Evaluated Preference Genetic Algorithm for Multiobjective Optimization and the Application

[[abstract]]本篇論文提出EPGA (Evaluated Preference Genetic Algorithm) 的方法來解決 多目標最佳化問題。在現代工程設計中，多目標最佳化問題通常是彼此衝突的。 一般來說，要找到最佳解是相當困難的，所以在大部分現實應用上，找到所想要 的解區域是比較合乎需求的。本論文以MOI (Method of Inequalities) 為基礎，讓 使用者能夠根據需求，尋求其所想要的解空間。相對於大多數的多目標進化演算 法，EPGA 提出一個新的指派適應值方式，而且計算複雜度為O(MNlogN)。藉由 Deb 所提出ZDT 測試問題來對EPGA 做測試。而且我們以一個動態最佳化問題 來說明EPGA 可以廣泛的運用於尋找所想要的解區域的問題。最後以控制器參數 設計問題為例，展現此演算法的效率及能力。由實驗得知，EPGA 不但可以找到最 佳解或是近似最佳解，而且比演算法SPEA (Strength Pareto Evolutionary Algorithm), NSGA-II (Elitist Non-dominated Sorting Genetic Algorithm), and PAES (Pareto Archived Evolution Strategy) 能有更好的計算效率。[[abstract]]In this thesis, an evaluated preference genetic algorithm (EPGA) is proposed to solve multiobjective optimization problems (MOPs). Most objectives of the optimization problems are usually conflicting with each other in modern engineering design. Therefore, it results in much difficulty in finding the optimal solutions. In addition, it’s more useful to find the solutions in the preferred conditions in most realistic applications. The proposed algorithm is based on the method of inequalities (MOI), and can search the solutions in the preferred spaces where the user can define the requirements. Compared with most multiobjective evolutionary algorithms (MOEAs), the EPGA proposes a new way to assign the fitness, where the run-time complexity is O(MNlogN). EPGA is tested with the ZDT test problems proposed by Deb [1], and can be widely applied to the kind of problems which find the preferred objective space, and an example of dynamic optimization problems is used to illustrate that. Finally, the design of the controller parameters is used to verify the efficiency and capability of EPGA. The experiments show that the proposed EPGA not only can find optimal or close-to-optimal solutions but also can compute more efficiently than the Strength Pareto Evolutionary Algorithm (SPEA), Elitist Non-dominated Sorting Genetic Algorithm (NSGA-II), and Pareto Archived Evolution Strategy (PAES)