| 氯代芳香族化合物结构电化学还原电位定量关系的贝叶斯规整化BP神经网络模型 |
| 摘要点击 1894 全文点击 2064 投稿时间：2004-04-23 修订日期：2004-08-10 |
| 查看HTML全文
查看全文 查看/发表评论 下载PDF阅读器 |
| 中文关键词 氯代芳香族化合物 QSPR 还原电位 贝叶斯规整化神经网络 权重平方和 |
| 英文关键词 chlorinated aromatic compounds QSPR electric potential Bayesian regularized neural model sum of square weights |
| DOI 10.13227/j.hjkx.20050205 |
|
| 中文摘要 |
| 将贝叶斯规整化误差反向传播神经网络(BRBPNN)应用于环境领域的 QSPR模型.采用ChemOffice2004内置的MOPAC 2000计算了6种量子化学参数(分子最高占据能EHOMO、分子最低占据能ELUMO、分子生成热HF、分子偶极矩DIP、分子的电子能量EE和分子的核核排斥能CCR)以及氯原子数(Cl)和分子量(MW),建立了87种氯代芳香族化合物结构与电化学还原电位定量关系的BRBPNN模型.最优网络模型结构为6-20-1,其电化学还原电位的拟合及预测能力明显优于逐步线性回归模型,其训练集和预测集的相关系数平方和均方根误差(MSE)分别达到0.999和0.000105,0.965和 0.00159.最优模型输入节点到隐含层权重平方和的分布规律揭示出各种描述符对还原电位的影响大小依次为: ELUMO>EHOMO>HF>CCR>EE>DIP.由散点图揭示出影响为正有EE;影响为负有ELUMO,HF,DIP;影响无明显正负性的有ELUMO,CCR.结果表明,贝叶斯规整化大大方便了网络规整化参数选择,保证了网络的优良概括能力和稳健性.本研究对氯代芳香族化合物采用电化学处理的适用性以及分析相应电化学降解机理提供了依据.
|
| 英文摘要 |
| Bayesian regularized BP neural network (BRBPNN) technique was applied in QSPR model in environmental field. The BRBPNN model for quantitative relationship between the electrochemical reduction potential (ERP) and chemical structures of 87 chlorinated aromatic compounds was established. The structure descriptor pool is consisted of Cl number(Cl), molecular weight(MW) and 6 quantum chemistry parameters which are calculated by MOPAC2000 built in ChemOffice2004, including energy of the highest occupied molecular orbital (EHOMO), energy of the lowest occupied molecular orbital (ELUMO), heat of formation(HF), dipole(DIP), electronic energy(EE), core-core repulsion(CCR). The achieved optimal network structure was 6-20-1, which possessed stronger fitting and prediction capacity than that of the stepwise linear regression and with the correlation coefficients square and the mean square error for the training set and the test set as 0.999 and 0.000105, 0.965 and 0.00159 respectively. The sum of square weights between each input neuron and the hidden layer of BRBPNN(6-20-1) indicate the effect of descriptor on the electric potential declining in the order of ELUMO>EHOMO>HF>CCR>EE>DIP. The scatter diagrams show that the EE descriptors had positive effect on ERP, and ELUMO, HF, DIP had negative effects, and EHOMO and CCR showed ambiguous effects. Results show that Bayesian regularized BP neural network is of automated regularization parameter selection capability and thus may ensure the excellent generation ability and robustness. This study threw more light on the applicability of electrochemical treatment for the chlorinated aromatic compounds and the analysis on electrochemical reduction mechanism. |
|
|
|
