一、迭代终止条件
1.椭球法 终止条件:
Xk为当前迭代点,x*是优化问题的最优解,f是目标函数
2.multi-start Levenberg-Marquardt终止条件:
3.multi-start sequential quadratic programming
存在
4.单纯形法能在有限步内找到最优解,因此没有停止条件
5.multi-start Gauss-Newton least-squares algorithm无约束最小非线性二乘问题
,优化问题的形式为
e代表错误函数
6.multi-start penalty function method + line search method with Powell directions and line minimization using parabolic interpolation
7.multi-start Lagrange method + Levenberg-Marquardt algorithm
h is the equality constraint function (when the constraint is written in the form h(x) = 0), andε1, ε2 are small positive numbers. where f and h are respectively the objective function and the equality constraint function of the original constrained problem (with the equality constraint written in the form h(x= 0)
8.改进单纯形法能在有限步找到最优解
a multi-start penalty function method with the Nelder-Mead algorithm (M7)
10.切平面法终止条件
11.interior point algorithm
二、判断凸函数:
(1)凸函数之和为凸函数
(2)存在非凸函数的函数之和就是非凸函数
(3)仿射函数的凸函数是凸函数
-log(x)是凸函数,
是凸函数,
是凸函数
(4)
是凸函数
三、非凸函数使用multi-start method:multi-start Levenberg-Marquardt
multi-start sequential quadratic programming
multi-start Gauss-Newton least-squares algorithm
两个函数相加,凸函数的这个公式取小于等于号,线性函数的这个公式取等号