摘要
本文利用曲线线性搜索法和最优化的微分梯度法的特点,提出了一种一般的曲线搜索方式:微分下降法。这种方法通过下降方向对确定迭代矩阵,由初值微分方程的解析解确定迭代搜索曲线。本文给出了算法的整体收敛性证明,并给出了满意的数值实验结果。
In this paper a general differential descent algorithm along curvilinear search path,which is based on curvilinear search method and differential gradient method,is proposed for soling the unconstrined minimization problem. In this method,an iterative matrix is generated by a pair of descent directi- onal vectors,and the iterative search paths are derived by the solution of differential epuation with the initial conditions.The convergence of the algo- rithm is proved.The numerical tests of the algoriahm is obtained,and it indicates that the convergence is very rapid.
