（毕业设计论文）《无约束优化问题算法初探》.doc

PAGE JIU JIANG UNIVERSITY 毕 业 论 文 题 目 无约束优化问题算法初探 英文题目 Unconstrained Optimization Algorithm is Discussed 院 系 理学院 专 业 数学与应用数学 姓 名 班 级 指导教师 二零一二年三月 摘要 对于无约束最优化问题可表述为，（1）其中f(x)是连续可微函数。 无约束条件下多变量函数的寻优方法大致分成两类。一类在迭代过程中仅用到函数值,不必计算函数的导数,这类方法称为直接法或搜索法。一般说来,直接法的收敛速度较慢,只是在变量较少时才适用。但迭代法的迭代步骤简单,特别是当目标函数的解析表达式十分复杂,甚至写不出具体表达式时,他们的导数很难求得,或根本不存在，这时,就只有用直接法了。另一类是在计算过程中要用到函数的一阶导数和(或)二阶导数,这时，在(1)中总假定f(x)有一阶或二阶连续偏导数.由于用到函数的解析性质,故称为解析法。 【关键词】关键词 无约束优化 牛顿迭代 信赖域算法 梯度法 Abstract For unconstrained optimization problem can be expressed as, (1) which f (x) is continuously differentiable functions. No constraint conditions of multivariate function optimization method roughly divided into two categories. A class in the iteration process only use function values, need not calculation function of derivatives, this kind of method is called direct method or search method. Generally speaking, the direct method of convergence speed is slow, just in the variable is small to apply. But iterative method of simple iterative procedure, especially when the analytical expression of the objective function is very complex, and even could not write a concrete expression, their derivative is difficult to be calculated, or simply did not exist, then, is only in the direct method. Another kind is in the process of calculation to use a function of derivatives and (or) second order derivative, at this moment, in (1) will always assume that f (x) have a one order or two order continuous partial derivative. Because of the use function analytic properties so called the analytic method. [Keywords] Unconstrained optimization, Newton iterative, Trust region algorithm，Gradient method 目 录 引 言1 第一章 直接法2 第二章 牛顿法及其修正算法3 2.1牛顿法3 2.1.1原始牛顿法4 2.1.2阻尼牛顿法6 2.1.3修正牛顿法7 2.2拟牛顿法8 2.3牛顿-拉弗森方法9 第三章 梯度法（最速下降法）12 3.1共轭梯