A REINFORCEMENT LEARNING-ENHANCED QBO-ATBNN MODEL FOR DYNAMIC VNF MANAGEMENT IN MULTI-ACCESS EDGE COMPUTING

Abstract

The management of Virtual Network Functions (VNFs) in Multi-Access Edge Computing (MEC) is critical for meeting the stringent latency and resource demands of modern applications. While our previous work introduced a hybrid Quantum Butterfly Optimization and Attention-Based Temporal Bayesian Neural Network (QBO-ATBNN) model to address this, its reactive-predictive loop lacked real-time decision-making agility. This paper proposes a novel enhancement: the integration of a Deep Reinforcement Learning (DRL) agent into the QBO-ATBNN framework. The resulting QBO-ATBNN-RL model leverages the proactive forecasting of ATBNN and the global optimization of QBO, while using DRL for instantaneous, fine-grained resource allocation in response to dynamic network states. We evaluate our model on a simulated MEC environment with real-world workload traces. Results demonstrate that the QBO-ATBNN-RL model achieves a further 15% reduction in latency and a 12% improvement in energy efficiency compared to the baseline QBO-ATBNN, while maintaining high scalability and robustness under rapidly fluctuating conditions.