Modeling Research On The Relationship Between Volatile Organic Compounds And Free Radicals In The Troposphere

A steady stream of volatile organic compounds are released into the atmosphere every day around the world.The rapid develoment of urbanization results in the increasement of the ampunt of VOCs into the atmosphere.The accumulation of volatile organic compounds into the atmosphere makes the air quality worse,which brings great challenges to human health and atmospheric environment.Volatile organic compounds are key precursors for the formation of secondary organic aerosols（SOA）.Secondary organic aerosols are an important part of particulate matter and a major contributor to haze pollution.Most volatile organic compounds can undergo degradation reactions with free radicals（such as Hydroxyl radicals,Nitro free radicals）and（such as ozone）active molecules present in the atmosphere.Therefore,studying the reaction mechanism of volatile organic compounds and free radicals is crucial for evaluating the fate of volatile organic compounds in the environment.In chemical kinetics,the rate constants of chemical reactions are affected by the properties of the reactants.Therefore,evaluating the persistence of volatile organic compounds requires studying the kinetic constants of the reaction between volatile organic compounds and free radicals.In this dissertation,the quantitative structure-property/activity relationship（QSPR/QSAR）method was used to explore the relationship between the kinetic constants of the reaction between volatile organic compounds and hydroxyl radicals and nitro radicals in the atmospheric troposphere and their chemical structures.Analyze the effect of molecular structure on reaction kinetic constants.The specific contents include:1.A QSPR Model for Predicting Reaction Kinetic Constants for the Reaction of Volatile Organic Compounds with Nitro Radicals.Based on DRAGON and CODESSA theoretical molecular descriptors of volatile organic compounds（VOCs）,quantitative structure–property relationship（QSPR）models were developed by using multi-linear regression model（MLR）,support vector machines（SVM）and Project pursuit regression（PPR）to study their reactivity with Nitro free radicals（NO₃·）in the troposphere.Structural features affecting the reaction between the VOCs and nitrate radicals were also instigated.Rate constant was transformed into negative logarithmic unit,-7)2)6)₃.Specific work includes:The training set and the test set were divided based on random grouping and principal component analysis（PCA）,and the DRAGON-MLR model was established by the multiple linear regression（MLR）method.It is obtained by analyzing the results of the DRAGON-MLR model with different proportions of random grouping and PCA grouping.By analyzing the results of the DRAGON-MLR models with different ratios of the two different division methods,it is concluded that the division is more reasonable when the grouping ratio of the training set samples and the test set samples is 4:1.Comparing the two grouping methods,it is found that the distribution of training set samples and test set samples divided by the PCA grouping method is relatively uniform under the same proportion,indicating that the PCA grouping method has more advantages in selecting the training set and the test set.After the grouping method and grouping ratio are determined,The CODESSA-MLR model was established by using the CODESSA descriptor as the input variable using the multiple linear regression method.Comparing the DRAGON-MLR model and the CODESSA-MLR model,it is found that the model built with the CODESSA descriptor is better than the model built with the DRAGON descriptor.Taking into account the complexity of influencing factors.Considering the complexity of the influencing factors,the MLR model was established by combining the DRAGON descriptor and the CODESSA descriptor using multiple linear regression.It is found that the model results of joint modeling have better generalization ability.For the test set,MLR gave a predictive squared correlation coefficient R² of 0.928,root mean square error（RMSE）of 0.548,absolute average relative deviation（AARD）of 3.556%.In addition to the linear modeling method,this paper also adopts the nonlinear modeling method for modeling.Modeling the training set and test set of PCA groupings with support vector machine（SVM）modeling method found that the results of the CODESSA-SVM model were significantly improved.The SVM model jointly established by the two descriptors outperforms the results of independent modeling by a single descriptor.MLR gave a predictive squared correlation coefficient R² of 0.950,root mean square error（RMSE）of 0.469,absolute average relative deviation（AARD）of 2.833%.Compared with the MLR model,the SVM model has better fitting ability,so the relationship between the reaction kinetic constant and the descriptor cannot be simply described by a linear relationship.Projection pursuit regression method is also used to model.For the test set,PPR gave a predictive squared correlation coefficient R² of 0.956,root mean square error（RMSE）of 0.434 and absolute average relative deviation（AARD）of2.563%,respectively.Finally,the optimal model was used to predict 56 compounds whose reaction kinetic constants had not yet been detected.2.A QSPR Model for Predicting Reaction Kinetic Constants for the Reaction of Volatile Organic Compounds with Hydroxyl Radicals.Based on DRAGON theoretical molecular descriptors of volatile organic compounds（VOCs）,quantitative structure–property relationship（QSPR）models were developed by using multi-linear regression model（MLR）,support vector machines（SVM）and Project pursuit regression（PPR）to study their reactivity with Hydroxyl radicals（·OH）in the troposphere.Structural features affecting the reaction between the VOCs and nitrate radicals were also instigated.Rate constant was transformed into negative logarithmic unit,-7)2)6).PPR model performs better than MLR model both in the fitness and in the prediction capacity indicating its good generalization capability.For the test set,MLR gave a predictive squared correlation coefficient R² of 0.866,root mean square error（RMSE）of 0.443 and absolute average relative deviation（AARD）of 3.236%,For the test set,SVM gave a predictive squared correlation coefficient R² of 0.883,root mean square error（RMSE）of 0.410 and absolute average relative deviation（AARD）of 2.854%,For the test set,PPR gave a predictive squared correlation coefficient R² of 0.898,root mean square error（RMSE）of 0.382 and absolute average relative deviation（AARD）of 2.645%.The PPR model has better fitting ability,indicating that the nonlinear model is more suitable for fitting the reaction rate constants of hydroxyl radicals and volatile organic compounds.