Intelligent Intrusion Detection of Measurement Automation System Based on Deep Learning

Abstract

In the smart grid with measurement automation system, the intrusion detection system judges the intrusion event by analyzing the transmission data in the grid. Aiming at the characteristics of traditional intrusion detection, such as loss of features, low detection efficiency and poor adaptability, an intrusion detection method based on stacked denoising convolutional autoencoders is proposed, which combine convolutional neural network and denoising autoencoder to strengthen feature recognition ability, using Dropout and regularization methods to prevent overfitting, and using Adam algorithm to obtain optimal parameters. Finally, the NSL-KDD data set is used to verify the proposed method. Experimental results show that the overall recognition rate of this method is 97.25%, which is 11.59%, 9.63% and 4.07% higher than the existing NN, SVM, and ICNN accuracy rates respectively.