Intelligent Resource Allocation via Hybrid Reinforcement Learning in 5G Network Slicing

摘要

Manufacturers are focusing on reconfigurable, resilient environments for Industry 5.0 paradigms. Applications like digital twins and mobile robots require communication networks to meet latency, bandwidth, and reliability requirements. Beyond 5G (B5G) networks provide unprecedented communications performance and flexibility through virtualization and network slicing, which generates various logical partitions for particular applications with specific requirements. RAN slicing is an essential section of 5G network slicing due to its vulnerability to errors, affecting its ability to meet stringent reliability requirements. This paper presents a novel framework for optimizing resource allocation in 5G network slicing by integrating Double Deep Q-Network with Prioritized Experience Replay (DDQN-PER) and Pointer Network-based Long Short-Term Memory (PtrNet-LSTM). The proposed framework dynamically adjusts the attention coefficient, balancing Service Satisfaction Level (SSL) and Quality of Experience (QoE), improving system efficiency, spectrum efficiency, and user connectivity across diverse user scenarios. The experiment illustrates that the combined PtrNet-LSTM framework within DDQN-PER outperforms the baseline methods in terms of spectrum efficiency and user connectivity, demonstrating scalability and the potential to address challenges in dynamic wireless networks.