基于非支配排序的遗传规划的建模方法

Genetic programming based on non-dominated sorting

分析了利用遗传规划进行复杂非线性系统建模中存在的过学习问题,提出了一个基于插值函数保护法和多目标非支配排序优化方法的遗传规划建模方法。文章利用NSGA所提出的非支配排序的思想结合传统的遗传规划来实现对于模型的精确度、复杂度和曲率的平衡。同时改进了传统遗传规划所使用的函数保护策略进一步降低了过学习现象,得到具有较高泛化能力和简洁性的最优模型。实验结果证明了该方法的有效性。

Based on the analysis of the overfitting phenomenon in genetic programming,a new protected approach based on interpolation for some mathematical functions is proposed.To get more accuracy model without losing generalization,a regularization term based on the curvature of a nonlinear model,the complexity of a model and the training errors should be considered.However,their effects on the training process are difficult to decide and estimate.Generally,the degrees of these effects are decided by the experience or the knowledge of the data.Here,a method of non-dominated sorting is used in GP,and then their effects can be selected more impersonally.Experiments show the effect of the method.