effects:

[0060] (1) The transmission line fault identification method based on Gram angle field and residual network of the present invention can realize the self-extraction of feature quantities, reduce the influence of factors such as voltage and current waveforms, fault distance and transition resistance on fault feature extraction, effectively deal with the problem of white noise in waveform data, and has low requirements for voltage and current sampling frequency.

[0061] (2) The present invention converts the fault three-phase voltage and current waveforms into Gram angle field diagrams, which amplifies the fault characteristic quantities, which is beneficial for the network to extract the fault waveform characteristic quantities, thereby improving the identification effect of the network.

[0062] (3) The present invention integrates the one-dimensional data of the three-phase voltage and current of A, B and C into a two-dimensional tensor matrix. In the subsequent computer processing, it avoids the need to perform repeated and tedious calculation steps for each electrical quantity. Therefore, the calculation process is simpler and the operation speed is faster.

[0063] (4) For the class imbalance problem, this method uses Smote oversampling technology to reduce the impact of class imbalance between data on network training performance and improve the network's recognition accuracy.

[0064] The following describes in detail the transmission line fault identification method and apparatus based on Gram angle field and residual network of the present invention with reference to the embodiments shown in the accompanying drawings and the reference numerals. Attached Figure Description

[0065] Figure 1 The flowchart of the transmission line fault identification method based on Gram angle field and residual network of the present invention is shown.

[0066] Figure 2 A flowchart illustrating the steps involved in constructing a transmission line fault identification model in this invention is shown.

[0067] Figure 3 The diagram shows the waveform recording of a lightning-induced ground fault in phase A of a line and its GAF transformation diagram in an embodiment of the present invention.

[0068] Figure 4 The diagram shows the waveform recording of a wind-induced ground fault in phase B of a certain line and its GAF transformation diagram in an embodiment of the present invention.

[0069] Figure 5 The diagram shows the waveform recording and GAF ​​transformation diagram of a ground fault caused by external force on phase C of a certain line in an embodiment of the present invention.

[0070] Figure 6 A schematic diagram of a multi-objective residual network preferably used in an embodiment of the present invention is shown. Detailed Implementation

[0071] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0072] It should be noted that all directional indicators (such as up, down, left, right, front, back, etc.) in the embodiments of the present invention are only used to explain the relative positional relationship and movement of each component in a specific posture (as shown in the figure). If the specific posture changes, the directional indicator will also change accordingly.

[0073] Furthermore, the use of terms such as "first" and "second" in this invention is for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined with "first" or "second" may explicitly or implicitly include at least one of that feature. Additionally, the technical solutions of the various embodiments can be combined with each other, but only on the basis of being achievable by those skilled in the art. If the combination of technical solutions is contradictory or impossible to implement, such a combination of technical solutions should be considered non-existent and not within the scope of protection claimed by this invention.

[0074] Combination Figure 1 As shown, this invention discloses a transmission line fault identification method based on Gram angle field and residual network, comprising: constructing a transmission line fault identification model; acquiring three-phase waveform data of the secondary side of the transmission line; converting the extracted three-phase waveform data of the secondary side of the transmission line into a two-dimensional color image based on Gram angle field, as a fault identification image; and identifying the obtained fault identification image according to the pre-constructed transmission line fault identification model, so as to output the fault type and fault cause as prediction results, thereby enabling accurate prediction of the fault cause and fault type of the transmission line under test.

[0075] In the method embodiments of this aspect, the transmission line fault identification model is a multi-output convolutional neural network architecture based on a multi-objective residual network, which can synchronously output the fault cause and fault type as prediction results.

[0076] In the above implementation process, the three-phase waveform data of the secondary side of the transmission line is converted into a two-dimensional color image using the Gram angle field method. This two-dimensional color image is then input into a pre-constructed transmission line fault identification model for fault identification. Because the transmission line fault identification model of this invention can simultaneously output the fault cause and fault type as prediction results, it enables simultaneous prediction of the fault cause and fault type of the tested transmission line.

[0077] In this invention, the three-phase waveform data is converted into a two-dimensional color image as a fault identification image using Gram angle field. Furthermore, the transmission line fault identification model is constructed based on a multi-objective residual network. This achieves an unexpected synergistic effect compared to existing methods that rely on signal processing for transmission line fault prediction. The multi-objective residual network itself has the advantage of strong data feature extraction capabilities, enabling self-extraction of features. Moreover, in this application, based on a two-dimensional matrix, the features of the two-dimensional color image converted from the fault waveform data using Gram angle field are further amplified, making the fault features more obvious. This further facilitates the self-extraction of the convolutional neural network based on the multi-residual network. In other words, this application combines a convolutional neural network based on the residual network with Gram angle field. By using the two-dimensional color image based on Gram angle field as the identification feature, the influence of factors such as voltage and current waveforms, fault distance, and transition resistance on fault feature extraction is reduced. This effectively addresses the problem of white noise in the waveform data and has low requirements for voltage and current sampling frequencies.

[0078] In one embodiment of the present invention, combined with Figure 2 As shown, the steps for constructing a transmission line fault identification model include:

[0079] Step S1: Construct the original recognition model;

[0080] Step S2: Obtain three-phase fault recording data from the secondary side of the faulty transmission line;

[0081] Step S3: Based on the Gram angle field, convert the acquired secondary side three-phase fault recording data of each faulty transmission line into a two-dimensional color image as a fault sample image;

[0082] Step S4: Classify the converted fault sample images according to fault type and fault cause, and perform Smote oversampling on the minority fault samples to obtain a fault sample set.

[0083] Step S5: Using the fault sample set based on two-dimensional color images as input, the original identification model is trained and tested to obtain the trained transmission line fault identification model.

[0084] In this invention, in step S1, the original identification model is a multi-output convolutional neural network architecture based on a multi-objective residual network; by means of the multi-output convolutional neural network architecture of the multi-objective residual network, it is possible to construct a fault cause original identification sub-model and a fault type original identification sub-model in the original identification model.

[0085] Specifically, such as Figure 5As shown, the original identification model is a multi-objective residual network model, which is specifically a multi-output convolutional neural network architecture based on multiple residual neural networks. Because a multi-output convolutional neural network architecture is used, it is possible to integrate the transmission line fault type identification sub-model and the transmission line fault cause identification sub-model into a single network architecture.

[0086] In a preferred embodiment, each sub-model includes multiple residual neural network modules, and each residual neural network module includes two identical basic computational units (i.e., Res_a unit and Res_b unit). The basic computational unit includes a convolutional layer conv_1 with a kernel size of 3×3×1 and a stride of 1; a convolutional layer conv_2 with a kernel size of 3×3×1 and a stride of 1; normalization layers bacthnorm_1 and bacthnorm_2; activation function layers relu_1 and relu_2; and a shortcut jump connection layer is connected between the two basic computational units to prevent model overfitting.

[0087] In the process of building the multi-objective residual network model, the initial model includes a convolutional layer conv_1 with an input size of 224×224×3, a kernel size of 7×7×1, a stride of 2, and padding of 3; a normalization layer bacthnorm_1; an activation function layer relu_1; and a max pooling layer MaxPooling_1 with an input size of 64×112×112, a kernel size of 3×3×1, and a stride of 1.

[0088] In the process of building the sub-model for identifying the fault type of transmission line, the sub-model includes a convolutional layer conv_2 with an input size of 64×112×112, a kernel size of 6×6×1, a stride of 2, and padding of 1; a normalization layer bacthnorm_2; an activation function layer relu_2, followed by two consecutive residual neural network modules; and an average pooling layer AvgPooling_1.

[0089] exist Figure 6 In this context, convolutional layers are abbreviated as Cov layers, normalization layers as BN layers, and activation function layers as Re1u layers.

[0090] In the process of building the sub-model for identifying the causes of transmission line faults, the sub-model includes a convolutional layer conv_3 with an input size of 64×112×112, a kernel size of 5×5×1, a stride of 2, and padding of 1; a normalization layer bacthnorm_3; an activation function layer relu_3, followed by four consecutive residual neural network modules; and an average pooling layer AvgPooling_2.

[0091] In a preferred embodiment, a feature fusion layer is constructed between the transmission line fault type identification sub-model and the transmission line fault cause identification sub-model. This fusion layer fuses the feature parameters trained by the two modules. After the parameters trained by each sub-model are fused by the feature fusion layer, the transmission line fault type is identified through the fully connected layer fc_1, the linear layer Linear_1, and the Softmax1 layer of the transmission line fault type identification sub-model. Similarly, the cause of the transmission line fault is identified through the fully connected layer fc_2, the linear layer Linear_2, and the Softmax2 layer of the transmission line fault cause identification sub-model.

[0092] In step S2, the extracted fault recording data comes from the fault recording data of phases A, B, and C of a transmission line in a certain area, totaling 354 sets. The obtained three-phase transmission line fault recording data includes: transient recording data of phase A voltage and current on the secondary side of the transmission line; transient recording data of phase B voltage and current on the secondary side of the transmission line; and transient recording data of phase C voltage and current on the secondary side of the transmission line.

[0093] Furthermore, in step S2, the sampling interval of the fault recording data is set to the period from one week before the fault occurred to two weeks after the fault occurred, so as to obtain the fault recording data required by this method.

[0094] In step S3, which converts the acquired secondary-side three-phase fault recording data of each faulty transmission line into a two-dimensional color image based on the Gram angle field, the following steps are included:

[0095] Step S301: Concatenate the one-dimensional time series arrays of the three-phase voltage and current on the secondary side of the faulty transmission line into a two-dimensional tensor matrix. The expression of this two-dimensional tensor is as follows:

[0096] .

[0097] Step S302: Normalize the transient fault voltage and transient fault current data of the three phases on the secondary side of the faulty transmission line. The normalization formula is as follows:

[0098]

[0099] In the formula: The fault data points are the fault voltages and fault currents collected for phases A, B, and C. These are the electrical quantity vectors corresponding to the fault voltages and fault currents of phases A, B, and C.

[0100] Step S303: Perform a polar coordinate transformation on the normalized data to obtain the radius and angle corresponding to each data point. The formula for the polar coordinate transformation is:

[0101]

[0102] In the formula: This represents the phase angle corresponding to each data point in the two-dimensional tensor matrix; The radius corresponding to each data point; The time number corresponding to each data point in the fault waveform data; N is the total number of sampling points of the acquired fault waveform data.

[0103] Step S304: Using the sum and difference angle relationships, calculate the corresponding Gram angle field transformations of the two-dimensional tensor matrices. The calculation formula used is as follows:

[0104] ;

[0105] ;

[0106] In the formula: I is a unit vector; GASF is the Gram angle and field transformation formula; GADF is the Gram angle difference field transformation formula.

[0107] In step S4, the converted two-dimensional color image is classified according to fault type, resulting in 10 fault types: single-phase ground fault (A phase), single-phase ground fault (B phase), single-phase ground fault (C phase), two-phase faults (A and B phases), two-phase faults (A and C phases), two-phase faults (B and C phases), two-phase ground faults (A and B phases), two-phase ground faults (A and C phases), two-phase ground faults (B and C phases), and three-phase faults (ABC phases). The converted two-dimensional color image is also classified according to fault cause, resulting in 9 fault causes: lightning strike, foreign object fault, wildfire fault, wind-induced fault, icing fault, external force damage fault, bird damage fault, pollution flashover fault, and tree flashover fault. (The appendix is ​​not included in the original text.) Figures 3-5 As an example, fault waveform data from lightning strikes, wind deflection, and external force damage, along with their converted GAF ​​diagrams, are shown and used as input for subsequent deep learning.

[0108] Step S4 also includes a step of performing Shote oversampling on minority class fault samples, which can reduce class imbalance between data. This step includes: for faults of small sample types in the fault dataset, converting their classified color images into pixel matrices; using the Shote oversampling algorithm to calculate the Euclidean distance from each sample in the minority class to all samples in the minority class sample set; calculating the k nearest neighbors of the minority class samples based on the Euclidean distance; randomly selecting a neighbor from the k nearest neighbors; selecting any position on the straight line between the neighbor and the minority class sample points; and generating a new fault sample using linear interpolation.

[0109] Step S5 also includes setting model training parameters. Preferably, the optimizer for local model training uses the SGD algorithm, and the loss function is the cross-entropy loss function combined with the Softmax classifier. The training step size is set to 64, and the initial learning rate is set to 0.001.

[0110] Step S5 also includes training the model, including: randomly selecting 70% of the fault type and fault cause identification sample set as the training sample set, and the remainder as the test sample set, inputting it as the target domain data into the multi-output convolutional neural network model, training with fault type and fault cause as output, setting a maximum number of training iterations and determining whether the loss function converges. If the loss function converges, the model training can be completed after reaching the maximum number of training iterations. If the loss function does not converge, the maximum number of training iterations is increased until convergence, and finally the fault type and fault cause identification model is obtained.

[0111] Embodiments of the present invention also disclose a transmission line fault identification device based on Gram angle field and residual network, comprising:

[0112] The acquisition unit is used to acquire the three-phase waveform recording data of the secondary side of the transmission line;

[0113] The Gram angle field conversion unit is used to convert the extracted secondary three-phase waveform data of the transmission line into a two-dimensional color image based on the Gram angle field, which can be used as a fault identification image.

[0114] The transmission line fault identification unit is used to identify the obtained fault identification graphics according to the pre-built transmission line fault identification model. The transmission line fault identification model is a multi-output convolutional neural network architecture based on a multi-objective residual network, which can output the fault cause and fault type as prediction results simultaneously.

[0115] The transmission line fault identification device based on Gram angle field and residual network of the present invention further includes:

[0116] The model building unit is used to build the original identification model, which is a multi-output convolutional neural network architecture based on a multi-objective residual network.

[0117] The second acquisition unit is used to acquire the secondary side three-phase fault recording data of the faulty transmission line;

[0118] The Gram angle field conversion unit is also used to convert the acquired secondary side three-phase fault recording data of each faulty transmission line into a two-dimensional color image based on the Gram angle field, as a fault sample image.

[0119] The fault sample classification unit is used to classify the converted fault sample images according to fault type and fault cause, and to perform Smote oversampling processing on the minority class fault samples to obtain the fault sample set.

[0120] The training unit is used to train and test the original identification model by taking a fault sample set based on two-dimensional color images as input, so as to obtain the trained transmission line fault identification model.

[0121] Embodiments of the present invention also disclose an electronic device, the electronic device including a memory and a processor, the memory for storing a computer program, the processor running the computer program to enable the electronic device to perform the transmission line fault identification method based on Gram angle field and residual network as described above.

[0122] Embodiments of the present invention also disclose a computer-readable storage medium storing computer program instructions. When the computer program instructions are read and executed by a processor, the transmission line fault identification method based on Gram angle field and residual network in the first aspect of the present invention is performed.

[0123] In the several embodiments provided in this application, it should be understood that the disclosed apparatus and methods can also be implemented in other ways. The apparatus embodiments described above are merely illustrative; for example, the flowcharts and block diagrams in the accompanying drawings show the architecture, functions, and operations that may be implemented according to the embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order than those marked in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0124] In addition, the functional modules in the various embodiments of this application can be integrated together to form an independent part, or each module can exist independently, or two or more modules can be integrated to form an independent part.

[0125] If a function is implemented as a software module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a portable hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0126] The above are merely embodiments of this application and are not intended to limit the scope of protection of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of protection of this application. It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures.

[0127] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0128] It should be noted that, in this document, relational terms such as "first" and "second" are used merely to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes the element. Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications and substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for identifying transmission line faults based on Gram angle field and residual network, characterized in that, Includes the following steps: Acquire three-phase waveform data from the secondary side of the transmission line; Based on the Gram angle field, the three-phase waveform data of the secondary side of the transmission line are converted into two-dimensional color images as fault identification images; The fault identification graphics obtained are identified according to the pre-constructed transmission line fault identification model. The transmission line fault identification model is a multi-output convolutional neural network architecture based on a multi-objective residual network, which can output the fault cause and fault type as prediction results simultaneously. Before acquiring the three-phase waveform data of the transmission line, the process also includes the step of constructing a fault identification model for the transmission line, including: constructing an original identification model, which is a multi-output convolutional neural network architecture based on a multi-objective residual network; Acquire three-phase fault waveform data on the secondary side of the faulted transmission line, including transient fault voltage waveform data and transient fault current waveform data of the three phases on the secondary side of the transmission line. Based on the Gram angle field, the secondary side three-phase fault recording data of each faulty transmission line is converted into a two-dimensional color image as a fault sample image; The converted fault sample images are classified according to fault type and fault cause, and Smote oversampling is performed on the minority fault samples to obtain the fault sample set. Using a fault sample set based on two-dimensional color images as input, the original identification model is trained and tested to obtain the trained transmission line fault identification model. The step of converting fault recording data into a two-dimensional color image based on the Gram angle field includes: The one-dimensional time series arrays of three-phase voltages and currents on the secondary side of the faulty transmission line are concatenated into a two-dimensional tensor matrix, the expression of which is: ;in, The fault data points are the fault voltages and fault currents collected for phases A, B, and C. Based on the two-dimensional tensor matrix, the transient recording data of fault voltage and transient recording data of fault current of the three phases on the secondary side of the faulted transmission line are normalized. The normalized data is then transformed into polar coordinates to obtain the radius and angle corresponding to each data point in the two-dimensional tensor matrix. The formula for the polar coordinate transformation is as follows: In the formula: This represents the phase angle corresponding to each data point in the two-dimensional tensor matrix; The radius corresponding to each data point; The time number corresponding to each data point in the fault waveform data; N is the total number of sampling points in the acquired fault waveform data; Using sum and difference angle relationships, the corresponding Gram angle field transformations are calculated for two-dimensional tensor matrices. The calculation formulas used are as follows: ； In the formula: Let I be the phase angle value corresponding to each data point in each electrical quantity, and let I be the unit vector; GASF is the Gram angle and field transformation formula; GADF is the Gram angle difference field transformation formula. , These are the electrical quantity vectors corresponding to the fault voltages and fault currents of phases A, B, and C.

2. The transmission line fault identification method based on Gram angle field and residual network according to claim 1, characterized in that, In the step of normalizing the transient fault voltage and transient fault current data of the three phases on the secondary side of the faulty transmission line, the normalization formula is as follows: 。 3. The transmission line fault identification method based on Gram angle field and residual network according to claim 1, characterized in that, The steps involved in performing Smote oversampling on minority fault samples include: The classified fault transient waveform image is converted into a pixel matrix. The Smote oversampling algorithm is used to calculate the Euclidean distance from each minority fault sample in the minority fault sample set to all samples in the minority fault sample set. The k nearest neighbors of the minority fault sample are calculated based on the Euclidean distance. Then, a neighbor is randomly selected from the k nearest neighbors. On the straight line between the neighbor and the minority fault sample, any position is selected, and a new fault sample is generated by linear interpolation.

4. The transmission line fault identification method based on Gram angle field and residual network according to claim 1, characterized in that, The convolutional neural network of the original identification model adopts a multi-output convolutional neural network architecture based on multiple residual network modules, which enables the original identification model to have original identification sub-models for fault types and original identification sub-models for fault causes. A feature fusion layer is built between the original fault type identification sub-model and the original fault cause identification sub-model to fuse the feature parameters obtained from the training of the two sub-models; with the help of the fusion parameters implemented based on the feature fusion layer, the original fault type identification sub-model and the original fault cause identification sub-model achieve their respective outputs.

5. The transmission line fault identification method based on Gram angle field and residual network according to claim 4, characterized in that, In the multi-output convolutional neural network architecture with multiple residual network modules, each residual neural network module contains two identical basic computational units, and the two basic computational units are connected by a jump connection layer that can prevent the model from overfitting.

6. The transmission line fault identification method based on Gram angle field and residual network according to claim 1, characterized in that, In the steps of training and testing the original identification model, a maximum number of training iterations is set and it is determined whether the model's loss function has converged. If the loss function has converged, the training of the model can be completed after reaching the maximum number of training iterations. If the loss function has not converged, the maximum number of training iterations is increased until convergence is achieved, and finally the transmission line fault identification model is obtained.

7. An apparatus for implementing the transmission line fault identification method based on Gram angle field and residual network according to any one of claims 1-6, characterized in that, include: The acquisition unit is used to acquire three-phase waveform recording data of the secondary side of the transmission line; The Gram angle field conversion unit is used to convert the extracted secondary three-phase waveform data of the transmission line into a two-dimensional color image based on the Gram angle field, which can be used as a fault identification image. The transmission line fault identification unit is used to identify the obtained fault identification graphics according to the pre-built transmission line fault identification model. The transmission line fault identification model is a multi-output convolutional neural network architecture based on a multi-objective residual network, which can output the fault cause and fault type as prediction results simultaneously. This also includes: The model building unit is used to build the original identification model, which is a multi-output convolutional neural network architecture based on a multi-objective residual network. The second acquisition unit is used to acquire the secondary side three-phase fault recording data of the faulty transmission line; The Gram angle field conversion unit is also used to convert the acquired secondary side three-phase fault recording data of each faulty transmission line into a two-dimensional color image based on the Gram angle field, as a fault sample image. The fault sample classification unit is used to classify the converted fault sample images according to fault type and fault cause, and to perform Smote oversampling processing on the minority class fault samples to obtain the fault sample set. The training unit is used to train and test the original identification model by taking a fault sample set based on two-dimensional color images as input, so as to obtain the trained transmission line fault identification model.

8. An electronic device, characterized in that, The electronic device includes a memory and a processor. The memory stores a computer program, and the processor runs the computer program to enable the electronic device to perform the transmission line fault identification method based on Gram angle field and residual network as described in any one of claims 1 to 6.

Images