Research On Individual Selection Mechanism In Differential Evolution

In various scientific and engineering fields, there are a lot of optimization problems. With the rapid development of the real world, many optimization problems become more and more complexity, and the traditional optimization algorithms have been difficult to meet the needs of various science and engineering. Therefore, to design optimization algorithms, which can effectively deal with highly complex problems, has profound scientific significance and great engineering application value. In this context, Evolutionary Algorithms(EAs) attract extensive attention and obtain great development, because of its advantages, such as self-adaptive, implicit parallelism and easy to use.As one of EAs, Differential Evolution(DE) solves the optimization problem by maintaining the evolution of a population. Since its simple structure, easy to use and robustness, DE has been successfully applied to various scientific and engineering fields,such as data mining, path planning, engineering design, and so on. However, at this stage, DE has still some shortcomings which roughly can be summarized into three aspects: 1) there are many strategies in the DE literature, however, how to choose the best one for a specific problem needs still an effective guidance; 2) the traditional DE is very sensitive to the control parameters, and there is still a lack of better solution for setting appropriate parameters; 3) DE is good at exploring the search space and locating the region of global minimum, but it is slow at the exploitation of the solution.Aiming at drawing these shortcomings, there are two aspects of DE are considered in this paper. On the one hand, DE is a typical swarm intelligence optimization algorithm, and its good optimization performance mainly depends on the interaction between individuals in the population. Furthermore, the individuals and methods of the interaction depend on how to use the population information. Therefore, how to extract and utilize the population information basically determines the optimization performance of DE. On the other hand, the mutation operator of DE plays an important role to lead the populaton to explore the search space, and how to design the individual selection mechanism directly affects the performance of DE.Based on these considerations mentioned above, the main objective of this paper is to extract the population information and use it to guide the individual selection in the mutation operator to improve the performance of DE. First, this study analyses the population information in DE and divids it into four major categories, namely topology information, fitness information, distance information and history information. Based on these population information, four individual selection mechanisms can be obtained.Then, for each individual selection mechanism, the advantages and disadvantages of the existing work is analyzed and a novel DE framework is proposed for effectively using the population information. Furthermore, its good performance is verified by extensive experiment. Finally, the comparison between the four DE frameworks has been done, and the effective reference information for scientific research and engineering application has also been provided.