摘要
本文提出1种新的分析正交设计实验方法,在不增加试验次数的前提下同时考察各因素单项、相互项及平方项对实验结果的影响。还提出1种在线性回归过程中各项的淘汰-增加-回归循环的优化参数方法,解决了长期困扰科技工作者对回归方程中各项影响因素的取舍难题。具体做法为:利用水平编码方式,改造正交设计L9(34),再扩展,增加所有可能相互作用的列和平方列。试验后,基于正交设计L9(34)扩展后的影响因素(含单项、相互作用项、平方项)与实验结果之间的线性回归。回归过程中逐步淘汰影响不显著的项目,同时增加第1步尚未包括的项目,然后2次回归。不断重复淘汰-增加-回归循环,直到回归方程与实验数据拟合较好为止(复相关系数R的显著性水平0.05或更低)。实例说明这方法是行之有效的。此外,回归方程在各因素取值范围内求最大值所获得的最佳条件,已远远优于直观分析获得的最优条件组合。最后,编码回归方程考察的项目更加全面。
A new method was proposed in this paper to study the influence of single items,interaction items,and square items of factors to the experimental results.In addition,the method eliminates and select the significant items in the regression of linear equation is also suggested,which always troubles scientific workers.the specific steps are：（1）Coding of the level of each factor,and forming the coded orthogonal table.（2）Expansion of orthogonal designing table to include single items,interaction items,and square items.（3）Regression of linear equation based on the experimental results.During equation regression,the most unimportant items will be eliminated and the most important items within the deleted items in the first round of regression will be added,and regress again,and repeat the elimination,adding and regression,until the correlation coefficient is significant,example showed that such calculation is feasible and effective.The optimistic parameter combination is much better than that obtained with direct analysis.The example also indicated that the coded equation regression studies more items than the coded equation regression.
出处
《计算机与应用化学》
CAS
CSCD
北大核心
2010年第11期1503-1508,共6页
Computers and Applied Chemistry
关键词
正交设计
编码扩展
线性回归
参数优化
orthogonal table
coding and expansion
linear equation regression
parameter optimization
