Dynamic Light Field Reconstruction and Digital Twin Surrogate Modeling for Multi-Objective Light Environment Optimization of a Vertical Circulating Rice Seedling Rack

Vertical circulating rice seedling racks in cold regions of northeast China face limited natural radiation, uneven light distribution, and high energy demand. This study develops a digital twin optimization framework integrating dynamic light field reconstruction, crop physiological response, surrogate modeling, and multi-objective analysis. A solar position algorithm-based simulator reconstructs the spatiotemporal PPFD distribution along the circulating trajectory and couples it with stage-specific rice seedling light response curves. An XGBoost surrogate model is then used to accelerate parameter evaluation for stage-wise optimization. Preliminary measurements using a plant light analyzer showed limited PPFD deviations from simulation, with a mean deviation of −3.4% and a maximum deviation of +11.4%. The surrogate model achieved a test R2 of 0.9955 and a five-fold cross-validation R2 of 0.9933 ± 0.0024. Under the engineering constraint of CV ≤ 0.10, stage-wise optimization identified max yield, best efficiency, and balanced operating strategies. The balanced strategy predicted a 7.7% increase in YI, a 6.0% reduction in energy use, and a decrease in CV from 0.0353 to 0.0058 relative to the fixed-parameter baseline. SHAP analysis further identified growth stage as the dominant structural factor, while lighting duration and rack speed mainly regulated within-stage performance. This framework provides a quantitative basis for stage-aware light environment optimization and subsequent experimental validation.