Adaptive Load Balancing In Sdns Using A Fuzzy-Driven Twin Delayed Deep Deterministic Policy Gradient Algorithm

Abstract

Efficient load balancing remains a critical challenge in Software-Defined Networks (SDNs) due to dynamic traffic patterns, unpredictable flow demands, and the risk of congestion at highly utilized links. Traditional load balancing methods and single-agent reinforcement learning approaches often fail to address the inherent uncertainty and instability in large-scale SDN environments. To overcome these limitations, this paper proposes a novel Fuzzy-Driven Twin Delayed Deep Deterministic Policy Gradient (Fuzzy-TD3) framework for adaptive load balancing. The proposed model integrates a fuzzy inference system with TD3 to enhance decision stability and robustness. Specifically, fuzzy logic is employed to preprocess network states such as link utilization, queue occupancy, and flow distribution, thereby reducing ambiguity in state representation. Furthermore, a fuzzy-based dynamic reward shaping mechanism is introduced to balance multiple objectives, including fairness, throughput, and end-to-end delay. By combining uncertainty handling with adaptive reinforcement learning, the hybrid framework achieves faster convergence and improved policy stability compared to conventional TD3. Experimental evaluation on Mininet-based SDN topologies demonstrates that Fuzzy-TD3 outperforms benchmark techniques, achieving lower flow delay, higher load fairness, and better link utilization under diverse traffic scenarios. The results confirm that the proposed hybrid model provides a scalable and adaptive solution for real-world SDN load balancing.