Font Size: a A A

Study Of Multi-objective Optimization Based On Evolutionary Algorithm And The Applications

Posted on:2006-07-08Degree:DoctorType:Dissertation
Country:ChinaCandidate:Z Z WangFull Text:PDF
GTID:1118360155961200Subject:Computer application technology
Abstract/Summary:PDF Full Text Request
In real world problems, one is faced with the problem, which often is composed of multiple, possibly competing, goals, while being optimized simultaneously, that is, the problem is a multi-objective optimization problem. Although the approaches for solving Multi-objective Optimization Problems (MOPs) have been available for many years, Evolutionary Algorithms (EAs) have been developed to be ideal techniques for solving MOPs, especially for very large scale ones. Therefore MOPs also become focus in the domain of EAs. Thereby Multi-objective Optimization Evolutionary Algorithms (MOEAs) are widely used such as follows in practice: multi-objective decision, transportation and portfolio problems for solving investment securities, inflation, economic growth in economics and management; multi-objective location, assignment, design and traffic problems in engineering design; multi-cast and Geo-cast problems for designing framework topological framework in network and communication. With the development of researches and applications, some complex problems in real world demand enhancing performance of MOEAs. Therefore the problems such as how to farther improve EAs, and in MOPs how to combine effectively local search strategies with optimization techniques for ultimately improving quality of solution, should be central in this work. The research in this work extends techniques used in MOEAs. Specifically this thesis is concentrated on the following areas:1. Diversity of population in GA has an important impact on GA's convergence and other performances. Some evolutionary properties, diversity and other factors, which influence GA's performances, are analyzed. An important approach for improving an EA's performance—EA based on Hybrid optimization strategy (HEA), and its feasibility and effective mechanism are analyzed. And then in the context of Vehicle Routing Problems (VRP), combined with local search strategy—2-optimal, an algorithm GAwith2-opt is proposed to solve VRP, and chromosome representation, genetic operators are discussed. Experiment shows GAwith2-opt is well performed.2. In contribution to solving goal of MOEAs, when constructing a MOEA, some techniques and strategies, used in fitness assignment, selection and genetic operations, are analyzed in details. Based on the mechanism of a Hybrid MOEA(HMOEA) and its frame, a HMOEA with enhancing convergence performance—HJMOEA, is proposed. HJMOEA combines MOEAs with a classical local search technique, Hooke and Jeeves search method. The experiment demonstrates effectiveness, feasibility, convergence performance of HJMOEA.3. Crossover is of importance to EAs, especially GA. Although theoretically there is not systematic analysis about its impact to GA, but from practical applications and experiences, crossover plays a important role in evolutionary process of GA running. Some properties about crossover in GA are discussed. Based on effective crossovers in GA and specific techniques used in the noted MOEAs, an algorithm with well performances, MOEADC (multi-objective evolutionary algorithm based on double crossovers) is proposed, where two kinds of crossovers are used effectively in a MOEA. Its convergence performance, solutions distribution and computing efficiency are analyzed and tested, and by proof it is convergent.4. According to the practical activities in engineering optimization and management, MOEAs are well suitable for solving and researching this class of problems. A simple engineering design problem, combined with the algorithm, MOEADC, proposed in chapter 4, is used to discuss the importance in practical application, and the useful insights about the relationship among decision variables corresponding to the Pareto-optimal solutions. Afterward based on MOEADC, a case about a class of portfolio selection problems is researched. Compared to other relevant results in literature, This method offers more comprehensive information, much more flexibility-to a decision-maker in making a portfolio selection, and therefore, to decision-makers or managers in engineering design optimization and management problems in practical applications.
Keywords/Search Tags:multi-objective optimization problems, evolutionary algorithm, multi-objective evolutionary algorithm, hybrid evolutionary algorithm
PDF Full Text Request
Related items