氮/氧杂环含能材料晶体密度的预测研究  

Research on Crystal Density Prediction for Nitrogen/Oxygen Heterocyclic Energetic Materials

作　　者：吕志涛 刘玉存 聂江稳 贾康辉 李强 LüZhi-tao;LIU Yu-cun;NIE Jiang-wen;JIA Kang-hui;LI Qiang(School of Environment and Safety Engineering,North University of China,Taiyuan 030051,China;Shanxi Xingxin Safety Production Technology Service Company,Taiyuan 030024,China)

机构地区：[1]中北大学环境与安全工程学院,山西太原030051 [2]山西兴新安全生产技术服务有限公司,山西太原030024

出　　处：《火炸药学报》2023年第4期299-305,共7页Chinese Journal of Explosives & Propellants

基　　金：山西省研究生教育创新项目(No.2021Y656)。

摘　　要：为了提高新型氮/氧杂环含能材料开发过程中晶体密度预测的精度,采用量子化学计算方法对所收集的已合成杂环含能材料,在3种不同的泛函理论下计算得到偶极矩、四极矩、极化率、最高占据轨道能量、最低空轨道能量等量化结构参数,并从PubChem数据库查到各化合物的氢键供体数、氢键受体数、可旋转键数3个参数,通过逐步回归方法,筛选出与密度关系密切的参数,分别建立多元线性回归(MLR)和反向传播(BP)神经网络模型。结果表明,3种泛函水平下,B3PW91泛函计算所得结果建立的模型优于其他两种,且BP神经网络模型较MLR模型预测杂环含能材料密度的误差更小、精度更高;与已有方法比较可知,在B3PW91/6-31G(d,p)水平下,BP神经网络模型的平均绝对百分比误差为3.1%,相关系数为0.956,精度最好,可以精准地预测杂环含能材料的密度。To improve the accuracy of crystal density prediction in the development process of new nitrogen/oxygen heterocyclic energetic materials,by using quantum chemical method,the quantitative structural parameters such as dipole moment,quadrupole moment,polarizability,the highest occupied orbital energy and the lowest empty orbital energy of the synthesized heterocyclic compounds were calculated under three functions.The numbers of hydrogen bond donor,hydrogen bond acceptor,and rotatable bond were obtained from the PubChem database.By using the stepwise regression method,the parameters closely related to density were selected,and the multiple linear regression(MLR)and back propagation(BP)neural network models were established,respectively.The results show that under the three function levels,the model established by B3PW91 function has the best prediction accuracy,and the BP neural network model has the higher precision accuracy in density predicting of heterocyclic energetic materials than the MLR model.Compared with the existing methods,at the B3PW91/6-31G(d,p)level,the average absolute percentage error and correlation coefficient of BP neural network model are 3.1%and 0.956,respectively,showing that the established model can be used to predict the density of heterocyclic energetic materials accurately.

关 键 词：量子化学 BP神经网络 晶体密度 定量结构-性质关系 杂环含能材料 

分 类 号：TJ55[兵器科学与技术—军事化学与烟火技术] O641[理学—物理化学]