Generalized Regression Neural Network Based Quantitative Structure-Property Relationship for the Prediction of Absorption Energy

作     者：Hui Li School of Computer Science and Information Technology Northeast Normal University Changchun,Jilin,China Yinghua Lu School of Computer Science and Information Technology Northeast Normal University Changchun,Jilin,China Ting Gao School of Computer Science and Information Technology Northeast Normal University Changchun,Jilin,China Hongzhi Li School of Computer Science and Information Technology Northeast Normal University Changchun,Jilin,China Lihong Hu School of Computer Science and Information Technology Northeast Normal University Changchun,Jilin,China 

会议名称：《2012 National Conference on Information Technology and Computer Science》

会议日期：2012年

学科分类：081704[工学-应用化学] 07[理学] 070304[理学-物理化学(含∶化学物理)] 08[工学] 0817[工学-化学工程与技术] 0703[理学-化学] 

基　　金：financial support from NSFC(20903020) 973 Program (2009CB623605) the Science and Technology Development Planning of Jilin Province (20110364,20100114) 

关 键 词：Generalized Regression Neural Network Sample subset partitioning Kennard and Stones algorithm Absorption energy Density functional theory 

摘      要：Generalized Regression Neural Network （GRNN） was used to develop a quantitative structure-property relationship （QSPR） model to improve the calculation accuracy of density functional theory （DFT）.The model has been applied to evaluate optical absorption energies of 150 organic molecules based on the molecular *** entire dataset was divided into a training set of 120 molecules and a test set of 30 molecules according to the method,termed SPXY （Sample set Partitioning based on joint x-y distances）,extended Kennard and Stones （KS） algorithm according to their differences in both x （instrumental responses） and y （predicted parameter） spaces in the calculation of inter-sample ***-propagation neural network with SPXY partitioning algorithm （BPNN-SPXY） and GRNN with KS algorithm （GRNN-KS） were also utilized to construct model to compare with the results obtained by GRNN with SPXY algorithm （GRNN-SPXY）.The root-mean-square errors in absorption energy predictions for the whole data set given by DFT,BPNN-SPXY,GRNN-KS and GRNN-SPXY were 0.47,0.21,0.17 and 0.13,*** GRNN-SPXY prediction results are in good agreement with the experimental value of absorption energy.