摘要
针对工程多目标优化求解耗时且全局高精度代理模型难以构造的问题,提出基于序列径向基函数的优化方法。该方法在每个迭代步,运用信赖域更新技术将整个设计空间分割成一系列信赖域,以降低优化问题的复杂程度。在每个信赖域上建立各个响应的代理模型,并采用微型多目标遗传算法进行优化。为避免序列近似优化引起的效率降低问题,运用继承拉丁超立方试验设计继承上一代样本点,以减少新样本;同时继承当前的非支配解,对代理模型局部加密,以减少迭代次数。在Benchmark测试问题中,与传统的代理模型方法相比,在实际模型相同调用次数情况下,该方法能更好地逼近全局前沿面。最后将该方法应用于某轿车车身结构轻量化问题中,验证了解决实际工程多目标优化问题的能力。
In view of the problem of time-consuming solving and the difficulty in constructing global high-accuracy surrogate models in multi-objective engineering optimization, an optimization scheme based on sequential radial basis function is proposed. In each iteration with the scheme, the entire design space is divided into a series of trust regions by applying trust region update technique to reduce the complexity of optimization problem. Then the surrogate model for each response is built in each trust region and an optimization is conducted with micro multi-objective genetic algorithm. To avoid low efficiency caused by sequential process, the sample points of previous generation is inherited by using inherited Latin hypercube design of experiment for reducing new samples, while with the current non-dominate solution inherited, the mesh of some localities in the surrogate model is densified to reduce the number of iteration. In benchmark testing, compared with traditional surrogate model method, the scheme proposed can better approximate global frontier with the same times the model is called. Finally, the proposed scheme is successfully applied to the body structure lightweight optimization of a car, verifying its capability in solving the practical engineering problem of multi-objective optimization.
出处
《汽车工程》
EI
CSCD
北大核心
2015年第9期1077-1083,共7页
Automotive Engineering
关键词
车身结构
多目标优化
序列径向基函数
耐撞性
轻量化
car body structure
multi-objective optimization
sequential radial basis function
crash- worthiness
lightweighting
