基于PM3量化参数的ANN-BP_X网络预测多氯代二苯并呋喃的logKow  

Predicting the logKow of PCDFs based on PM3 quantum chemical descriptors in radial basis function Neural Network

作　　者：胡树青[1] 

机构地区：[1]青海民族学院化学系,西宁810007

出　　处：《西南民族大学学报（自然科学版）》2006年第5期931-934,共4页Journal of Southwest Minzu University(Natural Science Edition)

摘　　要：应用G98W程序包中的PM3方法对51种多氯代二苯并呋喃的分子进行优化计算,所得QC1,QC3、QC8和EHOMO量化参数作为高性能四层反向传播人工神经网络的输入向量,隐含层分别采用双曲正切(tansig)S型和对数(logsig)S型传递函数,对35种PCDFS的logKow进行建模,16种PCDFS的logKow进行预测,所建模型预测值与其实验值吻合得很好,故这一新方法具有一定的参考价值.The quantificational parameters such as QC1、QC3、QC8 and EHOMO obtained from the optimized calculation to 51 polychiorinated dibenz of urans(PCDFs) molecular by PM3 method in G98w software package are used as the inputs of the Artificial Neural Network with error back-propagation(ANN-BPx network).The model about the logKow of 35 PCDFs is established by choosing df=25 and eg=10-3 and using tansig and logsig in the middle layer,and the logKow of the other16 PCDFs is predicted.There is excellent conformability between the value calculated by the model and the experimental data.Therefore,there’s some referenced value on a certain degree.

关 键 词：人工神经网络 多氯代二苯并呋喃 

分 类 号：O626[理学—有机化学] TP391.75[理学—化学]