Q-learning-based medium congestion control in WSN for smart farming applications

Abstract

This thesis proposes a novel Q-learning-based (MAC) protocol to address the challenge of medium congestion control in Wireless Sensor Networks (WSNs) for smart farming applications, where traditional static or centralized MAC protocols often fail to adapt to scalability and resource-constrained agricultural environments. By integrating reinforcement learning into a (TDMA) framework, the protocol enables sensor nodes to autonomously learn optimal transmission strategies, dynamically selecting time slots based on environmental feedback to minimize collisions and improve efficiency. The decentralized learning mechanism allows each node to maintain a Q-table, iteratively refining slot selection to enhance throughput and reduce energy consumption. Simulation results demonstrate significant improvements in collision probability, throughput, and fairness compared to conventional (TDMA) and contention-based protocols, making the approach particularly effective in dense deployments. The protocol’s adaptabil- ity and scalability highlight its suitability for rural agricultural settings, where reliability and energy efficiency are critical.