Improved Genetic Algorithms And Applications In Engineering Optimization

Evolutionary computation (EC), as a novel and powerful intelligent optimization technology, has been utilized extensively in almost every branch in engineering science. It has more advantages over traditional optimization methods when solving problems with global search, complex design domain and complicated target functions, and it is easier to use. Genetic algorithm (GA) is one of the most important algorithms in evolutionary computation. An extensive study of improved genetic algorithms in the context of engineering optimization design has been conducted in this dissertation.The dissertation is organized as following chapters.First, a general introduction to genetic algorithms is presented in chapter 1, and the past and recent developments in this field are briefly described. The framework of the dissertation is also figured out in the end of chapter 1.Next, theoretical aspects and implementation of genetic algorithms are focused on in chapter 2. The basic workflow of genetic algorithms is described at the beginning of this chapter, and then typical representation and genetic operators are discussed. Finally, the search mechanism and convergence are investigated in the end of the chapter.Next, a novel genetic algorithm called subdomain based genetic algorithms (SBGA) is proposed in chapter 3. In SBGA, design space is divided into many small and isolate subdomains, and distribution information in these subdomains will be traced in the process of evolution, which will be used to guide subsequent search. At the same time, SBGA also provides a new method to handle complicated constraints. The numerical results demonstrate that SBGA is effective and efficient to alleviate premature convergence.Next, GA based topology optimization methods of continuum structures are studied in chapter 4. A new compact representation called four direction chain code representation is proposed in this chapter, and case-based genetic algorithms enlightened by machine learning are developed, in which a new concept named "Target Vector" is utilized to calculate the distance between two structures withdifferent topology. The expected results are obtained when applying above approach to continuum beam and bike framework topology optimization.Next, practical engineering optimum problems are of more than one target functions and computational cost is very high, so multiobjective optimization methods are discussed in chapter 5. To improve computing performance, a parallel version of improved strength evolutionary algorithms (SPEA2) is presented to optimize beam topology with two conflicting targets on the environment of cluster of workstations connected with each other.Next, a hybrid genetic algorithm is proposed which combine niche and local search techniques in chapter 6. "Mutual Information" is introduced to calculate the matching degree of two medical images from different modalities and very good results are obtained.Finally, a summary of the research conclusions, a list of innovation points and a discussion on the most promising paths of future research are also presented in chapter 7.