Researches On Multi-Objective Genetic Algorithm Based On Adaptive ε Dominance

There exist considerable number of Multi-objective Optimization Problems (MOPs), which can't be well solved by traditional methods. Multi-objective Genetic Algorithm does well in solving highly complicated MOPs as well as searching for a set of tradeoff Pareto solutions by a single simulation run. Thus, it interests many researchers, who propose a great many MOGAs, such as, SPEA2, NSGAII, PESAII and etc., among which some need much time costs to gain favorable diversity, while other efficient algorithms possess unfavorable diversity. For example, SPEA2 gets better diversity set than NSGAII by more time cost.Based onε-dominance, Deb and the other researchers proposeε-MOGA, which usesεparameter to divide the whole Pareto optimal front into some hyper-boxes, each of which has no more than one non-dominated solution to maintain the diversity by usingε-dominance. This algorithm without truncation operator can achieve a good tradeoff between efficiency and diversity in certain conditions, but it needs to setεparameter according to Pareto optimal front and the number of non-dominated solutions desired by users. Theεparameter can't be reasonably set due to the failure to know the exact Pareto optimal front in many practical problems. Concerning the above problems, we propose a multi-objective genetic algorithm based on self-adaptationε-dominance, namely AEMOGA, by making use of advantages of other MOGAs. It doesn't need to setεparameter at initial running of the algorithm. It uses banker's principle to build non-dominated set and adopts a truncation operator by improving SPEA2's truncation operator. Furthermore, a self-adjusting method ofεparameter is proposed. Finally, my AEMOGA performs well in efficiency, diversity and convergence by comparing AEMOGA with some classic MOGA. Key words: self-adaptation;ε-dominance; multi-objective genetic algorithm;...