DT-GCN Based Digital Twin for 6G Self-Intelligent Network

With the rapid development of Sixth Generation (6G) self-intelligent network, higher requirements are put forward for intelligent decision, high-precision configuration, self-learning and self-optimization of the network. Digital twin, as digital representation of physical network, can enhance the capabilities of physical network in terms of data perception, intelligent decision-making, and pre-validation to enable the realization of closed-loop autonomy in 6G network. However, the precise and efficient construction of digital twin network (DTN) faces challenges due to the complexity of network, which in turn can affect the accuracy of artificial intelligence (AI) based intelligent policies pre-validation. Therefore, a data-driven Digital Twin architecture for 6G self-intelligent network is proposed in this paper. Based on this architecture, the method of digital representation based on Graph Sample and Aggregation (GraphSAGE) towards global network is innovatively proposed, aiming at mining and extracting the deep correlation of network nodes in global topology information. Moreover, a digital twin graph convolutional network (DT-GCN) which taking representation of digital twin as prior knowledge, is constructed for network decision-making pre-validation, and this provides more precision for intelligent policy. Finally, the DT-GCN is introduced to realize the application of network fault pre-validation with greater accuracy and speed. The results show that DT -GCN improves policies accuracy by 4.98% and convergence speed by 7.14% compared to traditional GCN method.