基于岭回归分析的烤烟焦油含量预测模型构建

为建立烤烟焦油含量的预测模型,对烤烟总糖含量(X1)、还原糖含量(X2)、总氮含量(X3)、烟碱含量(X4)、钾含量(X5)、氯含量(X6)、糖碱比(X7)、氮碱比(X8)和钾氯比(X9)9项化学指标进行基于相关分析和统计检验的多重共线性诊断,并对其与焦油含量(Y)进行了岭回归分析。结果表明,烤烟总糖、还原糖、总氮、烟碱、糖碱比和氮碱比6项化学指标间存在共线性,且9项化学指标与焦油含量间均存在极显著相关关系,表明采用岭回归方法建立的以该9项化学指标为自变量的烤烟焦油含量多元线性回归模型是合理的。当岭回归参数为0.08时的烤烟焦油含量预测模型为Y=18.800 9–0.000 7X1–0.034 2X2+1.625 2X3+0.691 1X4–0.968 68X5+0.292 70X6–0.029 9X7–3.519 3X8–0.056 87X9(R2=0.817 5)。回归方程通过显著性检验(P<0.01),且回归系数的符号均与相关分析结果相一致,有效地使运用最小二乘法估计时符号不合理的回归系数变得合理。

Model for predicting tar content of flue-cured tobacco based on ridge regression analysis

In order to build the model for predicting the tar content of flue-cured tobacco,multicollinearity of 9 chemical indexes including total sugar(X1),reduced sugar(X2),total nitrogen(X3),nicotine(X4),potassium(X5),chlorine(X6),total sugar to nicotine ratio(X7),total nitrogen to nicotine ratio(X8) and potassium to chlorine ratio(X9) was diagnosed through correlation and statistic test,while the ridge regression between tar content and chemical components was also conducted.The results indicated that collinearity existed among total sugar,reduced sugar,total nitrogen,nicotine,total sugar to nicotine ratio,and total nitrogen to nicotine ratio.And there was extremely significant correlation between the tar content and the 9 chemical indexes mentioned above.Therefore,it was reasonable to establish the multiple linear regression model using 9 chemical components as independent variables based on ridge regression.The prediction model for tar content using ridge regression at parameter K=0.08 was Y = 18.800 9– 0.000 7X1 – 0.034 2X2 + 1.625 2X3 + 0.691 1X4 – 0.968 68X5 + 0.292 70X6 – 0.029 9X7 – 3.519 3X8– 0.056 87X9(R2=0.817 5),and the regression equation passed the significance test(P<0.01).The regression coefficients properties of ridge regression equation were consistent with the results of correlation analysis,which made the unreasonable symbols of regression coefficients in the least square estimation reasonable.