目的 基于数据挖掘技术,建立三七叶面积生长预测模型,对于三七整个生长期的精准管理与决策提供参考。方法 基于粒子群-随机森林算法,采用2018、2019年4~10月云南省红河自治州泸西县三七种植基地棚内气象因子数据以及三七叶面积生长数据作为训练集和测试集构建生长预测模型。结果 通过特征工程中皮尔森系数分析可知,三七叶生长与土壤温度、上方水蒸气压和下方水蒸气压等气象因子呈正相关,其中土壤温度正相关程度最大,其皮尔森相关系数在0.75~0.90;下方土壤热通量与三七叶生长呈负相关,其皮尔森相关系数为−0.4~−0.3;通过粒子群优化随机森林算法训练的生长预测模型,其均方根误差(root mean square error,RMSE)收敛时值为0.021 82,模型优化后的三七叶生长预测模型决定系数R²达到0.999 97。结论 通过多种算法对比实验结果表明,粒子群-随机森林算法构建的三七叶面积生长预测模型具有较高的预测精度。该方法为三七叶的生长预测提供了新的研究思路。

Objective Based on data mining technology, the growth prediction model of Sanqi (Panax notoginseng) leaf area was established to provide reference for accurate management and decision-making of P. notoginseng during the whole growth period. Methods Based on the particle swarm-random forest algorithm, the meteorological factor data in the shed of P. notoginseng planting, Luxi County, Honghe Autonomous Prefecture, Yunnan Province from April to October 2018 and 2019 and leaf area growth data of P. notoginseng were used as the training set and test set of machine learning methods to build a growth prediction model. Results After doing the Pearson coefficient analysis of the characteristic engineering, the simulation results showed that the leaf growth of P. notoginseng was positively correlated with meteorological factors such as soil temperature, upper water vapor pressure and lower water vapor pressure. The positive correlation degree of soil temperature was the largest one with 0.75-0.90 Pearson correlation coefficient. On the contrary, the soil heat flux below was negative correlated with the leaf growth of P. notoginseng, and the Pearson correlation coefficient was −0.4- −0.3. For the prediction model trained by the proposed particle swarm-random forest algorithm, the convergence value of the root mean square error (RMSE) was 0.021 82, and the coefficient of determination R² of P. notoginseng leaf growth prediction model reaches 0.999 97 after model optimization. Conclusion The comparative results among different algorithms showed that the prediction model of P. notoginseng leaf area growth constructed by particle swarm optimization random forest algorithm has high prediction accuracy. Meanwhile, the combined algorithm proposed in this paper provides a new idea for the growth prediction research of stems and leaves of P. notoginseng.

R282

国家自然科学基金项目（62063011，51979134，51779113）；云南省科技厅科技计划项目（202001AU070032）