基于神经网络的丙酮产品质量分类挖掘

Data Mining for Product Quality Classification of Acetone Refining Process Based on Neural Network

针对丙酮精制过程的特点,提出一种基于神经网络的丙酮产品质量分类挖掘方法。首先,讨论了数据挖掘中自变量筛选的方法,包括相关性分析、Fisher指数分析、主成分回归分析以及偏最小二乘回归分析等,综合各种方法分析的结果,对丙酮精制过程中众多的工艺影响因素进行了重要性排序并据此筛选出重要的自变量;以选入的变量作为输入变量,构造基于神经网络的产品质量分类器。实验结果表明,训练后的神经网络分类器在丙酮产品质量分类挖掘中取得了良好的效果。

Considering the features of acetone refining process, a strategy of neural network based data mining for product quality classification is proposed. First, independent variables are analyzed by several methods, including correlation analysis, Fisher index analysis, principal component regression （PCR）, and partial least square （PLS） regression. Important independent variables are then selected according to the analysis results of these methods with an importance order of the technologic influence factors in acetone refining process. Using the selected variables as input variables, a product quality classifier based on neural network is constructed. Experiment results show that trained neural network classifier achieves good effects in the acetone product quality classification mining.