PER-DDQN-based Load Balancing Handover Strategy for Low Earth Orbit Satellite

In Low Earth Orbit (LEO) satellite communication systems, the dynamic variations in user demand and satellite resources make load balancing handover strategies a critical technology for improving network performance and service quality. This paper proposes a load balancing handover strategy based on Prioritized Experience Replay (PER) and Deep Reinforcement Learning (DRL), aiming to optimize the connection of User Equipment (UE) among multiple LEO satellites, ensuring efficient resource utilization and enhanced user experience. The study first establishes a system model that comprehensively considers factors such as Carrier-to-Noise Ratio (CNR), service time, and satellite load. Subsequently, the Double Deep Q-Network (DDQN) algorithm is employed, trained with a prioritized experience replay mechanism to enhance decision-making efficiency and accuracy. By dynamically evaluating satellite signal quality and load conditions, the proposed strategy adjusts UE connection choices in real-time to achieve balanced load distribution. Experimental results demonstrate that, compared to traditional load balancing methods, the proposed strategy significantly improves Quality of Service (QoS), and enhances load balancing. When the total number of users is 250, the QoS increases to 0.624. Notably, under high-load network conditions, the strategy effectively prevents single satellite overload, ensuring system stability and service continuity.