【摘要】 应用人工神经网络构造了2个直链聚合物的结构性质关系模型。一个是直链聚合物的基团均值法描述的结构参数与其12种性质间定量关系的模型（模型1A）;一个是直链聚合物的连接性指数描述的结构参数与其12种性质间定量关系的模型（模型2A）。讨论了2个模型的参数设置,而2个模型绘出的聚合物的12种性质的拟合误差（拟合值与实验值间的标准偏差）分别是:V（298K）为18.9（模型1A）/40.5（模型2A）cc/mole,E_coh为8.019/11.122 KJ/mole,δ为0.74/2.17（J/cc）^0.5 ,F_d为228/235 J^0.5cm^1.5/mole,T_g为27/52 K,P_s为25/37（cc/mole）（dyn/cm）^1/4,n为0.0140/0.5191,ζ为7.45/5.36 10^-6cc/mole,U_R为727/593 cm^10/3/(sec^1/3mole),U_H为568/674 cm^10/3/(sec^1/3mole),H_μsum为649/719 gJ^1/3/mole^4/3,Y_d,1/2为10.6/10.5 K*kg/mole。结果表明,所建立的模型可用于直链聚合物性质的预测,而人工神经网络确实是聚合物结构性质关系研究中的一个有利的数学工具。更多还原

【Abstract】 Two structure-property relationship models were made by artificial neural networks for linear chain polymers. Model 1A quantified the relationships between the 12 properties and the descriptors which were given by the group average method, and Model 2A quantified the relationships between the 12 properties and the descriptors which were given by connectivity indexes method. Standard deviations between the fitted value and the experimental value of two models were: 18.9 (model 1A) / 40.5 (model 2A)cc/mole for V (298K) , 8.019/11.122 KJ/ mole for E doh 0.74/2.17 (J/cc)0.5 for S , 228/235 J0.5cm1.5/mole for F d 27/52 K for Tg , 25/37 (cc/mole)(dyn/cm)1/’4 for P. , 0. 0140/0.5191 for n , 7.45/5.36 10-6cc/mole for ζ , 727/593 cm10/3/(sec1/3mole) for UR , 568/674 cm10/3/(sec’/3mole) for Uu , 649/ 719 gj’/3/mole4/3 for H sum 10.6/10.5 K* kg/mole for Kd,1/2. The results indicate that the models could be used to predict the properties of linear chain polymers, and ANNs is one of a useful tool for the polymer QSPR study.更多还原