基于特征选择的决策树方法在磷酸铝AlPO4-5定向合成中的应用

Decision Trees Combined with Feature Selection for the Rational Synthesis of Aluminophosphate AlPO4-5

分子筛类开放骨架材料的合成与结构关系研究对实现这类材料的定向合成起着至关重要的作用. 本文在建立开放骨架磷酸铝合成反应数据库的基础上, 提出了利用基于特征选择的决策树(C5.0)方法, 考察了不同反应条件(即各反应特征参数)对磷酸铝分子筛AlPO4-5生成的影响. 基于决策树模型, 利用8个反应特征参数,可以有效预测磷酸铝分子筛AlPO4-5的生成, 准确率达到88.18%, 接收者操作特性(ROC)曲线下面积(AUC)达到90%. 研究结果表明, 在众多的反应特征参数中, 有机模板剂的几何尺寸参数, 特别是模板剂的次长距离, 是影响AlPO4-5分子筛合成的重要因素.

The relationship between the synthetic features and the types of final product is critical for the rational synthesis of zeolite-type open-framework materials. In this paper, an AIPO4-5 prediction system based on C5.0 combined with a feature selection is proposed on the basis of the establishment of a database of AIPO syntheses. 26 synthetic parameters associated with gel composition, an organic amine template and a solvent were used as input to predict the formation of AIPO4-5. The effects of different synthetic parameters on the formation of AIPO4-5 were also studied. The performance of the method was evaluated using classification accuracy and a receiver operating characteristic （ROC） curve. The results show that the highest area under the ROC curve （90%） and the classification accuracy （88.18%） was obtained for the decision tree model that contains eight input features and some useful rules with high confidence degrees were extracted from the model. Among the various synthetic parameters the geometric size of the organic template, particularly the second longest distance of the template plays an important role in the formation of AIPO4-5.