基于半定规划问题的算法及其应用研究-运筹学与控制论专业论文.docxVIP

摘 要 万方数据 万方数据 摘 要 半定规划是线性规划的一种推广, 它是在满足约束“对称矩陈的仿射组合半正 定”的条件下使线性函数极大(极小)化的问题. 由于半定规划统一了许多类型的问题 (如: 二次规划和线性规划), 并在组合优化, 系统与控制理论, 电子工程等领域中得 到了广泛的应用, 半定规划备受人们的关注. 本文分别对线性半定规划和非线性半定 规划进行了讨论, 具体如下: 第一, 针对线性半定规划问题, 提出了一个具有二阶收敛性的光滑化牛顿法. 此 算法把带扰动量的F-B光滑函数应用到半定规划KKT方程组中, 避免了扰动量为零时 的不可微现象, 无需正定性假设和非退化条件, 获得了全局收敛性和二阶收敛性. 第二, 针对非线性半定规划问题, 提出了一个具有全局收敛性的滤子法. 该算法 通过减小违反约束度函数值或预估目标函数值来确定试探步是否被滤子接受, 并能 通过改变步长来实现充分下降性, 无需减小信赖域半径, 获得算法的全局收敛性. 最后, 针对上述算法进行了数值实验, 并将半定规划模型应用于无线传感网络定 位问题中, 数值仿真及应用结果表明所设计算法是可行有效的. 关键词: 半定规划; 牛顿法; 滤子; 全局收敛; 二阶收敛; 无线传感网络 -I- Abstract Abstract Semidefinite programming (SDP) is an extension of linear programming, which maximizing(minimizing) a linear function subject to the constraint that an affine combination of symmetric positive semidefinite matrices. Since SDP unifies many types of problems (such as, linear and quadratic programming) and has been found many applications in combinatorial optimization, systems and control theory, electronic engineering etc, which has attracted people’s attention. This paper discusses effective algorithms for linear SDP and nonlinear SDP, the achievements can be summarized into following two aspects. Firstly, we propose a linear SDP smoothing Newton algorithm with quadratic convergence for solving linear SDP. In this algorithm, Fischer-Burmeister complementary function with disturbance is employed in the complementary function of SDP KKT conditions, non-differentiability can be avoided as disturbance tends to zero. Global convergence and quadratic convergence of the algorithm are proved without positive definiteness assumption and nondegeneracy condition. Secondly, we propose a sequential quadratic semidefinite programming (QSD)-filter algorithm for solving nonlinear SDP. In this algorithm, a filter is accepted by reducing constraint violation value or predictive function value, and sufficient descent direction is ensured via changing the step size. Global convergence is obtained of the algorithm is proved