Smart and Sustainable Agriculture: Offline Reinforcement Learning for Intelligent Handover in Hybrid VLC/RF Agricultural IoT Networks

Smart agriculture faces critical challenges in establishing reliable, energy-efficient communication networks for real-time crop monitoring and precision resource management. Traditional Radio Frequency (RF) wireless sensor networks in agricultural environments suffer from electromagnetic interference, limited spectrum availability, and potential negative impacts on crop growth. This paper presents a novel hybrid Visible Light Communication (VLC) and RF heterogeneous network architecture specifically designed for smart agricultural Internet of Things (IoT) applications, integrated with offline reinforcement learning for intelligent handover management. Leveraging the dual-purpose functionality of LED grow lights for both plant cultivation and data transmission, our approach creates an RF-minimal environment that promotes sustainable precision farming. We implement and compare four offline RL algorithms—tabular Q-Learning, Deep Q-Network (DQN), Conservative Q-Learning (CQL), and Implicit Q-Learning (IQL)—using a realistic dataset of 1010 handover scenarios from agricultural IoT deployments. Experimental results demonstrate that tabular Q-Learning achieves 90.40% handover success rate, while IQL attains 88.91% among deep learning methods. The hybrid VLC/RF system reduces RF radiation exposure while maintaining Quality of Service (QoS) requirements for real-time soil moisture monitoring, climate control, and automated irrigation systems. Our findings contribute to sustainable precision agriculture by enabling energy-efficient, crop-safe wireless communication infrastructure that optimizes resource utilization and enhances agricultural productivity through intelligent network management.