几类神经优化问题的研究-计算数学专业论文.docxVIP

万方数据 万方数据 Classified Index: TP183/44 U.D.C: 510 Dissertation for the Master Degree in Science THE RESEARCH FOR SEVERAL NEURODYNAMIC OPTIMIZATION PROBLEMS Candidate： Wu Guangxi Supervisor： Associate Prof. Fan Dejun Academic Degree Applied for： Master of Science Speciality： Computational Mathematics Affiliation： School of Science Date of Defence： June, 2014 Degree-Conferring-Institution： Harbin Institute of Technology 哈尔滨工业大学理学硕士学位论文 哈尔滨工业大学理学硕士学位论文 摘 要 众所周知，最优化问题在许多领域中占有举足轻重的地位，很多决策问题 都可以转化为相应的最优化问题。继传统的优化方法暴露出越来越多的缺点之 后，人们开始寻求新方法、新途径来更加高效地求解最优化问题。由于神经网 络具有大规模并行处理及快速收敛的特性，因此用其求解优化问题受到了越来 越多学者的青睐。同时，有关神经动力学优化的研究更是方兴未艾，各种重要 的研究成果不断涌现。基于神经动力学优化与非光滑分析等理论，本文将重点 研究两类凸优化问题。 首先研究了带有等式与不等式约束的非光滑凸优化问题。基于 Tikhonov 正 则化方法，构造单层神经网络来求解这类非光滑优化问题，并利用 Lyapunov 稳 定性定理证明神经网络的稳定性和渐近稳定性。与目前已知的求解该类优化问 题的神经网络相比，我们构造的神经网络具有结构简洁、不依赖惩罚参数等优 点。特别地，神经网络的稳定性证明取消了可行域有界性或目标函数强制性的 要求。 其次研究了带不等式约束的二次凸规划问题。基于 Karush-Kuhn-Tuker 条 件，构造梯度型神经网络，并证明神经网络的平衡点即为二次规划问题的最优 解。最后利用 LaSalle 不变集原理和 ?ojasiewicz 不等式证明所构造的神经网络 的渐近稳定性。该神经网络的稳定性不依赖于其他附加条件，因此具有更广阔 的应用前景。 关键词：非光滑最优化；二次规划；Tikhonov 正则化；神经网络 - I - Abstract It is well-known that optimization problems play an important role in many engineer areas. In practical life, many problems can be considered as the related optimization problems. With many defects of the traditional optimization methods exposed, people begin to search for better methods to solve the optimization problems. The neural network has the paralleled distribution computation in large-scale and fast convergence, therefore, more and more scholars observe the importance of neurodynamic optimization. At the same time, the research on neurodynamic optimization is in the ascendant, and all kinds of typical research achievements are springing up constantly. Based on neurodynamic optimization and nonsmooth analysis theory, two classes of convex optimizations are studied in this paper. Firstly, we study a class of nonsmooth optimization problem with equality and inequality constraints. Based on the Tikhon