【作者】 王洪欣；

【导师】 于兰峰；

【作者基本信息】 西南交通大学 ， 机械设计及理论， 2008， 硕士

【摘要】 由于多目标优化技术在工程、经济、管理和军事等领域中具有重要的应用价值,多目标优化的研究越来越受到广泛的关注和重视,它已发展成为一门新兴的学科并在应用中显示出强大的生命力。遗传算法是借鉴生物的自然选择和遗传机制而开发出的一种全局优化自适应概率搜索算法,它在解决复杂系统优化时所表现出的独特的优越性和健壮性,使其成为解决多目标优化问题的一个非常有效的手段。由于龙门起重机结构系统的动态特性很难用设计变量精确显式表达,本文利用多层人工神经网络极强的非线性映射功能,来描述和处理动态系统中设计变量及其动态参数之间的关系。人工神经网络模型一旦建立,可取代有限元模型进行结构动态特性重分析,其分析过程简单而直接,且远比有限元模型计算速度快,尤其适用于工程技术人员使用。因此,利用遗传算法对所建立的神经网络模型寻优,可以得到可行区域内动态特性最优时的设计变量及目标值。本文在有限元分析的基础上,结合正交试验法,利用BP神经网络建立了振动系统快速重分析的数学模型,并利用遗传算法对神经网络模型寻优,最终得到多目标优化的部分Pareto最优解。更多还原

【Abstract】 In view of the importance of multi-objective optimization in engineering, economy, management, military and so on, the research on multi-objective optimization has been paid more attention. It has developed into a new branch of science and demonstrated powerful vitality in application. The genetic algorithm is a global optimization, auto-adapted, probability-based searching algorithm which uses the experience of biological natural selection and genetic mechanism for reference. Owing to its unique superiority and robustness in solving the complex system optimization, it becomes a very effective method in solving multi-objective optimization problems.The dynamic behaviour of gantry container crane structure system is very difficult to be explicitly expressed with design variables. As a result of its powerful nonlinear mapping ability, multi-layer neural networks are used to describe and deal with the relations between design variables and dynamic parameters of the structure system. Once the neural networks model is built, it can substitute the finite element model, and be used for the reanalysis of structure dynamic behaviour. The analysis process is simple and direct. Moreover, computation speed of the neural networks model is faster than the finite element model, which applies to engineers and technicians, especially. Therefore, genetic algorithm is used to optimum the built neural networks model, to get the design variables and target values in the feasible zone when the dynamic behaviour is optimal.In this paper, BP neural networks are employed in combination with finite element analysis and orthogonal experiment method, to build the mathematical models of the vibration system for a rapid re-analysis. And genetic algorithm is used to optimum the neural networks model. Eventually, we get part of the Pareto optimal solutions.更多还原