A field scale case study of digital twin framework for precision tomato cultivation: Integrating XGBoost-based prediction with model predictive control

초록

This study presents a digital twin-based framework for precision tomato cultivation, integrating XGBoost-based growth prediction with Model Predictive Control (MPC). The framework comprises three core components: a Virtual Control Module (VCM) for real-time monitoring and equipment actuation; an Environment Twin (ET) modeled on a physics-based simulator and validated with sensor data; and an XGBoost-based growth prediction engine. The VCM, developed using a commercial game engine, ensures seamless synchronization between physical and virtual entities. The ET accurately replicates greenhouse dynamics with errors well within standard thresholds. Correlation analysis identified key environmental drivers, which informed the growth model's configuration. Following a comparative benchmark against statistical and deep learning models, XGBoost was selected for its superior accuracy and robustness. The field-scale evaluation demonstrated that the MPC-driven approach significantly improved microclimate stability, reducing temperature and humidity deviations by 70-80 % compared to conventional VPD-based heuristic control. These improvements led to a 35.1 % increase in total yield, while premium-grade production more than doubled. Analysis of actual operation logs further indicates the enhanced resource and energy efficiency, with an inferred estimated return on investment (ROI) of approximately 2.5 years.

키워드