Research And Design Of Evolutionary Algorithms

There are a lot of optimization problems in our life and production management process. In recent years, intelligent computing for solving optimization problems has become a hotspot of international academia, where evolutionary computation is a bright prospect with many important areas. Evolutionary Computation has made remarkable development in theoretical analysis and industrial applications. It has been the first biological field it relates to the calculation of the development of various types of natural computing algorithms and techniques, including evolutionary computation, neural computing, computing ecology, social and economic calculation, and so on.In this paper, we made an intensive exploration on evolutionary algorithms (EA) proposed a single objective optimization algorithm and a multi-objective optimization algorithm. Specifically the study of this paper includes the following:In terms of single objective optimization, we propose a single objective optimization evolutionary algorithm based on Covariance Matrix Learning and Searching Preference (CMLSP) and design a switching method which is used to combine CMLSP and Cov-ariance Matrix Adaptation Evolution Strategy (CMAES) and named CMLSP/AES.The basic aim of CM-LSP/AES is to devote more resources in the vicinity of a good solution to search a better solution.To achieve this, we design of a covariance matrix of learning-based approach to produce higher quality solutions instead of the traditional evolutionary algorithm hybrid variation mathod. The design of CM-LSP/AES is consists of two parts. Firstly, the search preferences design is based on Covariance Matrix Learning and Gaussian distribution. Secondly, employing a switch framework to combine CMLSP and CMAES. To evaluate performance of the proposed algorithm we compare the CM-LSP/AES with the classical evolutionary algorithm CMAES and CoBiDE on sixteen test instances. The simulation results show that CM-LSP/AES is an effective solutions to complex problems, including most of the black box.In the multi-objective optimization, we proposed an evolutionary many-objective optimization algorithm based on population decomposition and reference distance, named EAPD-RD. The algorithm is designed mainly to solve the dimension disaster and population diversity maintaining low efficiency problems in the high dimension many-objective optimization. EAPD-RD employed the populations decomposition technology to ensure the diversity of population and reducing the amount of computation in evolution. And then, using the distance information from population and the Pareto dominance relation to judge the better non-dominated solutions. Finally, niching technique is used to measure the degree of congestion between individuals selecting the better solutions to ensure the diversity. Similarly, EAPD-RD compared with other EAs, such as MOEA/D, NSGA-III and GrEA by six test problems in IGD-metric and HV-metric. That test results show that EAPD-RD has certain advantages in handling high dimension multi-objective optimization problem.