Multiobjective Evolutionary Algorithms And Their Applications

Evolutionary Algorithm provides a new direction to complex optimization problems. Because of its intelligence, universality, robustness, global search ability and parallelism, it has been widely used in many fields. However, there still exist some open problems in the theories and the applications. This dissertation is focused on the competition and cooperation relationship among biology individuals and environment, individual and individual, and new evolutionary models and algorithms are proposed for multiobjective optimization problems and new performance metrics are presented for them. Difference evolution algorithm based on double population is given for constrained optimization problems. Finally, some works have been done on the multimodal function optimization to search all the peaks (both global and local ones), and new performances for approximating are also proposed. The main works can be summarized as follows:1. Two new evolutionary models are proposed for multiobjective optimization problems from a new point on the competition and cooperation relationship among biology individuals and environment, individual and individual. One is Intelligent Multiobjective Particle Swarm Optimization Based on AER Model and the other is Co-evolutionary Multiobjective Optimization Based on Particle Swarm Optimization. In the former, each particle is considered as a particle with intelligence, and new operators are also designed on the basis of the ability of agents in sensing and acting on the environment, which are used to implement the sharing of information among agent particles and maintain the diversity of the population. In the latter, new operators and selection mechanism are designed for the pending problem to guide the whole evolutionary process. By sharing and exchanging of information among all the particles, it does not only enhance the searching region but also maintain the diversity of the population. The quantitative and qualitative comparisons indicate that the new algorithm can quickly find the Pareto-optimal solutions with maximum possible converge and uniformity along the Pareto Fronts.2. Analysis and discussion are carried out on the existing multiobjective evolutionary algorithms. Since different algorithms may give different sets of nondominated solutions to the same problem. To evaluate the quality of these algorithms, metrics for approximation, uniformity and wide range are presented.3. At present, constrained optimization are the difficult and challenge problems inoptimization fields. Hence, constrained multiobjective and constrained numerical optimization problems are done in this thesis based on double populations. A modified difference evolution is proposed firstly, and difference evolution algorithms with double populations are designed for constrained optimization problems, where infeasible solutions with better performance are allowed to save and participate in the evolution. A large numbers of standard functions are taken to test its performance, which indicate that the proposed algorithm is effective and feasible.4. The halftoning of the gray image and color image is equivalent to a multiobjective optimization problem. Since K-mean algorithm converges quickly while easily getting into local optimum and difference evolution can avoid the disadvantage of the K-mean algorithm. Combining both of their advantage, some work based on multiobjective evolutionary algorithm is done to the halftoning of the gray image and color image. In addition, prior knowledge was taken to construct the initial population in order to improve the convergence velocity. Finally, the standard images are used to validate the performance of the algorithm.5. Generally, most work for multimodal function optimization is to search its global peaks, however, local peaks sometimes also play an important role in practical problems. Hence, a new algorithm based on clonal selection is presented to multimodal function and new operators are designed in it. In addition, to avoid peaks with different height swallowed and maintain the diversity, adaptive mechanisms are designed for both clustering radius and population size, and performance metric are also done on multimodal function. The results show the proposed algorithm can locate peaks as more as possible with the advantage of no resource wasting.