Construction and Analysis of Cross-layer Aggregation Neural Network for AMI Intrusion Detection

Abstract

Due to two-way communication with the internet, advanced measurement infrastructure (AMI) is vulnerable to network attacks. In order to improve the reliability of AMI intrusion detection, this paper proposes the cross-layer aggregation neural network. Firstly, the random forest algorithm is used to reduce the interference of redundant features. Then, the cross-layer model based on convolutional neural network and gated recurrent unit is used to extract local and temporal features, and concat() and multi-layer perceptron network are used for features aggregation to obtain comprehensive features with multi-domain features. According to the comprehensive features and softmax classifier, the communication state of AMI is detected. Finally, the four types of intrusion detection experiments are carried out on KDDCup 99 and NSL-KDD data sets. Experimental results show that the cross-layer model has better detection performance than multi layer perceptron, convolutional neural network, gated recurrent unit, serial neural network and others.