Network Security Based on GCN and Multi-Layer Perception

Abstract

With the continuous progress of network technology, network security has become a critical issue at present. There are already many network security intrusion detection models, but these detection models still have problems such as low detection accuracy and long interception time of intrusion information. To address these drawbacks, this study utilizes graph convolutional network to optimize multi-layer perceptron. An optimization algorithm based on multi-layer perceptron is innovatively proposed to construct an intrusion detection model. Comparative experiments are conducted on the improved algorithm. The accuracy of the algorithm was 0.98, the F1 value was 0.97, and the detection time was 1.1s. The overall performance was much better than comparison algorithms. Subsequently, the intrusion detection model was applied to network security detection. The detection time was 0.1s, the accuracy was 0.98, and the overall performance outperformed other comparison algorithms. The results demonstrate that the intrusion detection method on the basis of optimized multi-layer perceptron can enhance the detection ability of illegal intrusion information. This study optimizes the performance of detecting illegal network intrusion information, providing a theoretical basis for further development of network security. However, the types of intrusion information in this study are limited and there is still uncertainty. In the future, data augmentation techniques can be used to oversample minority class samples, synthesize new minority class samples, expand sample size, increase detection information, and improve the overall detection performance of the model.