The Artificial Bee Colony Algorithm For Multi-objective Optimization Problems

Multi-objective optimization problems(MOPs) are commonly used in practical engineering application and scientific research, which attract great attention of academic community around the world. Multi-objective Evolutionary Algorithms(MOEAs) have been recognized as efficient methods to deal with MOPs. The artificial bee colony algorithm belongs to the swarm intelligence optimization algorithm. This thesis aims at designing efficient and effective MOEAs for MOPs based on the artificial bee colony algorithm. The results are verified on handling benchmark test problems. The contributions of this thesis can be summarized as follows:1) The basic artificial bee mutation operator is easy to trap into local Pareto fronts, so a combined mutation operator is suggested.2) An artificial bee colony algorithm based on Pareto dominance is proposed. The main technologies of this algorithm are as follows.(a) The updating way in population is based on Pareto dominance.(b) The fitness of each individual is also based on Pareto dominance and the selection way for onlooker bees is based on roulette.(c) An archive is adopted in our algorithm to save the nondominated solutions generated in each generation.3) Within ?-MOABC, the updating way of individuals in population and archive are both based on ?-indicator. Our methods employ a fixed-sized archive to maintain the solutions generated in each generation. The selection scheme in population and archive is both based on the quality indicator I? ?. The fitness assignment based on quality indicator I? ? tries to rank the population members according to their usefulness regarding to the optimization goal. As to the uncertainty of the true Pareto front, which makes the contributions of each individual to the population different, it is necessary to allocate more computing resources for those much outstanding individuals. So the selection mechanism of individuals is based on power law when the onlooker bees select the food source.Compared with other state-of-art algorithms, the proposed algorithm proves to be competitive in dealing with a set of standard test problems, such as ZDT, UF and DTLZ test suites.