The power communication network has a complex structure, numerous equipment, and complex technology, with typical nonlinear random charac­teristics and multi-scale dynamic characteristics. In the field of communication network traffic prediction, the main focus is on feature extraction and model optimization of the network traffic time series itself to improve prediction accuracy. However, in the process of modeling, there is little analysis and research on net­work traffic prediction based on the spatial characteristics of learning traffic. This article proposes a twin prediction method for power communication network traf­fic based on GraphLSTM. This article mainly extracts information from specific network information such as predefined network topology and network traffic data in the power communication network to construct a digital twin network model. At the twin layer, the GraphLSTM neural network model is used to pre­dict network traffic, in order to simultaneously model the temporal and spatial dependencies in the input data. Then, it is compared with the actual flow data of the power communication network, with the goal of minimizing the difference. In the simulation section, this article tested the proposed network traffic predic­tion model using the mainstream GEANT public dataset in the field of network modeling, and compared its performance with traditional LSTM models. The ex­perimental results show that the network traffic prediction model proposed in this article has good prediction performance on power communication network traffic data.