Disturbance identification method based on Gramb angle difference field and transfer learning

Abstract

The invention relates to the technical field of disturbance identification of a power system, in particular to a disturbance identification method based on a Grubrum angle difference field and transfer learning, which comprises the following steps of: acquiring corresponding frequency and voltage amplitude time sequence data when a node of the power system IEEE39 is disturbed to obtain frequency and voltage amplitude data of a one-dimensional time sequence; the method comprises the following steps: respectively carrying out feature extraction and conversion into two-dimensional Grubrum angle field images, fusing feature extraction results based on an RGB image channel, inputting a feature two-dimensional image data set into a preset transfer learning-AlexNet network model for training, and constructing a disturbance recognition model based on an SE attention mechanism; and inputting the obtained current PMU disturbance data of the power system into the disturbance recognition model to obtain a disturbance recognition result output by the disturbance recognition model, so that the recognition capability and the anti-noise capability of the model can be improved, and the recognition precision of different disturbance types of the power system is further improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system disturbance identification, and in particular to a disturbance identification method based on Gram angle difference field and transfer learning. Background Art

[0002] With the increasing penetration of renewable energy and the integration of distributed generation and microgrids, modern power systems have shown high reliability and security, scalability and flexibility. At the same time, due to their large scale, complex structure and diverse equipment, they are also highly complex and nonlinear, making the types of power system disturbances complex and diverse, and the occurrence of disturbances uncertain. Accurately identifying the types of power system disturbances will help improve system protection and control strategies and enhance the robustness of the system.

[0003] In the existing technology, distribution network synchronized phasor measurement devices are used for disturbance identification. This method is usually based on frequency domain analysis, time domain feature extraction, etc. This type of method can directly obtain feature quantities with clear physical meanings, but in practical applications, the presence of noise may cause distortion of feature extraction results; at the same time, the feature representation of the existing technology is insufficient for the nonlinear and non-stationary characteristics in the disturbance data.

[0004] With the development of artificial intelligence technology, deep learning has been widely used in classification tasks. However, the classification training process of conventional deep neural networks in existing technologies usually requires a large amount of labeled sample data, and training the model from scratch requires a lot of resources. However, existing technologies lack means to efficiently obtain sample resources. In addition, existing technologies have poor noise resistance, resulting in low accuracy in disturbance recognition. Summary of the Invention

[0005] The purpose of the present invention is to provide a disturbance identification method based on Gram angle difference field and transfer learning, so as to solve the problems existing in the prior art of low efficiency in acquiring sample resources, poor anti-noise ability and low disturbance identification accuracy.

[0006] To achieve the above object, the present invention provides a disturbance identification method based on Gram angle difference field and transfer learning, comprising the following steps: S1. Obtain the frequency and voltage amplitude time series data corresponding to the disturbance of the IEEE39 node in the power system, and perform normalization preprocessing on the frequency and voltage amplitude time series data to obtain the frequency and voltage amplitude data of the one-dimensional time series; S2. Extract features from the frequency and voltage amplitude data of the one-dimensional time series and convert them into a two-dimensional Gram angle field image. Perform feature fusion on the feature extraction results based on the RGB image channels to obtain feature two-dimensional image data. S3. Input the feature two-dimensional image dataset into the preset transfer learning-AlexNet network model for training, and build a perturbation recognition model based on the SE attention mechanism; S4. Input the acquired current PMU disturbance data of the power system into the disturbance identification model to obtain a disturbance identification result output by the disturbance identification model.

[0007] In some embodiments of the present application, in S1, normalizing and preprocessing the frequency and voltage amplitude time series data includes: S11, get n The frequency and voltage amplitude time series data composed of the values at each moment , and perform normalization preprocessing on it. The expression is: ; Among them, the frequency and voltage amplitude data of the one-dimensional time series is the data point in the original time series, is the data point in the new time series obtained after normalization, X for The frequency and voltage amplitude time series data are composed of the values at each moment.

[0008] In some embodiments of the present application, in S2, converting the one-dimensional time series frequency and voltage amplitude data into a two-dimensional Gram angle field image includes: S21. Convert the data points in the new time series into angles and encode the timestamps into radius. The expression is: ; in, is the normalized time series, for The polar angle after the arc cosine function transformation is: for The polar radius of a point, for The timestamp of the point, is the constant factor of the regularized polar coordinate system; Get the transformed time series ,in, is the data point in the transformed time series; S22. Calculate the Gram matrix based on the phase angle difference of the converted time series G GADF , the calculation formula is: ; is the Gram matrix, For time series Point andj point angle difference; Based on the Gram matrix G GADF Get the two-dimensional Gram angle field image.

[0009] In some embodiments of the present application, in S2, performing feature fusion on the feature extraction results based on the RGB image channels includes: S23. Input the feature extraction result into the R channel and the B channel of the RGB image for feature fusion. The feature extraction result includes voltage and frequency data after the Gram angle difference field changes.

[0010] In some embodiments of the present application, in S3, constructing a perturbation recognition model based on the training results and the SE attention mechanism includes: S31. Based on the SE attention mechanism, feature weights are assigned to the feature channels in the training results to obtain global information. The expression is: ; in, z c For the c Global information of each channel, H is the height of the feature map, W is the width of the feature map, is the feature map, C is the number of channels; S32. Based on the first fully connected layer of the SE attention mechanism and the ReLU activation function, the channel feature value is calculated and expressed as: ; in, W 1 is the weight of the first fully connected layer, b 1 is the bias of the first fully connected layer, z is the global description vector of all channels in the first layer, s is the channel characteristic value; S33, based on the SE attention mechanism, the second fully connected layer and the sigmoid activation function are used to calculate the channel weight, which is expressed as: ; in, W 2 is the weight of the second fully connected layer, b 2 is the bias of the second fully connected layer, σ is the Sigmoid activation function, is the channel weight; S34. Optimize the feature channels of the training results based on the channel weights, and build a disturbance recognition model based on the optimization results.

[0011] In some embodiments of the present application, S34 further includes: S341. Use the improved cross entropy loss function FocalLoss to optimize the sample weights. The expression is: ; in, p t is the probability of the predicted target category; α t is the balance factor; γ is the regulating factor, is the improved cross entropy loss function; S342. Use the CenterLoss algorithm to optimize the distance of features within the model class. The expression is: ; in, For the The characteristics of the samples, For the The center of the sample class, N is the batch value, is the center loss function.

[0012] In some embodiments of the present application, in S4, the obtained current PMU disturbance data of the power system is input into the disturbance identification model, and the disturbance identification result output by the disturbance identification model includes: S41, obtaining voltage amplitude data and frequency data of the current PMU disturbance of the power system; S42, normalizing the voltage amplitude data and frequency data of the current PMU disturbance and performing Gram angle difference field image conversion and feature fusion in sequence to obtain an image to be recognized; S43: Input the image to be recognized into the disturbance recognition model, obtain a confidence score output by the disturbance recognition model, and obtain a disturbance recognition result based on the confidence score.

[0013] The advantages and beneficial effects of the present invention over the prior art are: 1. By converting one-dimensional time series data into two-dimensional Gram angular field images, the present invention can more intuitively reflect the dynamic relationship and pattern between frequency and voltage amplitude data, better capture the features in the data, and thus provide more valuable information for subsequent feature extraction and model training, thereby improving the accuracy of disturbance identification. By combining the SE attention mechanism with the transfer learning-AlexNet network model to jointly construct a disturbance identification model and apply it to the disturbance identification task, it can fully utilize the feature extraction advantages of the pre-trained model and pay more attention to the feature areas that are important for disturbance identification, thereby improving the recognition ability and noise resistance of the model, and further improving the recognition accuracy of different disturbance types in the power system.

[0014] 2. The present invention obtains different types of power system simulation data, such as simulation data of power grid node systems of different scales and simulation data under different operating conditions, integrates and processes multiple data to construct a more comprehensive sample data set, which can make full use of various data resources and improve the efficiency of obtaining sample resources.

[0015] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a flow chart of a disturbance identification method based on Gram angle difference field and transfer learning according to the present invention; Figure 2 Schematic diagram comparing the accuracy of disturbance identification in an embodiment of the present invention with that of the other two methods; Figure 3 Schematic diagram comparing the loss rate of disturbance identification in an embodiment of the present invention with two other methods; Figure 4 Schematic diagram of the frequency and voltage amplitude of a PMU in an actual power system according to an embodiment of the present invention; Figure 5 Schematic diagram of a two-dimensional Gram angular field image in an embodiment of the present invention; Figure 6 Schematic diagram of an RGB image after feature fusion in an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In the description of the present invention, it should be noted that the terms "upper", "lower", "inside", "outside" and the like indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, or the orientations or positional relationships in which the inventive product is usually placed when in use. These are only for the convenience of describing the present invention and simplifying the description, and are not intended to indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation. Therefore, they cannot be understood as limitations on the present invention. In the description of the present invention, it should also be noted that, unless otherwise expressly specified and limited, the terms "setting", "installation" and "connection" should be understood in a broad sense. For example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or an indirect connection through an intermediate medium, or it can be a communication between the internal parts of two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0018] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0019] like Figure 1 As shown, the present invention provides a disturbance identification method based on Gram angle difference field and transfer learning, comprising the following steps: S1. Obtain the frequency and voltage amplitude time series data corresponding to the disturbance of the IEEE39 node in the power system, and perform normalization preprocessing on the frequency and voltage amplitude time series data to obtain the frequency and voltage amplitude data of the one-dimensional time series.

[0020] S2. Perform feature extraction on the frequency and voltage amplitude data of the one-dimensional time series and convert them into a two-dimensional Gram angle field image. Perform feature fusion on the feature extraction results based on the RGB image channels to obtain feature two-dimensional image data.

[0021] S3. Input the feature two-dimensional image dataset into the preset transfer learning-AlexNet network model for training, and build a perturbation recognition model based on the SE attention mechanism.

[0022] Optionally, build the basic structure of the AlexNet network. The basic structure of the AlexNet network includes 5 convolutional layers, 3 maximum pooling layers, 3 fully connected layers, and 1 softmax output layer. In the basic structure of the AlexNet network, the SE attention mechanism is integrated to adaptively assign higher weights to important feature channels.

[0023] S4. Input the acquired current PMU disturbance data of the power system into the disturbance identification model to obtain a disturbance identification result output by the disturbance identification model.

[0024] The advantages and beneficial effects of the present invention over the prior art are: 1. By converting one-dimensional time series data into two-dimensional Gram angular field images, the present invention can more intuitively reflect the dynamic relationship and pattern between frequency and voltage amplitude data, better capture the features in the data, and thus provide more valuable information for subsequent feature extraction and model training, thereby improving the accuracy of disturbance identification. By combining the SE attention mechanism with the transfer learning-AlexNet network model to jointly construct a disturbance identification model and apply it to the disturbance identification task, it can fully utilize the feature extraction advantages of the pre-trained model and pay more attention to the feature areas that are important for disturbance identification, thereby improving the recognition ability and noise resistance of the model, and further improving the recognition accuracy of different disturbance types in the power system.

[0025] 2. The present invention obtains different types of power system simulation data, such as simulation data of power grid node systems of different scales and simulation data under different operating conditions, integrates and processes multiple data to construct a more comprehensive sample data set, which can make full use of various data resources and improve the efficiency of obtaining sample resources.

[0026] In some embodiments of the present application, in S1, normalizing and preprocessing the frequency and voltage amplitude time series data includes: S11, get The frequency and voltage amplitude time series data composed of the values at each moment , and perform normalization preprocessing on it. The expression is: ; Among them, the frequency and voltage amplitude data of the one-dimensional time series is the data point in the original time series, is the data point in the new time series obtained after normalization, X for The frequency and voltage amplitude time series data are composed of the values at each moment.

[0027] In some embodiments of the present application, in S2, converting the one-dimensional time series frequency and voltage amplitude data into a two-dimensional Gram angle field image includes: S21. Convert the data points in the new time series into angles and encode the timestamps into radius. The expression is: ; in, is the normalized time series, for The angle after the arc cosine function transformation is for The polar radius of a point, for The timestamp of the point, is the constant factor of the regularized polar coordinate system; Get the transformed time series ,in, is the data point in the transformed time series; S22. Calculate the Gram matrix based on the phase angle difference of the converted time series G GADF , the time correlation within different time intervals is expressed by the trigonometric difference between each point, and the calculation formula is: ; is the Gram matrix; the angle difference describes the relative positions of two points in the time series on the unit circle; and It reflects the influence of angle difference on the relationship between time series sample points, and its value fluctuates within [-1,1]. , the matrix is antisymmetric and has zero diagonals. In the GADF transformation process, the data mapping is one-to-one at each stage, and the matrix construction depends on the angle of the sequence sample point, which is uniquely determined by the value of the time series sample point. Therefore, the temporal information and nonlinear relationship of the perturbed time series are preserved; Based on the Gram matrix G GADF Get the two-dimensional Gram angle field image.

[0028] In some embodiments of the present application, in S2, performing feature fusion on the feature extraction results based on the RGB image channels includes: S23. Input the feature extraction result into the R channel and the B channel of the RGB image for feature fusion. The feature extraction result includes voltage and frequency data after the Gram angle difference field changes.

[0029] In some embodiments of the present application, in S3, constructing a perturbation recognition model based on the training results and the SE attention mechanism includes: S31. Build the basic structure of the AlexNet network. The basic structure of the AlexNet network includes 5 convolutional layers, 3 maximum pooling layers, 3 fully connected layers and 1 Softmax output layer.

[0030] Based on the basic structure of the AlexNet network, the SE attention mechanism is used to assign feature weights to feature channels in the training results, adaptively assigning higher weights to important feature channels, and obtaining global information. The expression is: ; in, z c For the c Global information of each channel, His the height of the feature map, W is the width of the feature map, is the feature map, C is the number of channels; S32. Based on the first fully connected layer of the SE attention mechanism and the ReLU activation function, the channel feature value is calculated and expressed as: ; in, W 1 is the weight of the first fully connected layer, b 1 is the bias of the first fully connected layer, z is the global description vector of all channels in the first layer, s is the channel characteristic value; S33, based on the SE attention mechanism, the second fully connected layer and the sigmoid activation function are used to calculate the channel weight, which is expressed as: ; in, W 2 is the weight of the second fully connected layer, b 2 is the bias of the second fully connected layer; σ Sigmoid activation function compresses the output into the range of [0,1]; is the channel weight; S34. Optimize the feature channels of the training results based on the channel weights, and build a disturbance recognition model based on the optimization results.

[0031] In some embodiments of the present application, S34 further includes: Specifically, before step S341, the channel weights obtained above are w It is used to reweight each channel of the original input feature map, that is, to "adjust" the importance of each channel by multiplying the output of each channel by its corresponding weight coefficient. The specific process is as follows: ; in, is the attention coefficient, i.e. channel c The channel weights, X c is the channel eigenvalue; S341. Considering the problem of sample imbalance, in order to enhance the weight of difficult-to-classify samples and reduce the weight of easy-to-classify samples, the improved cross entropy loss function FocalLoss is adopted: ; in, p tis the probability of the target category predicted by the model; α It is a balancing factor that adjusts the class imbalance problem; γ is an adjustment factor used to reduce the focus on simple samples. γ =0, FocalLoss degenerates into ordinary cross entropy loss.

[0032] S342. In order to further constrain the aggregation of intra-class features and optimize the feature space structure, the CenterLoss algorithm is used to optimize the distance of intra-class features of the model. The expression is: ; in, For the The characteristics of the samples, For the The center of the sample class, N is the batch value, is the center loss function.

[0033] Among them, the batch training method is used in the training process, the batch size is 32, the data set is randomly distributed into training set and test set, the stochastic gradient descent method is selected as the optimization function, the momentum is 0.9, L2 regularization is set to prevent overfitting, the number of iterations is 25, the initial learning rate is 0.0001, and the learning rate is adjusted every 10 epochs. Each time the learning rate is adjusted, the current learning rate is multiplied by 0.5. To suppress overfitting, the dropout layer dropout probability is set to 0.5, that is, 50% of the nodes are randomly dropped.

[0034] Fine-tune the AlexNet model trained on the ImageNet massive dataset, adjust the classification output layer to the target task, fine-tune the previous structure, and train it according to the aforementioned structure and parameters to achieve the new image recognition task to obtain a trained model.

[0035] In some embodiments of the present application, in S4, the obtained current PMU disturbance data of the power system is input into the disturbance identification model, and the disturbance identification result output by the disturbance identification model includes: S41, obtaining voltage amplitude data and frequency data of the current PMU disturbance of the power system; S42, normalizing the voltage amplitude data and frequency data of the current PMU disturbance and performing Gram angle difference field image conversion and feature fusion in sequence to obtain an image to be recognized; S43: Input the image to be recognized into the disturbance recognition model, obtain a confidence score output by the disturbance recognition model, and obtain a disturbance recognition result based on the confidence score.

[0036] To verify the effectiveness of the present invention, the present invention is compared with the AlexNet network and the CNN convolutional neural network without transfer learning. Under the same data set and the same related hyperparameters, the accuracy curves and loss value curves of the validation set obtained by training the three models are compared. Figure 2 and Figure 3 As shown in the figure, at the beginning of training, the accuracy of TL-AlexNet was 79.95%, while that of CNN was 51.04%, and that of AlexNet was 48.96%. TL-AlexNet had a higher starting point accuracy because the transfer learning network already had initial weights and biases, eliminating the need for retraining from scratch. By epoch 5, the accuracy of TL-AlexNet had reached 98.69%, while that of CNN was 79.17%, and that of AlexNet was 85.68%. Because AlexNet has deeper layers than CNN, it converged faster. At epoch 9, the accuracy of TL-AlexNet stabilized at 99.48%, while that of CNN stabilized at 91.67% at epoch 15, and that of AlexNet stabilized at 92.45% at epoch 12. In summary, AlexNet outperformed CNN convolutional neural networks. Compared to AlexNet without transfer learning, AlexNet using transfer learning had a higher starting point accuracy and converged faster and more accurately during iterations.

[0037] In order to verify the accuracy of the present invention, the method of the present invention is tested in combination with a specific disturbance analysis. The disturbance comes from an actual power grid PMU, and its voltage amplitude and frequency data are as follows: Figure 4 As shown, it is normalized and converted into Figure 5 The two-dimensional Gram angle difference field image is shown, and then it is placed in the R channel and G channel of the RGB image for feature fusion, as shown in Figure 6 As shown, the feature map is sent to the trained model for recognition, and the confidence scores are shown in Table 1. The highest confidence score of 0.982590 corresponds to the disturbance type of cutting. The recognition result of the disturbance recognition method of the present invention is consistent with the actual disturbance type.

[0038] Table 1 Confidence scores ;

[0039] The beneficial effects of the present invention are as follows: the present invention adopts the above-mentioned disturbance identification method based on Gram angle difference field and transfer learning-AlexNet, and on the basis of considering the small number of disturbance samples and the imbalance of disturbance sample categories, proposes a disturbance identification method that can capture complex disturbance patterns, has good noise resistance and strong generalization, improves the disturbance identification accuracy, and solves the problems existing in the prior art of not considering the insufficient disturbance samples of the actual power system, large noise, and low calculation accuracy.

[0040] In this application, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application belongs. In the event of any inconsistency, the meaning described in this specification or the meaning derived from the contents recorded in this specification shall prevail. In addition, the terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0041] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit the same. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solutions of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.

Claims

1. A disturbance identification method based on Gram angle difference field and transfer learning, characterized in that: The following steps are involved: S1. Obtain the frequency and voltage amplitude time series data corresponding to the disturbance of the IEEE39 node in the power system, and perform normalization preprocessing on the frequency and voltage amplitude time series data to obtain the frequency and voltage amplitude data of the one-dimensional time series; S2. Extract features from the frequency and voltage amplitude data of the one-dimensional time series and convert them into a two-dimensional Gram angle field image. Perform feature fusion on the feature extraction results based on the RGB image channels to obtain feature two-dimensional image data. S3. Input the feature two-dimensional image dataset into the preset transfer learning-AlexNet network model for training, and build a perturbation recognition model based on the SE attention mechanism; S4. Input the acquired current PMU disturbance data of the power system into the disturbance identification model to obtain a disturbance identification result output by the disturbance identification model.

2. The disturbance identification method based on Gram angle difference field and transfer learning according to claim 1, characterized in that: In S1, normalization preprocessing of the frequency and voltage amplitude time series data includes: S11, get The frequency and voltage amplitude time series data composed of the values at each moment , and perform normalization preprocessing on it. The expression is: ； Among them, the frequency and voltage amplitude data of the one-dimensional time series is the data point in the original time series, is the data point in the new time series obtained after normalization, X for The frequency and voltage amplitude time series data are composed of the values at each moment.

3. The disturbance identification method based on Gram angle difference field and transfer learning according to claim 2, characterized in that: In S2, converting the frequency and voltage amplitude data of the one-dimensional time series into a two-dimensional Gram angle field image includes: S21. Convert the data points in the new time series into angles and encode the timestamps into radius. The expression is: ； in, is the normalized time series, for The angle after the arc cosine function transformation is for The polar radius of a point, for The timestamp of the point, is the constant factor of the regularized polar coordinate system; Get the transformed time series ,in, is the data point in the transformed time series; S22. Calculate the Gram matrix based on the phase angle difference of the converted time series G GADF , the calculation formula is: ； is the Gram matrix, For time series Point and j point angle difference; Based on the Gram matrix G GADF Get the two-dimensional Gram angle field image.

4. The disturbance identification method based on Gram angle difference field and transfer learning according to claim 3, characterized in that: In S2, performing feature fusion on the feature extraction results based on the RGB image channels includes: S23. Input the feature extraction result into the R channel and the B channel of the RGB image for feature fusion, wherein the feature extraction result includes the voltage and frequency data after the Gram angle difference field changes.

5. The disturbance identification method based on Gram angle difference field and transfer learning according to claim 4, characterized in that: In S3, the perturbation recognition model based on the SE attention mechanism is constructed, including: S31. Based on the SE attention mechanism, feature weights are assigned to the feature channels in the training results to obtain global information. The expression is: ； in, z c For the c Global information of each channel, H is the height of the feature map, W is the width of the feature map, is the feature map, C is the number of channels; S32. Based on the first fully connected layer of the SE attention mechanism and the ReLU activation function, the channel feature value is calculated and expressed as: ； in, W 1 is the weight of the first fully connected layer, b 1 is the bias of the first fully connected layer, z is the global description vector of all channels in the first layer, s is the channel eigenvalue; S33, based on the SE attention mechanism, the second fully connected layer and the sigmoid activation function are used to calculate the channel weight, which is expressed as: ； in, W 2 is the weight of the second fully connected layer, b 2 is the bias of the second fully connected layer, σ is the Sigmoid activation function, w is the channel weight; S34. Optimize the feature channels of the training results based on the channel weights, and build a disturbance recognition model based on the optimization results.

6. The disturbance identification method based on Gram angle difference field and transfer learning according to claim 5, characterized in that: The S34 further includes: S341. Use the improved cross entropy loss function Focal Loss to optimize the sample weights. The expression is: ； in, p t is the probability of the predicted target category; α t is the balance factor; γ is the regulating factor, is the improved cross entropy loss function; S342. Use the CenterLoss algorithm to optimize the distance of features within the model class. The expression is: ； in, For the The characteristics of the samples, For the The center of the sample class, N is the batch value, is the center loss function.

7. The disturbance identification method based on Gram angle difference field and transfer learning according to claim 6, characterized in that: In S4, the obtained current PMU disturbance data of the power system is input into the disturbance identification model, and the disturbance identification result output by the disturbance identification model includes: S41, obtaining voltage amplitude data and frequency data of the current PMU disturbance of the power system; S42, normalizing the voltage amplitude data and frequency data of the current PMU disturbance and performing Gram angle difference field image conversion and feature fusion in sequence to obtain an image to be recognized; S43: Input the image to be recognized into the disturbance recognition model, obtain a confidence score output by the disturbance recognition model, and obtain a disturbance recognition result based on the confidence score.

Images