利用神经网络预测聚合物玻璃化温度

Predictions of glass transition temperature of terpolymers via neural networks

分析了影响聚合物玻璃化温度的各种因素，采用合成时进料组成中各组分的摩尔分数、组成成分的重量分数以及聚合度等共计６ 个变量来表达聚合物结构上的差别，通过神经网络的训练建立了描述硬脂酸乙烯酯＋ 乙酸乙烯酯＋ 氯乙烯三元共聚物链结构和组成等６ 变量空间到玻璃化温度空间的映射关系，进而对未知三元聚合物的玻璃化温度进行预测，预测结果与实验结果与比较吻合。

Mole fractions of feed materials, weight fractions of the components and degree of polymerization have been taken to describe the structure and composition characteristic of terpolymers based on the discussion of influential factors. Thirty six samples with six variables were used to train the neural network. A mapping relationship among 6 variable spaces and glass transition temperature (Tg) space was set up for the terpolymers of vinyl stearate+vinyl acetate+vinyl chloride system. The model was used to predict the Tg values of 12 unseen samples in the training set. The predictions were comparable to the experiments.