基于多目标优化NSGA2改进算法的结构动力学模型确认

传统结构动力学模型确认方法通常采用单目标优化,存在精度不足和稳定性差等缺点,难以满足实际工程需求。基于此,提出一种采用神经网络作为代理模型,建立以马氏距离和鲁棒性为不确定性量化指标的多目标优化模型,并将NSGA2多目标进化算法用于求解。针对NSGA2存在无法有效识别伪非支配解、计算效率低和解集质量较差等设计缺陷,提出一种基于支配强度的NSGA2改进算法INSGA2-DS。INSGA2-DS将支配强度引入非支配排序,采用新型拥挤距离公式和自适应精英保留策略,以提高收敛效率和解集质量。GARTEUR飞机算例的仿真结果表明,INSGA2-DS求解复杂工程问题时具有更好的收敛性和分布性,而考虑鲁棒性的结构动力学模型确认方法可以获得同时满足多种目标要求的Pareto解集,提高了模型确认的精度和稳定性。

Structural dynamics model validation based on NSGA2 improved algorithm

The traditional structural dynamics model validation methods usually use single-objective optimization.Due to poor accuracy and stability,it is difficult to meet the actual engineering needs.This paper uses neural network as agent model,and establishes multi-objective optimization model with Mahalanobis distance and robustness as optimization targets,which is solved by NSGA2.Since NSGA2 has some design defects,such as ineffectiveness in identifying pseudo non-dominant individuals,low efficiency,poor convergence and distribution,this paper proposes an improved NSGA2 algorithm based on dominant strength (INSGA2-DS).INSGA2-DS introduces dominant strength to non-dominated sorting method,and adopts a new crowding distance formula and the adaptive elitist retention strategy to improve the convergence efficiency and Pareto solution quality.The simulation results of GARTEUR airplane show that INSGA2-DS has better convergence and distribution when solving complex engineering problems.The structural dynamics model validation method considering robustness can provide a variety of Pareto solution sets which satisfy different target requirements,and improve the accuracy and stability of model validation.