| 177 | 2 | 467 |
| 下载次数 | 被引频次 | 阅读次数 |
提出了以Gibbs-Duhem方程为基架,引入前传网,由此将机理方程与神经网络这两种不同类的方法相集成,用以求解化工热力学的汽液相平衡问题,以克服经典的机理方法或单一神经元方法的欠缺与不足。将这种集成方法应用于恒温、恒压二元体系,由液相组成确定汽相组成,效果良好。应用实例表明该法计算简便,预测精度高,且稳定性好,显示出胜于两种单一方法的优越性。
Abstract：A hybrid model (GD-FFN) that combines Gibbs-Duhem equation (GD) and artificial neural networks (FFN) for vapor liquid equilibrium (VLE) calculation was presented. In which a simple expression for vapor composition was deduced and FFN was included in the expression to compensate inaccuracy of first principles. The model's predictions are restricted by GD, the deficiency, i.e. VLE calculation using pure FFN is independent on thermodynamics mechanism, is avoided. So the model has better universality and stability. Otherwise the model has simpler form and easier computation compared with traditional UNIQUAC method. The vapor compositions of the selected isobaric (isothermal) VLE binary systems were predicted by using GD-FFN. The results indicate the GD-FFN model can produce more accurate and consistent predictions, and can provide more reliable extrapolation than pure FFN or UNIQUAC method.
[1]HU Ying(胡英).NewAdvances in theApplicationStudy ofModernChemicalEngineeringThermodynamics(近代化工热力学应用研究的新进展)[M].Shanghai(上海):ShanghaiScientific andTechnicalLiteraturePress(上海科学技术文献出版社),1994.
[2]ThompsonM L,KramerM A.Modeling chemical processes using prior knowledge and neural networks[J].AIChE Journal,1994,40(8):1328-1340.
[3]CHEN Chong-wei(陈翀伟),CHEN De-zhao(陈德钊).Feed-forward networks based on prior-knowledge and its application in modeling the true boiling point curve of the crude oil(基于先验知识的前馈网络对原油实沸点蒸馏曲线的仿真)[J].J ChemEng ofChineseUniv(高校化学工程学报),2001,15(4):351-356.
[4]CHEN Chong-wei(陈翀伟),CHEN De-zhao(陈德钊).An improved differential evolution algorithm in training and encoding prior knowledge into feedforward networks with application in chemistry[J].Chemometrics andIntelligentLaboratorySystems,2001,64(1):27-43.
[5]HornikK,StinchcombeM,WhiteH.Multilayer feedforward networks are universal approximators[J].NeuralNetworks,1989,2:359-366.
[6]YAN Xue-feng(颜学峰),CHEN De-zhao(陈德钊),HU Shang-xu(胡上序).Combinational neural network with the similarrandom training and its application to design the molecular structure of substituted aromatic sulfur derivatives(用类随机训练组合网络以辅助含硫苯衍生物分子设计)[J].J ChemEng ofChineseUniv(高校化学工程学报),2000,14(6):577-582.
[7]ZHUANG Ling(庄凌),CHEN De-zhao(陈德钊),ZHAO Wei-xiang(赵伟祥),ZHANG Hong(张红),HU Shang-xu(胡上序).The radial basis functions-cyclic subspace regression approach and its application to cracker modeling(RBF-CSR方法及其用于裂解装置建模的研究)[J].J ChemEng ofChineseUniv(高校化学工程学报),2002,16(1):64-69.
[8]YAN Ping-fan(阎平凡),ZHANG Chang-shui(张长水).ArtificialNeuralNetworks andSimulatedEvolutionaryComputation(人工神经网络与模拟进化计算)[M].Beijing(北京):TsinghuaUniversityPress(清华大学出版社),2000.
[9]HU Ying(胡英),YingXu-gen(英徐根).A new method of determining equilibrium vapor composition via solution's vapor pressure or boiling point(由溶液蒸汽压或沸点确定平衡气相组成的新方法)[J].J ChemInd andEng(化工学报),1980,(1):27-36.
[10]GmehlingJ,OnkenU.Vapor-LiquidEquilibriumDataCollection:OrganicHydroxyCompoundsAlcohols[M].Gesellschaft:DECHEMA,1977.
[11]GmehlingJ,OnkenU.Vapor-LiquidEquilibriumDataCollection:Aqueous-OrganicSystems[M].Gesellschaft:DECHEMA,1977.
基本信息:
中图分类号:O642
引用信息:
[1]陶少辉,胡望明,陈德钊.Gibbs-Duhem方程与前传网的集成方法及其在汽液相平衡中的应用[J],2004(03):357-361.
基金信息:
国家自然科学基金项目资助(20276063)