一种近正交试验设计方法

为提高对多维输入变量模型有效分析研究的能力,找出试验设计中正交性与填充空间性质的结合点,减少试验设计算法本身的复杂度,结合正交试验设计和均匀试验设计,应用对正交性及均匀性的多重度量方法提出了一种近正交试验设计方法,利用该方法所设计的近正交试验具有近正交性和良好的填充空间性质.应用本方法设计了一个8-11个变量,每个变量33个水平的近正交试验,通过对比分析,证明了使用该方法设计的近正交试验具有近正交性和良好的填充空间性质,从而验证了该文提出方法的正确性和可行性.

A Method of Nearly Orthogonal Experimental Design

To improve on the ability to efficiently explore and approximate large subspaces of the multidimensional input variable models, find out the orthogonal and uniform combining point in experimental design, reduce the complexity of the experimental design algorithm, this research develops a new set of experimental designs for metamodeling. The traditional experimental designs are of orthogonal experimental design and uniform design, but the orthogonal experimental design has poor space-filling and the uniform design has a big correlation. The paper combines orthogonal Latin hypercube and uniform designs to create the designs with near orthogonality and excellent space-filling properties based on multiple measures. By using the method the nearly orthogonal experiment with eight to eleven variables is designed, each of which has thirty-three levels, to testify the nearly orthogonal experimental design. Through comparison and analysis, the results show that the nearly orthogonal experiment designed by this method has near orthogonality and excellent space-filling properties. So this method is feasible and of significance in improving the effect of computer experiments