Abstract:
Many-objective optimization is very important for numerous practical applications. It, how- ever, poses a great challenge to the Pareto dominance based evolutionary algorithms. In this thesis, a fuzzy dominance based evolutionary algorithm is proposed for many-objective opti- mization. The essence of the proposed algorithm is that it adaptively determines the fuzzy membership function for each objective of a given many-objective optimization problem. Fur- thermore, it emphasizes both convergence and diversity of all the evolved solutions in the same way by using one selection criterion. This is why our algorithm employs the reference points for clustering the evolved solutions and selects the best ones from different clusters in a round-robin fashion. The proposed algorithm has been tested extensively on a number of benchmark problems in evolutionary computing, including eight Waking-Fish-Group (WFG), three Deb-Thiele-Laumanns-Zitzler (DTLZ) problems having 2 to 25 objectives and three in- stances of Rectangle problem. The experimental results show that the proposed algorithm is able to solve many-objective optimization problems efficiently, and it is compared favorably with the other evolutionary algorithms devised for such problems. A parametric study is also provided to understand the influence of a key parameter of the proposed algorithm.
