论文部分内容阅读
应用ChemOffice 8.0中的MOPAC-PM3算法计算得到多溴联苯醚(PBDEs)的6个量子化学参数,采用基于多项式核,径向基核及Sigmoid核的支持向量机(SVM)方法建立了23种PBDEs蒸汽压的QSPR模型。三类核函数对训练集拟合的相关系数R~2分别为0.994,0.996,0.994,均方误差MSE分别为0.0102,0.0081,0.0095;留一法交叉验证(LOO)的相关系数分别为0.992,0.991,0.991.对测试集进行回归的相关系数分别为0.994,0.986,0.991,均方误差MSE分别为0.0225,0.0458,0.0247。结果表明SVM回归算法在PBDEs蒸汽压的QSPR建模上表现出色,核函数的选择对SVM模型性能影响不显著。文章还从模型拟合效果、预测能力及稳定性三方面比较了三类核函数的整体性能,并比较了支持向量数目对核函数预测能力的影响。结果表明多项式核与Sigmoid核性能相当,优于径向基核。
Six quantum chemical parameters of polybrominated diphenylethers (PBDEs) were calculated by using MOPAC-PM3 algorithm in ChemOffice 8.0. SVP methods based on polynomial nuclei, radial base nuclei and Sigmoid nuclei were used to establish 23 PBDEs vapor pressure QSPR model. The correlation coefficients R ~ 2 of the three kinds of kernel functions to the training set were 0.994, 0.996 and 0.994, respectively. The mean square error (MSE) was 0.0102, 0.0081 and 0.0095, respectively. The correlation coefficients of LOO were 0.992, 0.991, 0.991. The regression coefficients of the test set were 0.994, 0.986 and 0.991, respectively. The mean square error (MSE) was 0.0225, 0.0458 and 0.0247 respectively. The results show that the SVM regression algorithm performs well on the QSPR modeling of PBDEs vapor pressure, and the choice of kernel function has no significant effect on the performance of SVM model. The article also compared the overall performance of the three types of kernel functions from the model fitting effect, prediction ability and stability, and compared the effect of the number of support vectors on the prediction ability of kernel function. The results show that the polynomial nucleus and Sigmoid nuclear performance is better than the radial basis.