神经网络辅助的多目标粒子群优化算法在复杂产品设计中的应用  

Application of ANN Assisted MOPSO in Optimization Design of Complex Products

作　　者：冯国奇[1] 崔东亮 张琦[3] 代学武 FENG Guoqi;CUI Dongliang;ZHANG Qi;DAI Xuewu(School of Business Administration,Northeastern University,Shenyang 110819,China;State Key Laboratory of Synthetical Automation for Process Industries,Shenyang 110819,China;Signal and Communication Research Institute,China Academy of Railway Sciences,Beijing 100081,China)

机构地区：[1]东北大学工商管理学院,沈阳110819 [2]流程工业综合自动化国家重点实验室,沈阳110819 [3]中国铁道科学研究院通信信号研究所,北京100081

出　　处：《系统管理学报》2019年第4期687-696,707,共11页Journal of Systems & Management

基　　金：国家自然科学基金资助项目(U1834211,61790574);国家铁路智能运输系统工程技术研究中心开放课题(RITS2018KF03)

摘　　要：复杂产品有限元分析(Finite Element Analysis,FEA)费用很高,给多目标优化(Multi-ObjectiveOptimization,MOO)带来很大困难.提出一种人工神经网络(Artificial Neural Network,ANN)辅助的多目标粒子群优化算法(Multi-Objective Particle Swarm Optimization,MOPSO)处理这类计算密集的设计问题:以基于噪声的虚拟样本丰富ANN的训练样本集,通过虚拟样本的控制参数和ANN模型参数的协同优化提高ANN泛化能力;以此ANN为代理模型支持多目标粒子群算法的进化,并采用基于网格邻域信息的拥挤指标提高Pareto前沿的收敛性、多样性及均匀性.最后,以航空发动机高压涡轮盘(High Pressure Turbine Disc,HPTD)多目标优化案例验证该策略的有效性和可用性.试验证明,这种面向成本的MOO方法降低了复杂产品多目标优化的工程应用难度,提高了设计质量.Computational simulation of finite element analysis (FEA) for complex products is highly timeconsuming , making it a great challenge to find a multi-objective optimization (MOO) solution. An artificial neural network (ANN) assisted multi-objective particle swarm optimization (MOPSO) was proposed in this paper to solve this computing intensive problem. The proposed approach has the following features. The size of training set for ANN is increased by noise-based virtual samples. The control parameters of virtual samples and neural model were optimized together to improve the generalization ability of ANN that is used as the fitness function of the MOPSO. The diversity, uniformity, and convergence of the proposed MOPSO are improved by exploiting the information of neighborhood grids. Finally, the feasibility and effectiveness of the MOPSO are validated by the MOO of high pressure turbine disc. The experiment shows that the engineering application difficulty of MOO is mitigated by this computation-efficient method.

关 键 词：复杂产品 小样本数据 人工神经网络 多目标粒子群优化算法 

分 类 号：N945.15[自然科学总论—系统科学]