基于人工神经网络的多学科优化设计研究

Multidisciplinary design optimization based on artificial neural network

多学科优化设计的两大难点是子学科间的信息交换和系统分析计算的复杂性。为此,在一致性约束算法和并行子空间算法基础上,提出了一种基于人工神经网络响应面的多学科优化设计算法,它是一种二级结构的优化方法,即学科层仅满足局部约束,系统层提供一种协调学科间冲突的机制,保证在相关变量和耦合变量上的一致性,使设计方案不断改进。通过某型号飞航导弹系统的优化实例,验证了算法的有效性。

There are two challenges facing the Complex Multidisciplinary Optimization Design, which include information exchange among coupled subsystems and complexity of system analysis. Based on Simultaneous Analysis and Design （SAND） algorithm and Concurrent Subspace Optimization （CSSO） algorithm, the use of Artificial Neural Network （ANN）based Response Surface （RS） approximations in Multidisciplinary Design Optimization （MDO） was proposed. It was a two-level optimization architecture, that is to say, the discipline level only satisfied the local constraints and the system level offered some collaborative mechanism to guarantee all of the discipline designs agreements on linking/coupled variables so as to facilitate improvements on design. A winged missile system was adopted as an example to verify the effectiveness of this algorithm.