| Summary: | 碩士 === 逢甲大學 === 資訊工程學系 === 87 === This thesis proposes an efficient generalized multiobjective evolutionary algorithm (GMOEA) using orthogonal arrays for finding all nondominated solutions of multiobjective optimization problems (MOOPs). In the proposed algorithm, a novel generalized Pareto-based scale-independent (GPSI) fitness value with no use of a weighted-sum of multiple objectives is applied for coping with the difficulty involved in making the trade-off decisions to arrive at a single measure of performance. Furthermore, orthogonal arrays are used to achieve intelligent crossover (IC) that the chromosomes of the children are formed from the best combinations of the better genes representing variables of a function from the parents rather than the random combinations of parents’ genes. The proposed IC operation with the GPSI fitness value can economically find all nondominated solutions of MOOPs without additional use of traditional local search procedures. The simple and efficient general-purpose algorithm GMOEA is more superior for solving MOOPs with a large number of parameters in convergence speed and accuracy. High performance of our algorithm capable of handling any generally formulated MOOPs is demonstrated by applying it to test functions gleaned from literature and three real practical applications, finding the minimum reference set for nearest neighbor classification, solving the flexible process sequencing problem, and bicriteria network topological design problem. From the encouraging experimental results, it is shown empirically that GMOEA outperforms the existing superior methods.
|
|---|
