基于多级代理模型的优化算法

Simulation Optimization Based on Multilevel-Surrogate Models

在仿真优化中高精度的仿真模型大多难以实现,本文提出一种由全局和局部代理模型共同作用的多级代理模型,并与仿真优化相结合提出基于多级代理模型的仿真优化算法。运用Kriging近似理论和RBF神经网络分别构建全局代理模型和局部代理模型,并在仿真优化的过程中在线更新代理模型。通过算例对算法进行验证,结果表明多级代理模型具有良好的逼近能力,基于多级代理模型的仿真优化方法具有良好的鲁棒性和寻优性能。

The high precision simulated models are difficult to be actualized during simulation optimization. In this paper, a new multilevel-surrogate model constituted by global and local surrogate model is proposed, and a new simulation optimization based on multilevel-surrogate model is given. The global and the local surrogate models by using Kriglng model and RBF neural network are constructed and are renewed in time during the optimization process. The computation results indicate that multi-level surrogate model has the capacity of good approximations, and the simulation optimization based on multilevel-surrogate models has the performances of good robustness and optimization.