Research On Modeling For Methanation Catalysts Based On Data Mining Technology

The production of synthetic natural gas through methanation reactions has attracted increasing attention because of the shortage of natural gas.Thus,exploring excellent methanation catalysts is of significant interest.During the development of new efficient catalysts,besides economic,environmental friendly and sustainable,accelerating the discovery of catalysts is also required.Due to the complexity of the catalytic mechanism and the high non-linearity of the catalytic reaction,the traditional catalyst development based on the experience design is a process of trial and error.In the process of searching the optimal catalysts,it is contingent and difficult to meet the needs of industrial production.With the improvement of the ability of computer and information processing,the application of data mining technology to the modeling of catalytic processes is a feasible method for the development of catalysts,and it has become major concerns.The catalytic process has many influencing factors,large fluctuations,and complex reaction mechanisms,and the traditional mathematical methods is not sufficient to solve this problem.The data mining method starts from some samples and draws the knowledge that can not be obtained through the principle analysis at present.Then the knowledge is used to analyze the objective objects and predict the future data or the data that cannot be observed.They can avoid the blind mechanism of the catalyst and accelerate the discovery of new catalysts.In this study,data mining technology was employed to the modeling of methanation catalysts.The major research contents includes the design of modeling framework and prediction based on data mining technology,and the modeling prediction based on Gaussian process regression(GPR)with expected improvement(EI).The main research contents of this dissertation are as follows:(1)A new modeling framework and prediction was built by data mining technology.Faced with the high-dimensional,discrete and complex catalytic data,the principal component analysis and K-means algorithm are used to reduce the dimension and cluster the data.Then the radial basis function neural network(RBFN)with simple structure and fast convergence rate is used for modeling.The experimental results showed that the catalyst composition model was obtained and gave good prediction results was produced.It provides a scientific and efficient method for catalyst design and screening with avoiding the huge and blind experimental process.(2)An effective relevance prediction algorithm based on GPR with expected improvement was presented.In view of the deficiency of neural network modeling for small dataset,GPR is used to replace RBFN,and the EI is introduced to further improve the accuracy of the model.This method was used to construct the composition model and deactivation model of the methanation catalysts,and new catalysts were screened.This results was not obtained by RBFN.It provides a new approach for high-dimensional,small dataset,non-linear catalysts modeling.(3)Based on the well-established model,the effect of physicochemical properties on the catalytic performance is calculated by the virtual element method,which could provide important information for the design of the catalyst.