【作者】 王亮；

【导师】 杨茂林；

【作者基本信息】 华中科技大学 ， 计算机技术， 2020， 硕士

【摘要】 随着科学技术的发展,工程领域中涉及到的一些函数优化问题越来越复杂,它们往往具有不可导、不连续、多峰值等特点,传统的数学方法已经很难得到理想的结果。函数最优化问题可以通过演化算法进行求解。其中,差分演化算法已经被证明是最强大的演化算法之一。然而,差分演化算法中一些参数的设置对性能有非常大的影响。为了解决这个缺陷,研究者们已经提出了很多不同的参数适应技术。不同的参数适应技术有着不同的特点,各自适用于不同类型的函数优化问题。差分演化算法的集成框架可以将差分演化算法不同的参数适应技术的联合起来,从而提高算法原本的性能。在框架中,整个种群被分成若干个子种群,每个子种群使用不同的差分演化算法的变体,它们有各自的参数适应技术。在演化过程中各个子种群的演化互不影响,只有产生变异个体的时候从整个种群进行选择以便交换信息。作为一个例子,将两个已经存在的参数适应技术的差分演化算法应用到框架中,进行了相关实验。通过对旅行商问题以及国际上通用的一个函数测试集的测试,最终的实验结果表明,通过使用提出的框架,集成起来的差分演化算法比之前各自单独的算法在稳定性和收敛性上均有了一定的提升,并且可以解决更多类型的函数优化问题。更多还原

【Abstract】 With the development of science and technology,some function optimization problems involved in the field of engineering are becoming more and more complicated.They often have the characteristics of unguided,discontinuous,and multi-peak.Traditional mathematical methods have been difficult to obtain ideal results.For the function optimization problem,it can be solved by evolutionary algorithm.Among them,the differential evolution algorithm has proved to be one of the most powerful evolution algorithms.However,the setting of some parameters in the differential evolution algorithm has a very large impact on performance.In order to solve this defect,researchers have proposed many different parameter adaptation techniques.Different parameter adaptation techniques have different characteristics,and they are suitable for different types of function optimization problems.The integrated framework of the differential evolution algorithm can combine different parameter adaptation techniques of the differential evolution algorithm,thereby improving the original performance of the algorithm.In the framework,the entire population is divided into several subpopulations,and each subpopulation uses different variants of the differential evolution algorithm,and they have their own parameter adaptation techniques.During the evolution process,the evolution of each sub-population does not affect each other.Only when mutated individuals are generated,they are selected from the entire population in order to exchange information.As an example,the differential evolution algorithm of two existing parameter adaptation techniques was applied to the framework,and related experiments were conducted.Through the test of TSP(Travelling salesman problem)and a function test set commonly used internationally,the final experimental results show that by using the proposed framework,the integrated differential evolution algorithm is more stable and convergent than the previous separate algorithms.A certain improvement,and can solve more types of function optimization problems.更多还原