DFT Calculation And Machine Learning Assisted Analysis Of Ammonia Nitrogen Adsorption Structure Of Nitrogen-oxygen Doping In Biochar

In this study,the adsorption behavior and mechanism of different modified structures of nitrogen-oxygen doped modified biochar on ammonia and nitrogen nutrients are unknown.Based on the DFT calculation data,an intelligent machine learning modeling method was used to develop a prediction model for the adsorption energy,minimum and maximum electrostatic potential of nitrogen-oxygen modified biochar on ammonia nitrogen,which provides a model for the rapid calculation and accurate prediction of key parameters such as adsorption energy and electrostatic potential in the process of ammonia nitrogen adsorption on biochar.It provides a model basis for the rapid calculation and accurate prediction of key parameters such as adsorption energy and electrostatic potential during ammonia-nitrogen adsorption on biochar.The main research contents and conclusions are as follows:（1）In an attempt to reveal the influence and mechanism of the fugitive morphology of nitrogen-oxygen doped structures and their binding sites on the biochar skeleton on the ammonia-nitrogen adsorption behavior of biochar,DFT calculations were used to investigate the adsorption mechanism of NH₄⁺by nitrogen-oxygen monodoped groups on the biochar skeleton and the change pattern.These results show that the interaction of NH₄⁺with nitrogen-containing groups and oxygen-containing groups depends more on the type of functional groups and the adsorption sites play a synergistic role.Among the oxygen-containing groups,the best average adsorption energy was observed for nitrogen oxide,with a 9.3 fold enhancement compared to pristine biochar;among the oxygen-containing groups,the best adsorption energy was observed for carbonyl groups,with a 5 fold increase in adsorption energy.In addition,the influence of adsorption sites on the adsorption energy ranged from 3.2 k J/mol to 16.95 k J/mol.The edge sites are mainly covalent adsorption,while the in-plane sites are van der Waals force interactions.（2）In order to elucidate the coupling behaviors and mechanisms of ammonia nitrogen adsorption by nitrogen and oxygen doped structures at different positions of biochar skeletons and their different grouping forms,the adsorption of ammonia nitrogen by biochar skeletons in the case of nitrogen and oxygen double doping was considered and analyzed in this study.It was shown that the adsorption of NH₄⁺by biochar was promoted when the spacing of nitrogen-oxygen double-doped groups was less than 5?.On the contrary,the adsorption of NH₄⁺by biochar was inhibited.The adsorption of NH₄⁺by the synergistic effect of nitrogen groups and oxygen groups mainly relies on cation-π,van der Waals interaction with electrostatic attraction and hydrogen bonding.The adsorption process leads to the change of electron density around NH₄⁺,resulting in the local electron density of the carbon skeleton being remodeled.The adsorption of NH₄⁺by biochar singly doped with nitrogen and oxygen depends mainly on the properties of the dopant groups themselves,with the adsorption sites acting synergistically.Adsorption of NH₄⁺by the nitrogen-oxygen double-doped biochar was limited by the spacing of the two groups,and the group with the stronger adsorption capacity played the dominant adsorption role,and the adsorption site had no significant effect on the adsorption capacity of the group.（3）Three data set partitioning methods（RS,KS,and SPXY）were used to perform the cut-off,and the results showed that the SPXY partitioning method could quickly find the optimal distribution training set for the biochar nitrogen and oxygen modification data set in this study.The Spearman correlation coefficient was selected to characterize and filter the biochar adsorption performance dataset to predict the adsorption energy,electrostatic potential maxima and electrostatic potential minima.Compare the performance of five types of machine learning,select the optimal prediction model and test the robustness of the model.The importance of using features for each type of prediction models was analyzed,and the results showed that group type and doping concentration were the key features affecting the adsorption energy;group spacing and group type were the key features affecting the prediction models of electrostatic potential maximum and electrostatic potential minimum.