Descriptive statistical analysis.

Stable nanofluid dispersion with SiO2 particles of 15, 50, and 100 nm is generated in a base liquid composed of water and glycerol in a 7:3 ratio and tested for physical characteristics in the temperature range of 20-100oC. The nanofluid showed excellent stability for over a month. Experiments are undertaken for the flow of nanofluid in a copper pipe and measured for their heat transfer coefficient and flow behavior. The convection heat transfer coefficient increases with the flow Reynolds number in the transition-turbulent flow regime. The experimental results further reveal that the friction factor enhancement with 0.5% concentration has increased by 6% as compared to the base liquid. It was employed for prognostic model development using XGBoost and multi-gene genetic programming (MGGP) to model and predict the complex and nonlinear data acquired during experiments. Both techniques provided robust predictions, as witnessed by the statistical evaluation. The R2 statistics of the XGBoost-based model was 0.9899 throughout the model test, while it was lowered to 0.9455 for the MGGP-based model. However, the change was insignificant. The mean squared value was 8.37 for XGBoost, while it increased in the MGGP model to 45.12. Similarly, the mean absolute error (MAE) value was higher (6.623) in the case of MGGP than in XGBoost at 2.733. The statistical evaluation, Taylor diagrams, and violin plots helped determine that XGBoost was superior to MGGP in the present work.

本研究以体积比7:3的水与甘油混合作为基础液，制备得到分散稳定性优异的SiO₂纳米颗粒（粒径分别为15、50、100 nm）纳米流体，并在20~100℃的温度区间内对其物理特性开展测试。该纳米流体可保持长达一个月以上的稳定分散状态。针对纳米流体在铜管内的流动特性展开实验，测定其对流传热系数与流动行为。实验结果表明，在过渡-湍流流动区域内，对流传热系数随流动雷诺数的升高而增大。进一步的实验结果显示，当纳米流体浓度为0.5%时，其摩擦阻力系数相较于基础液提升了6%。本研究采用极限梯度提升树（XGBoost）与多基因遗传规划（MGGP）构建预测模型，对实验采集的复杂非线性数据进行建模与预测。经统计评估验证，两种方法均能输出鲁棒性良好的预测结果。在全部模型测试中，基于XGBoost的预测模型的决定系数（R²）为0.9899，而基于MGGP的模型决定系数降至0.9455，但该差异并不显著。XGBoost模型的均方误差为8.37，而MGGP模型的均方误差升至45.12。类似地，MGGP模型的平均绝对误差（MAE）为6.623，高于XGBoost模型的2.733。通过统计评估、泰勒图与小提琴图的综合分析，本研究证实XGBoost模型的整体表现优于MGGP模型。