Detonation Prediction And Screening Of Energetic Materials By Machine Learning Combined With Quantitative Structureproperty Relationship(QSPR)

As the core materials of national defense technology and its weaponry,the precise molecular design and efficient green synthesis of high-safety and high-energy energetic materials are the major strategic needs and technical challenges in the field of energetic materials.The combination of machine learning(ML)and quantitative structure-property relationship(QSPR)is an effective tool for the research and development of new energetic materials.It is aiming to investigate the variation of molecular structure and physicochemical properties,and then realize the performance prediction and intelligent screening of high-energy materials,which can effectively improve the development and application of energetic materials.This thesis provides the theoretical basis and technical support for the practical needs of performance prediction and intelligent screening of molecular structures of novel energetic materials,which.The main research contents are as follows:Taking density and detonation velocity as research objects,a LSSVM-QSPR model was established to achieve performance prediction.First,the molecular descriptors and performance parameters of 179 energetic materials were obtained,and the calibration set and test set of the QSPR model were selected by K-S method;then,146 molecular descriptors were using to constructe three QSPR calibration models based on RF,PLS and LSSVM,and cross-validation method combined with the MRE minimization principle for model parameter selection and optimization;finally,the three QSPR calibration models based on the optimal model parameters are constructed separately and the two performances are predicted to verify their prediction performance.The results show that the LSSVM-QSPR calibration model exhibits better prediction performance compared to both RF and PLS QSPR models,both in terms of density(MREP=0.0159,R_P~2=0.9712)and detonation velocity(MREP=0.0735,R_P~2=0.6854).Therefore,the LSSVM-QSPR based burst performance prediction method for energetic materials was established,which provides new ideas and methods for the screening of new energetic materials.In order to improve the modeling efficiency of QSPR model,a variable selection method(VIM and VIP)combined with LSSVM is established for the prediction of energetic materials.First,the importance thresholds of each molecular descriptor are calculated by VIM and VIP;then,they are selected and optimized using the 10-CV method combined with the MRE minimum principle;finally,the LSSVM-QSPR calibration model is constructed under the optimal input variables and model parameters.The results show that the optimal calibration model is VIP-LSSVM-QSPR(MREP=0.0038,R_P~2=0.9968)for density and VIM-LSSVM-QSPR(MREP=0.0620,R_P~2=0.8757)for detonation velocity.Therefore,a variable selection strategy combined with LSSVM-QSPR was established for the prediction of the burst performance of energetic materials,which effectively improves the modeling efficiency of the QSPR model.Based on the above study,an energetic materials screening method was further developed based on the detonation velocity(RE<5%).First,the LSSVM-QSPR calibration model was constructed with 146 molecular descriptors as input variables,and the parameters of the model were selected and optimized using 10-CV;then,the optimized LSSVM-QSPR was used to predict the detonation velocity and calculate its common backbone and functional groups of these compounds;finally,the energetic materials with RE<5%were screened,and the molecular descriptors of the test compounds were further weighted and screened by VIM.The results showed that the R~2 and MRE between the experimental and test values of the screened energetic compounds were 0.9851 and 0.0184,respectively,and their common backbone and functional groups were benzene ring and nitro group,respectively,and there were 25 molecular descriptors with weights>1.This study provides the theoretical basis and technical support for the intelligent design and molecular structure screening of energetic materials.