神经网络法研究多溴代二苯胺热力学性质.pdfVIP

CHINESEJ0URNAL0FCHEMICALPHYSICS V0LUME 28．NUMBER 1 FEBRUARY 27．2015 ARTICLE ResearchonThermodynam icPropertiesofPolybrominatedDiphenylam ine byNeuralNetwork xi—huaDu，Wen—changZhuang，Xiao—qinShi，Chang-junFeng SchoolofChem~tryandChemicalEngineering,XuzhouInstituteofTechnology,Xuzhou221111，China (Dated：ReceivedonJune22，2014；AcceptedonJanua~10．2015) Basedonthelocationofbrominesubstituentsandcon ．a 一 ． iugationmatrix newsubstituentpo sitionindexuX notonlywasdefined，butalsomolecularshapeindexes m andelectronega- tivitydistancevectorsM ofdiphenylamineand209kindsofpolybrominateddiphenylam ine fPBDPA1moleculeswerecalculated．Thenthequantitativestructure—propertyrelationships (QSPR)amongthethermodynamicpropertiesof210organicpollutantsand X，K3，M29， M36werefoundedbyLeaps—and—Boundsregression．Usingthefourstructuralparametersas inputneuronsoftheartificialneuralnetwork，threesatisfactoryQSPRmodelswithnetwork structuresof4：21：1，4：24：1，and 4：24：1respectively，wereachievedbytheback-propagation algorithm．ThetotalcorrelationcoemcientsR were0．9999．0．9997．and0．9995respectively and thestandard errorsS were 1．036，1．469，and 1．510 respectively．Therelativemean deviationbetweenthepredictedvalueandtheexperimenta1valueofS ．△ ，日 and△ fG were0．11％，0．34％and0．24％respectively,whichindicatedthattheOSPRmodelshadgood stabilityandsuperiorpredictiveability．Theresultsshowedthatthereweregoodnonlinear correlationsbetweenthethermodynamicpropertiesofPBDPAsandthefonrstructuralpa— rameters．Thus，itwasconcludedthattheANN modelsestablishedbythenew substituent positionindexwerefullyapplicabletopredictpropertiesofPBDPAs． Keywords：Polybrominateddiphenylam ine，Neuralnetworks，Molec