9无约束优化法分析报告.ppt

第二章无约束优化解析法 School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU 牛顿法的基本结构（解析搜索法） 牛顿法基本原理:运用目标函数的一阶、二阶导数，构造在近似点的二阶Taylor展开式，逼近目标函数。运用逼近函数在某点的梯度为零，可以得到新的近似点。重复上述步骤，直到近似点满足精度要求。 School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU * School of Economic Management , NCEPU School of Economic Management , NCEPU 无约束极小化方法 直接搜索法 （不需要求函数的导数） 解析搜索法 最速下降法 共轭梯度法 牛顿法 阻尼牛顿法 变尺度法 …… 坐标轮换法 模式搜索法（步长加速法） 方向加速法（Powell法） 单纯形搜索法 …… 2.1 引 言 School of Economic Management , NCEPU 2.2 最速下降法(梯度法) 最速下降法是在迭代中选择负梯度方向作为搜索方向而形成的一种下降迭代算法。 最速下降法的理论依据： 定理：设函数 在某点可微，则负梯度方向 是函数在该点的最速（快）下降方向。 School of Economic Management , NCEPU 最速下降法的基本步骤（解析搜索法） 初始点 精度检查 Y 停止迭代 最速下降法基本原理: 从初始点出发，沿着函数负梯度方向进行搜索，得到一个新的近似极小点，使目标函数值减小。重复上述步骤，直到近似点满足精度要求。 N 搜索规则:目标函数的负梯度方向为搜索方向,按最优步长确定 School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Management , NCEPU School of Economic Mana