基于信赖域技术和修正拟牛顿方程的非单调超记忆梯度算法  

A non-monotone super-memory gradient method based on trust region technique and modified quasi-Newton equation

作　　者：宫恩龙[1] 陈双双 孙清滢 陈颖梅 

机构地区：[1]青岛酒店管理职业技术学院,山东青岛266100 [2]中国石油大学理学院,山东青岛266580

出　　处：《中国石油大学学报（自然科学版）》2013年第2期191-196,共6页Journal of China University of Petroleum(Edition of Natural Science)

基　　金：国家自然科学基金项目(61201455);中央高校基本科研业务费专项(10CX04044A;11 CX06087A)

摘　　要：基于信赖域技术和修正拟牛顿方程,结合Neng-Zhu Gu非单调策略,设计新的求解无约束最优化问题的非单调超记忆梯度算法,分析算法的收敛性和收敛速度。新算法每次迭代节约了矩阵的存储量和计算量,算法稳定,适于求解大规模问题。数值试验结果表明新算法是有效的。Based on trust region technique and modified quasi-Newton equation, by combining with Neng-Zhu Gu non-monotone strategy, a new super-memory gradient method for unconstrained optimization problem was presented. The global and convergence properties of the new method were proved. It saves the storage and computation of some matrixes in its iteration, and is suitable for solving large scale optimization problems. The numerical results show that the new method is effective.

关 键 词：超记忆梯度算法 非单调规则 收敛性 收敛速度 数值试验 

分 类 号：O221.2[理学—运筹学与控制论]