中文摘要：

对于输入输出之间作用关系复杂且输出质量特性存在多极值的复杂过程，生产实践中通常拥有部分先验知识。以往研究中忽略这些先验知识进行实验设计建模导致样本浪费。提出一种基于先验知识的分区域式实验设计与建模方法。首先根据先验知识对各因子取值区间进行划分；其次利用模糊评价方法对波动大小进行度量；然后利用均匀设计方法在由各维度因子分段组合形成的子区域上安排试验点；最后，利用全部样本信息建立复杂作用关系过程的支持向量机回归模型。算例研究表明，与传统均匀实验设计建模方法所建模型相比，所提方法所建模型的三个预测误差指标值平均降低了18．1％，说明所提方法建立的模型具有更好的预测性能。

英文摘要：

Generally,in production practices there is some prior knowledge of the complex process featured with multi-extremums of output quality characteristics and complex relationship between input parameters and output characteristics.Ignoring this prior knowledge could cause a waste of the sample in the design of experiments and modeling of the past researches.A sub-regional design of experiments and modeling approach based on the prior knowledge is proposed in this paper.Firstly,the value range of each factor is divided according to prior knowledge of a process;Secondly,fuzzy evaluation method was applied to measure the volatility;Thirdly,a uniform design （UD）was used to arrange trial points for each sub-region formed of the combination of sections of each factor in different dimensions;Lastly,a SVM regression model of the complex process was constructed using the information obtained from the sample.Case study results illustrate that,compared with the model obtained by traditional UD and modeling method, selected three prediction index values of the model obtained by the proposed approach are reduced by an average of 18.1%.It indicates that the model constructed by the proposed approach has better predictive performance.