基于动力学演化样本的代理模型优化方法

Surrogate model optimization method based on dynamic evolution sample

为提高代理模型的精度,将动力学演化样本生成方法引入代理模型优化设计。采用拉丁超立方抽样得到初始样本点集,基于动力学演化计算生成均匀性更佳的样本点集,并用于Kriging代理模型构造,结合遗传算法给出一套基于动力学演化的代理模型优化方法。通过经典测试函数和某实际机床横梁的优化分析,证明演化后点集在代理模型构建和模型全局寻优上的优越性。

To improve the accuracy of the surrogate model, the dynamic evolution sample generation method is introduced into surrogate model optimization design. The initial sample point set is obtained by Latin hypercube sampling, and the sample point set with better uniformity is generated based on the dynamic evolution calculation, which is used to construct Kriging surrogate model. Combined with genetic algorithm, an optimization method of surrogate model based on the dynamic evolution is proposed. The superiority of the evolved point set in the surrogate model construction and the model global optimization is proved by the classical test functions and the optimization analysis of an actual machine tool beam.