Optimization of the Decision-making Process of Digital Twins in Network Security Based on Graph Neural Networks

Abstract

The network security model based on static rules lacks flexibility and adaptability, and it is difficult to adapt to and respond to dynamically changing network threats in real-time. Therefore, this paper proposes a method for optimizing the decision-making process of digital twins (DTs) in network security based on graph neural networks (GNNs). First, this paper applies the Adam optimizer to adjust the learning rate by optimizing the GNN structure, combines the cross entropy loss function to improve the attack recognition ability, and uses L2 regularization and Dropout to prevent overfitting to enhance the model’s performance in complex network data. Then, real-time network threat detection and attack path prediction are performed based on the optimized GNN model. To further improve the intelligent level of network security protection, this paper applies the online learning (OL) algorithm to continuously update the model to adapt to changes in the network environment and threat patterns. At the same time, combined with the policy gradient (PG) method, an intelligent decision-making module is designed to automatically adjust the defense strategy to achieve dynamic protection against changing network threats. Experimental results show that the optimized GNN model’s accuracy in network threat detection reaches 93.3%, which is 9.7% higher than that of the non-DT model, and the malware’s precision is increased by 6.9%. The system’s response time is reduced to 50ms, which significantly improves the real-time performance and decision accuracy of network security protection and demonstrates the excellent performance and broad application prospects of this method in dynamic network environments.