Application Of Doped TiO₂ In Photocatalysis Based On Machine Learning

TiO₂ photocatalytic technology,as a protection technology of great significance,can be applied to environmental protection,such as the photocatalytic degradation of organic dyeing and printing wastewater from textile industry,to solve the issues of low efficiency and high cost of traditional wastewater treatment technology.It can also be applied to modern new energy,for instance,photocatalytic water splitting for hydrogen production,developing clean production of hydrogen energy,saving resources,reducing energy consumption,and achieving green and low-carbon cycle development.In this thesis,machine learning technology is applied to the modeling of doped TiO₂ photocatalysis,avoiding the complex mechanism of photocatalytic reactions.According to data,a fitting model is established to discover the hidden relationship between input variables（dopants and experimental conditions）and output data（catalytic performance）.The most obvious advantage of this technique is that it can easily approach any nonlinear continuous function,and automatically discover hidden relationships and patterns from large amounts of raw data to predict possible outcomes before the experiment,thus helping the researcher to reduce trial blindness and optimize the experimental plan.Regression models are developed to predict the degradation rate of photocatalytic pollutants by doped TiO₂.Using a cross-validation method and evaluating the fit of the four models in the training set,we found that the Light GBM model had the best confidence and highest correlation of prediction results.The application of machine learning provided a viable route to rank the importance of experimental variables affecting catalytic activity in the degradation of doped TiO₂.The Light GBM model proposed the following influence sequence:illumination time>dopant/Timole ratio>catalyst/pollutant mass ratio>calcination temperature>light wavelength>dopant>p H>pollutant>experimental temperature.We also designed and developed a fusion model based on stacking algorithm.By comparison,the fusion model has higher reliability and better correlation of prediction results than a single linear model or decision tree model in the study of hydrogen production rate of doped TiO₂ photocatalytic splitting water.In the data collection stage,15 physicochemical properties were used as inputs to describe the doping elements.The little information in the data set and insufficient doping elements can also be used for effective predictive analysis.Considering the importance of these characteristic variables,we found that the experimental conditions had a greater impact on the hydrogen production rate than the properties of the doped elements.The decision tree classification algorithm can also provide some heuristic and regular data information for achieving high hydrogen production rate.The research strategies and methods in this thesis provide an effective path and scheme for the selection of environmental and energy materials and the design of experimental routes,as well as for the high-throughput calculation of textile fiber material genomics.