两种Rosenbrock改进算法及其比较

Two Improved Rosenbrock Algorithms and Their Comparison

Rosenbrock旋转坐标法是一种有效的直接搜索方法,它分为坐标轮换和转轴两部分,可以追随着弯曲和陡峭的深谷前进。这种方法对于在搜索范围寻找近似极小点的位置是十分有效的。而在多极值问题中,选定的初始点对结果是有影响的,解的点序列只能收敛于局部极值,而不能在全局范围内找到最小点。笔者对Rosenbrock优化算法在多极值问题中避免局部最小方面提出了两种改进方法,并通过实验数据对该两种方法进行了比较。

Rosenbrock rotating coordinates method is a direct searching method which is effective.It can be divided into two parts which are alternating coordinates search and rotating coordinates.It can advance in the same direction as the flexural and sharp coulee.This method is very effective to find the approximative minimum in the searching area.But the chosen initial points have influence on the results while solve multi-extreme value problems.It can not find the global extremum and the points sequence of solution can only converge to local extremum.In this paper,two improved methods are proposed to avoid the local minimum in the multi extremum problem of the Rosenbrock optimization algorithm,and the two methods are compared by experimental data.