Multi-Agent Deep Reinforcement Learning for Handover Management in Massive Industrial Internet of Things Networks

Abstract

The industrial internet of things (IioT) is considered one of theIapplications in the fifth generation (5G) networks. In this application, users’ high mobility in a typical industrial scenario needs high reliability. The high mobility creates frequent handover, creating extra control signalling to a new base station (BS). The users’ congestion to the new BSs can lead to an outage. In this paper, we investigate how to manage the handover of users to improve reliability in a high-mobility scenario using deep learning. We first use an offline centralized algorithm to create labels for user association to a BS that is done without adding handover coefficient. Then, we train the neural network and use the trained parameter to make the multi-agent deep reinforcement learning (RL) learns better. This is done to avoid long iterative methods in reinforcement learning. The results show that our method can outperform the offline centralized algorithm by 40% when the handover coefficient increases.