结合机理先验约束的Kriging模型参数优化方法研究

摘要

Since traditional optimizing methods of Kriging model are easy to result in local optimum, a new method was proposed considering both evolutionary algorithm and priori knowledge. Monotonicity and concave-convexity priori knowledge were used as the constraints in the optimization process. With a reasonable number of samples, the Kriging model becomes more consistent with priori knowledge, which can improve prediction accuracy. Finally, a function simulation experiment and a forecasting model of the average particle size of ZrO2-TiO2 colloidal sols were used to demonstrate the effectiveness of the proposed method.%基于过程先验知识和Kriging模型,针对Kriging模型在求解参数过程中容易陷入局部最优的缺陷,提出一种利用进化优化算法求解模型参数的方法.在参数求解过程中,将过程变量对应的输出函数的单调性和凹凸性先验知识作为约束条件,在合理的样本数目下,得到更符合先验知识的Kriging模型,提高模型的预测精度.通过函数仿真实验和复合纳滤膜的溶胶粒径的预测估计验证了此建模方法的有效性.