Efficient Handover Management and Resource Allocation in Dual-Connectivity Optical Wireless/RF Networks Using Multi-Criteria Deep Reinforcement Learning

Dual Connectivity (DC)-enabled Optical Wireless Communication and Radio Frequency Heterogeneous Networks (OWC/RF HetNets) offer a promising solution for achieving high data throughput, low latency, and energy-efficient wireless communication. However, the integration of fundamentally distinct technologies introduces significant challenges in coordinating handover management and resource allocation, especially under dynamic network conditions and user mobility. This paper formulates the joint handover and resource allocation problem in DC-enabled OWC/RF HetNets as a multi-objective optimization framework, explicitly modeling the trade-offs between transmission power, handover cost, and throughput across both technologies. We employ goal programming to unify the objectives, enabling balanced trade-offs among objectives. To address the limitations of conventional Reinforcement Learning (RL) methods, which rely on weighted summation to integrate multiple objectives into a single reward function, we propose a novel framework, termed Multi-Criteria Double Deep Q-Learning (MC-DDQL). The framework employs parallel deep Q-networks to learn policies for each reward function independently. It integrates the Technique for Order Preference by Similarity to Ideal Solution (TOPSIS) together with Boltzmann annealing to achieve effective multi-criteria action selection. Simulation results demonstrate that the proposed MC-DDQL framework addresses the challenge of objective weight selection, particularly in dynamic environments. The framework achieves accelerated convergence and superior performance relative to state-of-the-art benchmarks, while maintaining a performance profile comparable to the optimal solution. © 2002 IEEE.