Research On Classification And Recognition Of Voltage Sag Based On Balancing Generative Adversarial Network

In recent years,with the development of new energy sources and the construction of new power systems,the operation and control of power systems are facing many challenges.Among them,voltage sag,as one of the most serious power quality disturbances,has received a lot of attention.At present,the voltage sag classification and identification model based on artificial feature extraction method and artificial intelligence method has been widely used in the field of voltage sag identification.However,in the actual production and operation process,the distribution of voltage sag events is very random.Due to the influence of region and time,there are serious category imbalances in the data sets of various monitoring systems.Seriously affected the engineering application effect of the existing method.To solve the above problems,this paper proposes a data augmentation algorithm based on balancing generative adversarial networks.This algorithm combines the advantages of the autoencoder and the generative adversarial network,learns the public knowledge in the data set through the initialization training of the autoencoder,and uses the characteristics of the majority class samples to assist in the generation of minority class samples.It can effectively solve the problem of unbalanced distribution of voltage sag samples.The balancing generative adversarial network model is designed for the sample generation of unbalanced data sets,which ensures the authenticity and diversity of generated samples.For the sample generation of voltage sag data,this paper further optimizes the model.Firstly,an embedding layer is introduced in the autoencoder architecture to add label information to the feature vectors,which solves the problem of strong coupling of the voltage sag eigenvector distribution,improves the discreteness of the distribution of sag samples in the feature space,and enhances the generation effect of samples.Secondly,the Wasserstein distance optimized by the gradient penalty term is used to guide the model confrontation training,which overcomes the problems of the original model’s gradient disappearance and mode collapse.Finally,the convolutional attention module is integrated in the discriminator of the generative adversarial network to improve the feature extraction ability and the recognition effect of the discriminator.Through experimental analysis,the improved model can fully extract the characteristics of minority samples under the unbalanced data set,ensure the distribution consistency between the generated samples and the original samples,and take into account the authenticity and diversity of the generated samples.At the engineering application level,in order to verify that the enhanced dataset based on the balancing generative adversarial network model can provide data support for the deep learning classification model,this paper uses the current mainstream deep learning model as the classifier to verify the impact of the enhanced dataset classifier on the recognition accuracy.A classifier model based on convolutional neural network and long short-term memory network is designed.After experimental verification,the classification recognition accuracy of the model has been effectively improved under the enhanced data set.The model in this paper can provide strong data support,improve the recognition effect of the deep learning model on minority classes,and improve the overall recognition accuracy.