Prediction Model of Bursty Network Traffic for Cloud Data Center Based on GAN-TrellisNet

Aiming at the nonlinearity and burstiness of network traffic in large cloud data centers, this paper proposes a GAN-TrellisNet model to improve the accuracy of traffic prediction. Firstly, the model uses a multi-layer TrellisNet construction generator to obtain the long-term and local features of traffic data and generate traffic prediction values. Then, the discriminator selects a convolutional neural network (CNN) with strong feature extraction ability to distinguish the difference between the predicted value and the real value. The experiment shows that compared with the XGBoost, LSTM and TCN models, the proposed model is optimal in RMSE, MAE and R
²
performance indicators, and has better prediction effects.