Hybrid GA-DQN for opportunistic cluster head selection and routing in resource constraint Internet of Things networks
TL;DR
Results confirm that GA-based cluster optimisation combined with DQN-based adaptive routing balances energy expenditure and sustains data delivery in the IoT network.
Abstract
Energy efficiency and reliable data transmission are two important challenges in Internet of Things (IoT) networks because of the limited energy resources of sensor nodes and their dynamic nature. Accordingly, this research develops a Genetic Algorithm Deep Q-Network (GA-DQN) based hybrid framework for intelligent cluster head selection and adaptive IoT Network routing to counter these threats. A Genetic Algorithm (GA) selects the best cluster heads based on residual energy, coverage ratio, inter-cluster distance, and node density. In turn, this ensures balanced cluster formation and efficient energy utilisation. Next, the Deep Q-Network (DQN) learns the forwarding decision from a network state, including the node’s energy level, node location, information of a neighbour node, link quality, and distance to the sink. The reward mechanism has been formulated to reduce energy consumption, enhance communication reliability, and optimise routing efficiency. The results show over a 100 m × 100 m field containing 100 nodes shows that the proposed framework postpones the first node death to approximately 700 rounds and the half-node death to approximately 4200 rounds, against 550–600 and 2200–3000 rounds respectively for EERRL, ICR-HHODRL and CGA-GWO, so that the overall network lifetime increases by 7.69%, 2.94% and 1.45% with respect to the same baselines. Average energy consumption per node falls to 0.18 J from 0.28 J, 0.30 J and 0.35 J, the average packet delay falls to 3 × 10⁻³ s, a reduction of 25.0%, 33.3% and 40.0%, and the end-to-end latency falls to 0.24 s, a reduction of 29.4%, 36.8% and 42.9%. Notwithstanding a lower peak per-round throughput of 1.7 × 10⁵ bits against 2.6 × 10⁵ to 3.5 × 10⁵ bits for the baselines, delivery is sustained beyond 6000 rounds, whereas the throughput of the baseline’s decays to a residual level within approximately 2500 rounds. A scalability campaign spanning 50 to 200 nodes in fields of 50 m × 50 m to 200 m × 200 m extends the last node death from 5385 to 15,763 rounds without any modification of the learning architecture. These results confirm that GA-based cluster optimisation combined with DQN-based adaptive routing balances energy expenditure and sustains data delivery in the IoT network.