摘要
引入神经网络BP算法研究了36种炸药分子的撞击感度与其分子特征量间的关联关系.所有分子的特征量均采用DFTB3P86/6-31G**方法计算所得.共设计了8个不同的输入方案,训练和预测结果表明,在均方误差允许范围内(0.6245-4.4900),网络是可靠的.同时,含特征量(HOMO-LUMO)*BDE的方案训练预测结果最理想,说明在网络结构和训练参数基本相同的情况下,(HOMO-LUMO)*BDE与撞击感度的关联度最强,仅次于它的特征量是(HOMO-LUMO).
Backpropagation neural networks are used to study about the correlation between impact sensitivity and molecular properties of twenty-nine explosives molecules. All the molecular properties are calculated via B3P86/6-31G^＊＊ Eight different sets of molecular properties are utilized to train and test our net. The training and testing results show that the input vector with the descriptor （HOMO-LUMO） ＊ BDE can obtain the relatively better outcomes than other descriptors after training and testing several times. It further indicates that with the same net structure and training parameters, molecular descriptor （HOMO-LUMO） ＊ BDE has the strongest correlation with impact sensitivity of explosives.
出处
《四川大学学报（自然科学版）》
CAS
CSCD
北大核心
2006年第3期627-631,共5页
Journal of Sichuan University(Natural Science Edition)
基金
国家自然科学基金
中物院联合基金(10376021)
关键词
神经网络
撞击感度
BP算法
炸药分子
neural networks
impact sensitivity
backpropagation algorithm
explosive molecules
