正交试验设计在粗汽油干点多辅助变量选择中的应用

APPLICATION OF ORTHOGONAL EXPERIMENTAL DESIGN IN THE SELECTION OF CRUDE GASOLINE END-BOILING-POINT SECONDARY VARIABLES

在粗汽油干点的软测量过程中,往往有很多的辅助变量,导致辅助变量难以选择。辅助变量的选择由过程特性所决定,同时在实际应用中还应考虑经济性、可靠性、可行性以及维护性等额外因素的制约。如果直接使用过多辅助变量会出现过参数化问题,至于如何选取最佳辅助变量个数是一个有待研究的问题。在本文中提出了将正交试验设计和多元逐步回归结合的思想,来确定基于BP算法的粗汽油干点软测量模型的输入神经元个数。

In modeling the soft-sensing of crude gasoline end-boiling-point, too many secondary variables led to the difficult position of selecting the secondary variables for the modeling. The selection of the secondary variables depends on the process characteristics, At the same time, the constraints of the process economics, reliability, feasibility, maintenance, etc. are also to be considered in application. Over-paramenterization problem was expected to occur if too many secondary variables were directly used. How to select an optimum number of secondary variables? A method combining the multiple stepwise regression method (MSR) and the orthogonal experimental design was proposed to determine the number of input nerve cells of crude gasoline end-boiling-point soft-sensing model based on back propagation (BP).