Spatial-Temporal Prediction Model for Optical Fronthaul Network Traffic based on ASTFN

This paper proposes a traffic prediction model for optical fronthaul network called attention-based spatial-temporal fusion neural network (ASTFN), combining convolutional neural network (CNN) capturing the spatial dependencies between fronthaul traffic, and long short-term memory (LSTM) neural network obtaining temporal feature. Fusion strategy is adopted to reduce the phenomenon of information loss due to preference expression in CNN by concatenate the original traffic matrix and spatial feature extraction result. Self-attention mechanism in LSTM captures the noteworthy information in sliding window, improving the expressing capability of temporal dependency. The experimental results show that the model proposed in this paper has better prediction results and higher prediction accuracy.