Fixed-set Learning for Cluster-head Selection in Multi-hop Wireless Sensor Networks

Wireless Sensor Networks (WSNs) have remained an active research field in both military and civilian domains, driven by the expanding diversity of their applications. In recent years, there has been a progressive shift toward integrating Artificial Intelligence to address the persistent challenge of energy optimization in WSNs. We introduce a novel adaptation of a Fixed Set Search (FSS) mechanism to WSNs. FSS adds a learning phase to the well-known GRASP metaheuristic. FSS-WSN approach guides the Base Station (BS) in a centralized multi-hop environment to select the optimal cluster-heads, thereby maximizing the global utility of the network. We evaluated our approach under documented fairness conditions, against a wide range of established baselines including classical clustering protocols (LEACH, HEED, SEP), widely used swarm optimizers (PSO, GWO, ABC), and recent SO-GJO-family variants (SO, GJO, EMO–GJO, and ESO–GJO), as well as the recently proposed EEM-LEACH-ABC. The results demonstrate a statistically significant improvement (paired Wilcoxon test with Holm correction) over the best baseline regarding two key metrics–the number of delivered reports and the CPU time required for decision-making. These results suggest that our approach is a strong, practical option for many WSN use cases.