清香型优质烟叶物理特性指标预测分析

探究清香型优质烟叶物理特性指标与生态因子间的定量关系，构建物理特性指标预测模型。选择云南省玉溪市2019-2020年优质烟叶的物理特性指标数据与生态数据（气象、土壤和海拔），建立多元线性统计预测模型与BP神经网络预测模型，并分析各生态因子对烟叶物理特性指标的相对贡献率；利用均方根误差（Root Mean Square Error，RMSE）与归一化均方根误差（Normalized Root Mean Square Error，NRMSE）对两种预测模型模拟效果进行检验分析。结果显示，气象因子平均相对贡献率明显高于土壤、海拔的相对贡献率，气象因子对清香型优质烟叶物理特性指标尤为重要；统计预测模型的RMSE、NRMSE值均高于神经网络预测模型，神经网络预测模型预测准确性更高。利用统计方法与神经网络构建物理特性指标预测模型，可以为不同生态条件下提升烟叶品质、促进烟叶品质评价智能精准化提供一定的科学理论依据。

Prediction and analysis of physical characteristics indexes of light flavor high-quality tobacco

In order to explore the quantitative relationship between physical characteristics and ecological factors of high quality tobacco leaves with clear flavor, and to establish a prediction model of physical characteristics. Based on the physical characteristics index data and ecological data (meteorological, soil and altitude) of high-quality tobacco leaves in Yuxi city, Yunnan Province from 2019 to 2020, a multivariate linear statistical prediction model and BP neural network prediction model were established to analyze the relative contribution rate of ecological factors to physical characteristics of tobacco leaves. Normalized Root Mean Square Error (RMSE) and Root Mean Square Error (NRMSE) were used to test and analyze the simulation effects of the two prediction models. The results showed that the average relative contribution rate of meteorological factors was significantly higher than that of soil and altitude. The RMSE and NRMSE values of statistical prediction model are higher than those of neural network prediction model, and the prediction accuracy of neural network prediction model is higher. Statistical methods and neural networks were used to construct prediction models of physical characteristics, which could provide a scientific theoretical basis for improving tobacco quality and promoting intelligent and accurate evaluation of tobacco quality under different ecological conditions.