基于Kriging模型的多点加点准则和并行代理优化算法

针对并行仿真环境下复杂工程系统的优化设计问题,提出一种基于Kriging模型、多目标策略和聚类方法的并行代理优化算法.该算法的多点加点准则,以同时优化期望改进准则和可行性概率准则为目标,首先生成兼具目标响应改进和可行域边界刻画功能的备选试验点集;再利用聚类方法从备选点集中选取多个有代表性的新试验点.通过两个数值算例和一个工程算例,将所提并行优化算法与已有算法做比较,结果表明所提算法具有更高的优化精度、效率和稳健性.

A multi-points infill sampling criterion and parallel surrogate-based optimization algorithm based on Kriging model

To solve the design optimization problems of complex engineering systems in parallel simulation environment, a parallel surrogate-based optimization algorithm is proposed based on Kriging model, multi-objective strategy and cluster analysis method. The multi-points infill sampling criterion of the proposed algorithm, aims to optimize the expected improvement and the probability of feasibility simultaneously, so as to generate a candidate trials set in which the trials have the ability to balance exploration of the optimal solution vs. exploitation of the feasible region boundaries. Then, clustering method is adopted to select multiple representative new trials from the candidate trials set. In the end, the proposed algorithm is tested on two numerical and one engineering benchmarks and is compared with the existed algorithms. The numerical results indicate that the proposed algorithm is more accurate, efficient and robust.