First Principles Combined With Machine Learning Design Of Pure Organic Dye Photosensitizer For Photocatalytic Hydrogen Production System

Hydrogen energy,as an ideal clean renewable energy,plays an important role in the national energy system.At present,the development of hydrogen energy in China still faces the problems of low production efficiency and high comprehensive application cost,which hinder the development of hydrogen energy industry chain.Hydrogen production from renewable energy sources,such as photocatalytic hydrogen production,is a promising technology that has attracted extensive attention from researchers.Photosensitizer is the energy catcher of hydrogen production reaction.It is very important to find visible light absorption photosensitizer with with excellent performance in hydrogen production reaction to promote the development of photocatalytic hydrogen production technology.Studies have proved that the hydrogen production activity of pure organic dyes can be higher than that of noble metal complexes under the same experimental conditions.And pure organic dyes have the advantages of variety,easy synthesis and structural modification,and low application cost.Therefore,the construction of non-noble metal photocatalytic hydrogen production system by using pure organic dyes instead of noble-metal based photosensitizer is expected to bring new breakthroughs in the development of photocatalytic hydrogen production system.In this paper,we take the organic photosensitizer in the photocatalytic hydrogen production as the research object,and systematically study the hydrogen production performance of photosensitizer by using first-principles calculation combined with machine learning technology.The main research content is as follows:（1）At present,there are few theoretical researches on photosensitizers for hydrogen production,and there is lack of calculation standard for theoretical simulation of photosensitizers.Therefore,we optimized the configuration of rhodamine S150 by first-principles calculation.Among them,the function LSDA combined with base group6-31G（d,p）is the most suitable calculation setting for the experimental value.The difference between the calculated energy level and the experimental value is only 0.04e V,and the relative error is less than 1%,which provides theoretical support for studying the physical parameters of the newly designed organic photosensitivities.（2）After that,we taked dye S150 which has high hydrogen production activity as the prototype and studied the electronic and optical properties of rhodamine dye S150 by modifying its structure with extended side chain.The results show that both S156 and S162 molecules have broad spectrum absorption spectra and good visible light response.In particular,the molar extinction coefficient of S156 molecule is as high as2.65×10^4-1?（88）^-1,and they have larger photoinduced electron transfer driving force and more negative reduction potential.The reduction potential of S162 is 22.6%higher than that of S150.Therefore,dyes S156 and S162 extended by side chains are promising candidates for photosensitizers in homogeneous photocatalytic hydrogen production systems.（3）The material design method based on first principles has made remarkable progress in discovering new materials,studying material properties and evaluating material performance parameters,however,it also has the problem of high demand for computing resources and long time consuming.Therefore,it is urgent to develop new technologies to speed up the design and development of new materials.Machine learning technology can accelerate the development of new materials by integrating the understanding of the properties and mechanisms of molecular materials into data science models.In this paper,we build a machine learning model for photosensitivities in photocatalytic hydrogen production.After learning and training the model,the mean square error（MSE）,root mean square error（RMSE）and mean absolute error（MAE）of Bagging regression algorithm in the machine learning model are 0.067,0.259 and 0.196respectively.Bagging regression algorithm shows higher learning accuracy than the other machine learning methods in the photosensitizer machine learning model.The results show that the proposed machine learning model of photosensitizer is reliable.Our study is expected to provide data reference and theoretical guidance for the experimental synthesis of organic photosensitizer molecules.In this paper,we explored the relationship between molecular structure and properties of organic photosensitizers in photocatalytic hydrogen production system based on first principles.We achieved an optimized cycle of accelerating the discovery of novel photosensitizers by establishing a performance prediction model for the properties of photosensitizers and their hydrogen production activity,and broke through the limitations of experimental science in the development of new materials.It will be the trend in the future of materials research by first-principles computing and machine learning techniques to assist the development of new materials.