Abstract:

A genetic algorithm-based BP neural network method is constructed to predict the lower flammability limit temperature(LFLT) of organic compounds.The parameters in the model are molecular descriptors of Mv, CID, EEig02d, GGI1, nROH, nHDon.They are selected from 1664 molecular-based parameters obtained from Dragon software.For 1171 organic compounds used in this work, the average relative error and the average absolute error of this model are 3.23% and 10.28 K respectively.The correlation coefficient is 0.9833.The results are better than those obtained by genetic algorithm multiple linear regression(GA-MLR) analysis.The present model can be used to reveal the quantitative relation between LFLT and molecular structures of organic compounds and predict the LFLT of a wide range of organic compounds.

摘要：

由1171种有机物组成样本集，将利用Dragon软件计算出来的分子结构描述符Mv、CID、EEig02d、GGI1、nROH、nHDon等数值与有机物的易燃下限温度进行关联，借助遗传神经网络方法建立了相应的定量关系模型。结果表明，在给定条件下，由该模型获得的预测值平均相对误差为3.23%，平均绝对误差为10.28 K，相关系数为0.9833。新建立的有机物易燃下限温度预测方法具有模型建立简便、预测精度较高、适用面宽等优点，有望在有机危险物品的火灾性能预测及其安全使用方面发挥重要作用。