Integrated Research For Isoinorphic Differential Evolution

AS an emerging computing model of evolutionary computation, differential evolu-tion(DE) has achieved significant performance in optimization domain and subsequent CEC competitions. However,several aspects of DE still need to be improved for solving complex optimization problems. Generally speaking, there are three demerits of DE as follows:(1) The ability to solve diverse problems of DE,is still relatively poor;(2) The much mutation strategies in DE family have extremely different performance. Unfor-tunately, we have observed that these mutation strategies have not yet systematically exploited in DE algorithm design;(3)Because that most integrated frameworks depend on special optimizer and have high coupling degree,it's hard to achieve the goal of the transplantation and mutual operation of framework reuse completely.For the first two drawbacks, we design more robust and effective DE by integrating differential mutation strategies. For the last one drawback,we construct a general and flexible integrated framework shared with others.The main contributions of which can be summarized as follows.1. The theory of integrated evolution is proposed. We start the original motivation of integrated evolution and then give the formal definition of integrated evolution, de-lineate the difference between the integrated evolution with other similarity concept.We propose the integrated framework, describe the classification and property of integrated evolution.2. For solving control parameters selected problem of DE, we introduce a self-adaptive scaling factor parameter F for each individual by using chaos.Experimental results on different benchmark functions show that the new algorithm is superior to the other related algorithms.3.To combine the mutation schemes "DE/rand/1" and "DE/current-to-best", we propose a new hybrid mutation scheme "elite-to-rand" that utilizes an explorative and an exploitative mutation operator, with an objective of balancing their effects. The experiments are designed to determine whether there was any benefit to the use of the novel mutation scheme as opposed to the other two DE-variants.4. To solve the problem of slow convergence speed before reaching the global opti-mum in the conventional differential evolution (DE), an effective approach, called elite opposition-based learning, is proposed, in which the generalized opposition-based learn-ing strategy is improved by the elite members. A novel differential evolution algorithm (EODE) which integrated the elite opposition-based learning strategy is presented in this paper.In the proposed algorithm, it use a novel process-divided framework proposed in the thesis which we divide the optimization process into2stages. It is also proven that the proposed algorithm can guarantee the convergence towards the global optimum. Experimental results show that the elite opposition-based learning strategy has much better search performance than the generalized opposition-based learning strategy and the novel EODE algorithms can obtain better efficiency.5. We propose a multi-population differential evolution which integrate three mutation schemes(MMDE).The improvements in MMDE mainly focus on three as-pects.(1)We apply the population-divided integrated framework which divide the pop-ulation into different sub-populations according to the indices of the individuals in the population;(2)We choose three mutation schemes and three control parameter settings to balance exploration versus exploitation:(3)Subpopulations are interconnected accord-ing to different communication topologies and can exchange information periodically by migrating individuals from one subpopulation to another.6.We propose a multi-space multi-mutation strategy DE(MMDE) based on the space-divided integrated framework. The proposed MMDE algorithm divides the whole search space into multiple subareas according to the fitness of the individuals, each of which may cover one or a small number of local optima, and then separately search within these subspaces by differential strategies.Extensive experiments have been con-ducted in the thesis to compare it with other algorithms.