Energy-efficient clustering and routing protocol for wireless sensor networks based on improved HOA and attention-guided D3QN

Abstract Clustering and routing protocols are widely regarded as effective methods for improving energy efficiency and extending network lifetime in wireless sensor networks. However, existing protocols often insufficiently consider dynamic energy distribution and topology characteristics, which may lead to imbalanced cluster structures, uneven energy consumption, and additional communication overhead. To address these issues, this paper proposes HOADRL, an energy-efficient clustering and routing protocol based on the hiking optimization algorithm (HOA) and deep reinforcement learning. First, an improved HOA with an integrated encoding strategy is developed to optimize cluster formation by jointly determining cluster head (CH) selection and cluster membership, where a multi-objective fitness function considers residual energy, intra-cluster distance, load balancing, and distance to the base station. Second, an attention-guided dueling double deep Q-network (D3QN) is introduced to determine adaptive routing paths under dynamic network conditions, where each CH aggregates neighboring information through attention weights to improve next-hop selection. Finally, a lightweight feedback mechanism is designed to adaptively adjust HOA fitness coefficients according to network states. Extensive simulation results demonstrate that HOADRL achieves superior performance compared with representative clustering and routing protocols in terms of network lifetime, average energy consumption, end-to-end delay, and network throughput.