Please use this identifier to cite or link to this item:
標題: 使用演化策略的多目標最佳化演算法
Multiobjective Optimization Algorithm Using Evolution Strategies
作者: 徐永勝
Hsu, Yung-Sheng
關鍵字: multiobjective optimization
evolutionary computation
evolution strategy
Pareto optimal
genetic algorithm
出版社: 機械工程學系
摘要: 為了克服傳統多目標最佳化方法的缺點,近年來多種基於演化式計算的多目標最佳化方法相繼被提出及改良,並廣泛的進行比較,在許多文獻中已證實這些設計方法的效率很高,可以快速得到多組最佳解。 本研究提出一種新的演化式方法來求解多目標最佳化問題,與其他方法相異之處在於本文方法是以基於排名與擁擠指標來指定適應值,與其他根據凌勝等級來指定的方式不同。本文選擇用演化策略ES方法作為演化法則,同時配合菁英保留和避免優越個體過度集中的技巧,以獲得大量既收斂又分佈均勻的多目標最佳解。本文方法除了可解無限制條件的問題外,亦可處理有限制條件的問題。 本研究對十數個包含有限制條件及無限制條件問題進行測試,並在部分問題上與其他方法作比較。測試結果顯示出本方法對大部分測試題目而言可以得到不錯的Pareto最佳解,但是也有一些缺點。本文除了探討可能的原因之外,亦提供了一些改進之建議。
In order to overcome the shortcomings of traditional multi-objective optimization(MO) methods, several new methods based on evolutionary computation(EC) have been designed and continually improved in these years. Through intensive tests and comparisons, these new methods have been proved to be able to provide high efficiency in producing MO solutions. This paper trys to develop a new MO problem solver based on EC to solve MO problems. The main difference between proposed and other method is that the fitness is calculated based on rank and crowdness instead of domination level used by others. The evolution strategies(ES) is employed to be the tool to simulate the evolution process. Furthermore, both elitism strategy and anti-crowdness skill are used to get high-quality solutions which converge to Pareto front and are evenly distributed on the front. Both constrained and unconstrained MO problems can be solved by the proposed method. The method will be tested against several MO problems including constrained and unconstrained ones. Comparison are made among three methods for some test problems. The results have shown that the proposed method can provide satisfactory Pareto solutions for most test problems, and some drawbacks are observed. The possible reasons behind these drawbacks are discussed and some recommendations are given for further research.
Appears in Collections:機械工程學系所



Show full item record
TAIR Related Article

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.