Agentic Centralized Clustering for Energy-Efficient Wireless Sensor Networks

Although several algorithms have been proposed for clustering in wireless sensor networks, the energy constraint of nodes remains a significant issue affecting network performance. Algorithms such as LEACH and its centralized versions have provided effective strategies to reduce energy consumption. However, due to static decisions or restrictive criteria in the clustering process, these methods are unable to adapt to changing network conditions. Recent advances in large language models and agent-based architectures have enabled the use of an intelligent management layer that can adjust high-level decisions according to the current network state. In this paper, we present a modular and upgradable architecture for centralized clustering management in WSN. In this architecture, an LLM-based agent plays the role of the control and decision-making layer, while the main clustering operation is performed by the k- means algorithm. The agent dynamically optimizes clustering parameters including the number of clusters, re-clustering intervals, and feature weights based on real-time network performance metrics. Experimental results show that the proposed architecture, using an agent based on large language models, can adaptively manage the clustering process and increase the network lifetime by an average of more than ۱۷% compared to reference methods. In all tested scenarios, the network under the control of the agent reached energy depletion later than traditional methods, indicating higher efficiency and greater flexibility of the proposed architecture in dynamic conditions.

کلیدواژه ها:

Clustering ، Energy Efficiency ، Agentic Architecture ، Wireless Sensor Networks ، Large Language Models