CAE-based classification method of electric power business

Abstract Aiming at the problem that abnormal data in the business classification of power data communication networks will reduce the business classification accuracy, a classification method based on convolutional autoencoder combined with lightGBM was proposed. First, the flow features of the electric power business were extracted through the convolutional autoencoder. Then, anomaly detection was realized according to the relationship between the reconstruction loss of the convolutional autoencoder and the set threshold. Finally, the electric power business was classified through the LightGBM classifier. The Moore data set used for simulation verification. The results show that the proposed method can effectively detect abnormalities, thereby improving the accuracy of the electric power business classification.