Integrated satellite-terrestrial edge computing networks (ISTECNs) have been developed to advance the existing wireless communication systems by combining multi-access edge computing (MEC) with radio access network (RAN) slicing. This paper proposes an adaptive slicing resource optimization (ASRAO) method for satellite-terrestrial edge computing networks

aims to deliver the MEC services with high bandwidth

low latency

broad coverage

and low power consumption by creating four end-to-end isolated RAN slices. First

the resource optimization problem for the four RAN slices into two subproblems

the terrestrial layer (TL) and air-space layer (ASL) optimization subproblems

is designed. Then

a parallel traffic prediction model with feature fusion (PTPFF) is developed by integrating an improved convolutional neural network (ICNN) and a long short-term memory (LSTM) model to conduct predictions of network traffic for different service types. Finally

a two-layer traffic adaptive multi-agent reinforcement learning-based (TAMARL) model is introduced. It adaptively adjusts resource allocation for each slice based on the traffic volume and performs an alternating iterative optimization of the TL and ASL subproblems. The experimental results demonstrate that the proposed ASRAO can enhance the throughput by 38%

reduce the average latency by 33%

improve coverage by 20%

and decrease the average energy consumption by 21.5%.