Conversational LLM Framework for Secure and Energy-Efficient Wireless Sensor Networks

A conversational orchestration framework named A-LLM is presented in which large language models (LLMs) are employed to jointly optimize energy efficiency and security in wireless sensor networks (WSNs). Sink–sensor interactions are modeled as structured dialogues, through which compressed network state summaries are translated into actionable policies for cluster-head selection, duty-cycle scheduling, and anomaly mitigation. A risk-aware energy optimization objective that incorporates the cost of LLM queries is formalized, and a secure conversational reconfiguration protocol (SCR-LLM) is proposed to isolate compromised nodes and refresh group keys under adversarial conditions. To improve practicality, a prompt-distillation approach is introduced so that LLM behavior can be transferred to a lightweight edge policy, thereby reducing both latency and query overhead. The framework has been designed for evaluation through extensive simulations and an emulated testbed under heterogeneous traffic patterns and attack scenarios. Improvements in network lifetime, energy consumption, and detection accuracy have been observed when compared to classical baselines. All datasets used were either publicly available or derived from public sources, ensuring reproducibility and methodological transparency.