Neural Computing and Applications Software
SAC-AC: a soft actor-critic based framework for adaptive clustering in wireless sensor networks
Abstract
WSNs (Wireless Sensor Networks) continue to face challenges related to scalability, efficiency, and adaptability in dynamic environments. Traditional clustering algorithms often rely on fixed heuristics, which frequently result in uneven energy consumption and a reduced network lifetime. The current paper presents a new adaptive clustering framework, which is called SAC-AC, that uses the Soft Actor-Critic (SAC) algorithm, which aims at maximizing cluster formation and scheduling communication in WSNs. SAC-AC makes use of a continuous action space to enable precise selection of cluster-heads, considering residual energy, distance to the base station, and queue backlog. The framework incorporates traffic-aware TDMA scheduling and dynamic power control, which enable real-time adaptation to varying network loads. A multi-objective rewarding mechanism trades off throughput, reliability, and load distribution to enhance the overall network performance. Various simulations with traffic loads of 100 to 1000 packets per round show that the SAC-AC protocol consistently outperforms the established protocols under the compared scenarios, such as LEACH, HEED, RL-Static, and LEACH-RLC. SAC-AC can reach up to 62% higher throughput under high-load conditions; it reduces node failures by up to 90% and has better energy efficiency. These improvements are statistically validated using non-parametric tests (Kruskal–Wallis H test, $$p 2.6$$ ).