AI-driven IoT routing: A hybrid deep reinforcement learning and shrike optimization framework for energy-efficient communication

Abstract

The expansion of Internet of Things (IoT) networks has intensified the need for intelligent and adaptive routing strategies capable of handling frequent topological changes, energy limitations, and application-specific performance requirements. Existing routing protocols often struggle to simultaneously achieve scalability, energy conservation, and reliability. To address these challenges, this paper introduces a novel hybrid routing framework, DRL-SOA, which fuses Deep Reinforcement Learning (DRL) with the Shrike Optimization Algorithm (SOA) to enable real-time, congestion-aware, and energy-efficient routing in IoT environments. The DRL component incrementally learns optimal routing paths by interacting with dynamic network conditions, while SOA enhances the convergence of Q-learning by identifying the most promising action sequences using a nature-inspired hunting mechanism. The proposed method employs a multi-parameter fitness function that considers link stability, link duration, remaining energy, bandwidth availability, and node connectivity to determine optimal routing paths. Extensive simulations using NS-3 demonstrate that DRL-SOA significantly outperforms existing approaches, including RIATA, DRL-IRS, and DOACAR. Notably, the proposed approach achieves up to a 25% increase in network lifespan, reduces routing overhead by 22%, and enhances packet delivery and energy efficiency across different node densities and mobility rates. These results establish DRL-SOA as a scalable and robust routing protocol for next-generation IoT systems.