Proximal Policy Optimization for Adaptive Cluster Head Selection in Wireless Sensor Networks

Abstract

The energy-efficient operations of WSNs have become important due to the limited battery life of the sensor nodes. WSNs are increasingly used increasingly for mission-critical applications like environmental monitoring, precision agriculture, and industrial automation. Choosing the correct cluster head is one of the core deficiencies in the implantation of any clustering algorithm which greatly affect network longevity and energy consumption. Protocols designed to achieve LEACH, iLEACH, SEP, TEEN, and Sim-TEEN use static heuristics or probability models. They do not adapt to dynamic variations of remaining energy and changing topology. These lead to early energy depletion of nodes. They apply unequal utilization of nodes. To address these limitations, a CH selection method based on Proximal Policy Optimization (PPO) using reinforcement learning is proposed in this paper. A WSN (wireless sensor network) environment, compatible with OpenAI Gym, is created to simulate a field deployment with 200 sensor nodes in a 300×300 m² area. The PPO agent's policy runs for 100,000 timesteps for a single episode of 1000 rounds which helps the policy learn about temporal patterns and long-term energy. The method adaptively chooses CHs based on energy thresholds, node distribution and communication range constraints while minimizing energy loss and ensuring fairness. PPO significantly outperforms conventional algorithms in terms of savings of residual energy as well as node survival, as shown by the simulation results. The article compares the performance of a novel protocol for wireless sensor networks with previous methods. The results indicate the robustness, flexibility and applicability of PPO in energy efficient WSN protocol design with a scalable and intelligent solution to the traditional CH selection scheme.