Research On Multi-objective Optimization Algorithms Based On Evolutionary Algorithm And Their Applications

Multi-objective optimization is quite common in daily life. Studies in this field have great significance for both research and application. Evolutionary algorithm is a random optimization method which simulates natural evolutionary process. By using population-based search strategy, this algorithm has strong capability for global optimization, so that it is particularly suitable for multi-objective optimization. At present, the research on multi-objective evolutionary algorithm has become a priority and highlight in the field of evolutionary computation.In this thesis, the development of multi-objective optimization algorithm is outlined, the concept, basic framework and features of evolutionary algorithm is presented. The focal points are set on introducing and analyzing two classical multi-objective evolutionary algorithms. Simultaneously, simulation results of these classical multi-objective evolutionary algorithms on three test functions demonstrate not only the rapidity and capability of the algorithm, but also the distribution and dominance of the optimal solution set.This thesis designed a fast diversity measurable multi-objective evolutionary algorithm, so that better convergence performance of the algorithm and uniform distribution of the optimal solution will be achieved. The algorithm associates the status of the population with populations'selection and the searching mechanism of optimal solutions. A diversity measurement named NDM based a minimum Euclidean distance of the target space is designed. Dual-elitism-mechanism and a new method based on reserving minimum boundaries is introduced to calculate crowding distance. The proposed algorithm is applied to six different types of test functions to verify that it can effectively speed up the search and improve the diversity of population. Its application on PID optimal design further validates its competence in solving practical problems.With regard to multi-objective optimization problems with complex constraint conditions, an algorithm based on Pareto sorting is proposed. An improved constrain-domination criterion is used, in which feasible and unfeasible solutions are fuzzily distinguished. Unfeasible solutions can then be used to enhance the distribution of population. In the end, Simulation results of the proposed algorithm on three representative test functions confirm its efficiency.Conclusion is made in the end and prospects of future study is formulated.