The Application Of Data Mining Methods In Aluminophosphate Syntheses Data Analysis

Recently, the rational synthesis of new microporous materials has become a significantlyimportant topic in the field of current molecular sieve research. However, the designedsynthesis of microporous materials is of great challenge because of unclear the complexity ofhydrothermal chemical reactions such as the gel composition, the template, the solvent, thecrystallization temperature and time. In the past years, researchers have tried to establishprediction models of new synthesis methods. Specially, they applied some data miningmethods to the rational designs of target materials in order to obtain a good prediction modelfor a specific structure, which was used for the synthesis of new materials and obtained goodpredictive results in chemical material analysis. But the study on analysis and prediction ofopen-framework aluminophosphates (AlPOs) is not enough. In view of the rich chemicalstructure of open-framework AlPOs, the theories and methods of machine learning based onstatistics are employed to analyze and predict the structures of AlPOs molecular sieves in thisthesis.The detailed study is divided into the following two parts:Part I: Aiming to deal with the problem of missing values, four missing value estimationmethods are employed on the database of AlPO syntheses for the first time.Database of AlPO syntheses contains missing values about29%of total. For moreaccurate analyze of the data, the fuller use of existing data to avoid the impact of the work onthe follow-up research, to complete the database, we employed four missing value estimationmethods on the database of AlPO syntheses to solve the miss value problem. A large numberof experimental results show that the estimation methods are competent for microporousaluminophosphates and the BPimpute is recommended. We also found that during theexperiment, these methods also have the role of correcting the value of the parameter of theexisting data.Part II: To analyze the relationship between the synthetic factors and the specificresulting structure on the database of AlPO syntheses, two novel feature selection methods areproposedA novel fusion feature selection model and an integrated feature selection model havebeen presented to analysis the relationship between the synthetic factors and the specificresulting structure on the database of AlPO syntheses. In particular, a weighted voting schemeis applied to order the significance of synthetic factors. A large number of experimental resultsshow that the proposed model is efficient and feasible. The result also provides importantguidance for the rational design and synthesis of microporous materials.Based on a survey of feature selection methods, a new feature selection model (random subspace method combined with fisher score, FRS) is proposed to select and order thesynthetic factors which affect the predictive results importantly and directly on database ofAlPO syntheses. A large number of experimental results show that the proposed model isefficient and feasible.In this thesis, data mining methods based on statistics are employed to analyze andestablish a series of predictive models of synthesis parameters to resultant specific structureon AlPOs database. The research in this thesis will provide important guiding significance forrational designing the molecular sieves framework with specific structures.