Adaptive Slicing Allocation for Satellite-terrestrial Edge Computing Based on Traffic Prediction

To meet the diverse requirements of the multiaccess edge computing (MEC) systems’ users, this paper proposes an adaptive slicing resource allocation (ASRAO) for satelliteterrestrial edge computing networks based on traffic prediction. The ASRAO method aims to deliver the MEC services with low latency as well as low power consumption by creating two end-to-end isolated radio access network (RAN) slices. First, a space-time traffic prediction (STTP) is developed by integrating an improved convolutional neural network (ICNN) and a long short-term memory (LSTM) model to conduct interval predictions of network traffic for different service types. Then, a two-layer traffic adaptive multi-agent reinforcement learningbased (TAMARL) model is introduced to solve the optimization problem by adaptively adjusting resource allocation for each slice based on the traffic volume. The TAMARL model performs an alternating iterative optimization of the terrestrial layer (TL) and air-space layer (ASL) subproblems. The experimental results demonstrate that the proposed ASRAO outperform the existing algorithms.