25种持久性有机污染物飞灰-水分配系数的定量结构性质关系研究

Study on the quantitative prediction of the soot-water partition coefficient of 25 persistent organic pollutants

分别采用多元线性回归(MLR)和径向基函数人工神经网络(RBF-ANN)建立了2个不同的持久性有机污染物飞灰-水分配系数(Ksc)的定量结构性质关系(QSPR)模型,并应用留一交叉验证方法对所建立的模型进行了检验.用所建立的模型研究了25种有机污染物的飞灰-水分配系数的定量结构性质关系.结果表明:MLR模型预测的lg Ksc均方相对误差为6.31%,全部样本lg Ksc的预测值与实验值之间决定系数(R2)为0.823 3;RBF-ANN模型预测的lgKsc均方相对误差为3.03%,决定系数为0.962 3.这说明MLR和RBF-ANN方法都能够用于建立被研究化合物Ksc的QSPR模型,RBF-ANN模型的预测准确度更高.

Two quantitative structure property relationship(QSPR) models for predicting the soot-water partition coefficient(Ksc) of persistent organic pollutant(POP) were established using multivariate linear regression(MLR) and radial basis function artificial neural network(RBF-ANN) respectively.The obtained models were assessed with leave one out cross validation.The soot-water partition coefficients(Ksc) of 25 persistent organic pollutants were predicted using two models.The relative square-error and the correlation coefficient between the data predicted by using MLR model and the measured data are 6.31% and 0.823 3 separately,while those by using RBF-ANN model are 3.03% and 0.962 3 separately,which shows that two methods all can be used for establishing the soot-water partition coefficient prediction model,but the prediction accuracy of the model established by RBF-ANN is higher than that by MLR.