基于遗传算法的参数投影寻踪回归及其在洪水预报中的应用

Application of parametric projection pursuit regression based on genetic algorithm in flood forecasting

针对传统非参数投影寻踪回归方法在应用时存在的问题 ,提出了三点改进措施 :(1 )采用实数编码的遗传算法代替高斯 牛顿算法优化投影方向 ;(2 )采用参数非线性Hermite多项式代替非参数逐段线性回归来拟合岭函数 ;(3)参数优化同时进行无需分组 ,逐个增加岭函数 ,不实行返回拟合的优选策略 .新的方法被称为参数投影寻踪回归 ,将其用于紫坪埔洪水预报 ,取得了满意的预报效果 ,而且所提出的方法在实际应用时 ,具有简单。

The traditional projection pursuit regression model has been improved in three aspects. The fist, genetic algorithm based on real coding is used to replace Gauss Newton method in optimizing projection direction. The second, nonlinear Hermit polynomial is used to approach the ridge function. The third, the method of constructive procedures is used to determine the numbers of ridge functions and all parameters without back fitting are optimized at the same time. The new model is applied to the flood prediction of Zipingpu Project and satisfied results are obtained.