基于响应面法的并行子空间优化算法改进研究  

Research on Concurrent Subspace Optimization Based on Response Surface Method

作　　者：刘明航 LIU Minghang(AVIC The First Aircraft Institute, Xi＇an 710089, China)

机构地区：[1]航空工业第一飞机设计研究院,陕西西安710089

出　　处：《航空科学技术》2017年第9期51-55,共5页Aeronautical Science & Technology

摘　　要：为了更好地解决飞行器多学科设计优化问题,对传统基于响应面的并行子空间优化算法(RS-CSSO)进行改进:改进基于近似模型,在具有同等计算精度的情况下减少学科分析的次数,采用均匀试验设计代替学科级优化来直接获得性能优良的初始设计样本点;在系统级优化过程中引入自适应近似模型算法,在迭代过程中对两种近似模型的精度进行对比,以相对误差的均值和标准差作为判据,选用精度更高的近似模型来提高系统级优化效率。采用改进的RS-CSSO算法对飞翼布局无人机进行了设计优化,并与传统RS-CSSO算法进行了对比。结果表明,改进的RS-CSSO不但有着更小的计算量,而且得到了更优的结果,可以应用于飞行器设计多学科优化。To solve the problem of aircraft multidisciplinary design optimization, the traditional concurrent subspace optimization based on response surface method （RS-CSSO） was studied. Then the RS-CSSO based on self-adaptive approximation model were developed. In order to reduce the amount of disciplinary analysis and keep the accuracy of the approximation models, uniform experiment design was introduced to replace the disciplinary optimization to obtain a set of design points directly, and self-adaptive approximation algorithm was introduced in system level optimization. The accuracies of two approximation models were compared in each iteration and the better model was used. The improved RS-CSSO algorithm was validated by a UAV test. In comparison with the traditional RS-CSSO optimization, the better optimum with much less computation cost was found by the improved RS-CSSO.

关 键 词：多学科优化设计 并行子空间优化 径向基神经网络 遗传算法 

分 类 号：V221[航空宇航科学与技术—飞行器设计]