Reinforced-LSTM Trajectory Prediction-driven Dynamic Service Migration: A Case Study
Official URL
Publisher DOI
Abstract
Mobility prediction is an essential enabler to provide intelligent network systems and services in the upcoming B5G/6G era. Artificial Intelligence (AI) models such as Long Short Term Memory (LSTM) offer great performance at predicting users’ locations. However, model training can be time-consuming, which brings obstacles to practical applications. In this article, we present a mobility predictor based on Long Short Term Memory (LSTM), which is a variant of Recurrent Neural Networks (RNN) to reduce the network traffic for the sake of service migration improvement and handover (HO) optimization. To speed up the model convergence rate, we employ a Reinforcement Learning (RL) technique to automate the selection procedure of the best neural network architecture. To further accelerate the RL environmental search procedure, we transfer the architecture knowledge learned from a teacher LSTM to a student LSTM via a Transfer Learning (TL) framework. We propose a HO algorithm and a service migration algorithm based on the proposed LSTM predictor. We deploy the AI models on a mobile edge computing architecture using a real-world dataset collected from Paris, and evaluation results prove the efficiency of the predictor, and its impacts on improving ping-pong handover, and the service migration performance.
Date Issued
2022-04-25
Publication Type
Article
Subjects
Service migration
•
handover optimization
•
trajectory prediction
•
recurrent neural network
•
reinforcement learning
•
transfer learning
Language(s)
en
Author(s)
Braud, Arnaud | |
Radier, Benoit | |
Tamagnan, Philippe |
Additional Credits
Journal
Transactions on Network Science and Engineering
Publisher
IEEE
Access(Rights)
open.access