基于自适应代理模型的序列迭代优化

The Application Sequence Iterative Optimization Based on Adaptive Surrogate Model

针对现有较为成熟的代理模型通用性差的问题,提出自适应代理模型方法,该方法能够根据优化问题自适应地选择出性能最优的代理模型.对某一优化问题分别构建不同类型的代理模型,然后对其进行性能测试,挑选性能较好的模型并进一步组合,从而提高模型性能.在此基础上进行序列迭代优化,很好地解决了传统的序列迭代优化方法无法在优化过程中实时更新模型类型的问题.通过标准测试函数以及活塞优化仿真实例分析表明:较之传统优化和基于单一代理模型的序列迭代优化,此方法能在保证求解精度的情况下拥有更高的优化效率.在活塞实例分析中,调用仿真模型的次数仅为传统优化的46%.

Aiming at the poor versatility of the existing more mature surrogate models,an adaptive surrogate model method was proposed,which could adaptively select the surrogate model with the best performance according to the optimization problem.Aiming at a certain optimization problem,different types of surrogate models were constructed,and their performance was tested.The models with better performance were selected and further combined to improve the model performance.Carrying out sequence iterative optimization on this basis could well solve the problem that the traditional sequence iterative optimization method cannot update the model type in real time during the optimization process.The analysis of standard test functions and piston optimization simulation examples shows that compared with traditional optimization and sequential iterative optimization based on a single a surrogate model,this method can have higher optimization efficiency while ensuring the accuracy of the solution.In the piston case analysis,the number of calls to the simulation model is only 46%of the traditional optimization.