基于序列响应面方法的协同优化算法

分析了响应面协同优化的计算结构，指出影响计算结果准确度的因素，即计算过程中存在选择响应面样本点及响应面更新的问题，这是两个相互耦合的问题。分析了产生这个问题的原因并提出了一种将修正的置信域方法与中心组合设计相结合的序列响应面算法，系统地解决了这个问题。仿真算例表明由该算法得到的结果有很高的精度与鲁棒性，完全可以满足工程计算需要。

On Revised Collaborative Optimization Using Sequential Response Surface Method

Due to its parallel execution, the collaborative optimization algorithm using response surface estimation is very promising for solving large scale engineering problem. However, there exist some difficulties: selection of the sample points and refinement of the response surface, which are coupled and have very strong influence on the precision. In this paper, the cause for these difficulties is pointed out by analyzing the unique collaborative optimization algorithm structure and a comprehensive method called sequential response surface algorithm is given which combines revised trust region method and central composite design. The computation results illustrate that this algorithm has enough precision and robustness. Figs.3 and 4 illustrate clearly the contrast between the original constraints and the response surface constraints for a 2 dimensional problem. Figs.5 and 6 illustrate the iteration history from different start points for a 5 dimensional problem.