Abstract
Intrusion detection in Software Defined Networks (SDNs) faces critical challenges due to evolving attack surfaces and increasing traffic complexity. This paper proposes a novel Adaptive Transformer-based Quantum Intrusion Detection System (ATQ-IDS), integrating four core components: Quantum-Inspired Evolutionary Selection (QIES) for optimal feature reduction, a Transformer-Spatial Temporal Network (TSTN) for deep traffic context modeling, Hierarchical Reinforcement Learning-based IDS (HRL-IDS) for adaptive policy control, and a Federated Learning-enabled IDS (FL-IDS) for decentralized, privacy-aware deployment. The QIES component minimizes model overhead by selecting a reduced, high-utility feature set, while the TSTN captures intricate spatial and temporal patterns using attention mechanisms. HRL-IDS ensures decision adaptability in dynamic traffic environments, and FL-IDS supports real-time distributed detection with minimal communication cost. Experimental evaluations on benchmark SDN datasets demonstrate that ATQ-IDS achieves state-of-the-art accuracy of 99.84%, with a false negative rate of 0.12% and inference time below 7 ms. Ablation studies and robustness analyses with confidence intervals confirm the contribution of each module and the model's consistency across runs. This architecture demonstrates high accuracy, adaptability, and real-time applicability, making it a robust solution for modern SDN-based security systems.