Maintaining adequate root-zone temperature in solar greenhouses during extreme cold is crucial for crop production. This study investigated the optimization of an auxiliary biomass heating system in a solar greenhouse. The heating performance was evaluated using an integrated methodology that combined orthogonal experimental design, Computational Fluid Dynamics (CFD) simulation, and Machine Learning (ML) surrogate modeling. First, a reliable CFD model, validated against experimental data (Index of Agreement, IA = 0.954), was used to generate high-fidelity temperature field data for nine layout schemes. Parameter sensitivity analysis revealed that the burning cave Diameter is the dominant factor (R = 6.01), followed by burial Depth (R = 2.00), with inter-pool Spacing having the least impact (R = 0.89). Subsequently, six ML algorithms were compared for use as a predictive surrogate model, with Lasso Regression demonstrating superior performance (R2 = 0.934). Comprehensive optimization focused on maximizing the Suitable Area Ratio (Rs) in the critical 0.2 m depth root zone. The analysis conclusively identified the 2.5 m diameter group as optimal, achieving a maximum Rs of 90% and the lowest temperature standard deviation. The final recommended optimal design (2.5 m diameter, 0.7 m depth, 10 m spacing) significantly improves heating uniformity and efficiency. This integrated CFD-ML approach provides a scientific basis and a rapid assessment tool for the design and structural optimization of similar underground thermal systems in cold-climate agriculture.
Xu et al. (Mon,) studied this question.