Optimization of Airline Scheduling and Industry Design Problems Using Hybrid Multiobjective Genetic Algorithms

[[abstract]]大多數的工業應用問題，很少能有足夠的知識預先了解其適應值的分佈。因此，通用化之多目標演化計算方法，對於實際上具有不同限制及時效性需求之多元應用問題，經常無法提供單一有效的解決方案。本論文提出兩種混合式方法改進基於不等式之多目標最佳化遺傳基因演算法（MMGA）來解決實務上之應用問題。 第一個混合方法利用可調式評估向量(AEV)導引演化收斂並與MMGA共同合作以作為全域搜尋解決方法。此混合方法有效地平衡多樣性與搜尋效率，可用來解決高時效需求的飛航擾動回復問題；另一方面，混合式MMGA以歐式距離門檻值混合於柏拉圖(Pareto)解次序區分中以強化多樣性，相當適合用來解決離線式但高精度要求之工業設計問題。 數種基於航空排程及工業設計實務上需求的指標，將用以建立相關研究案例，驗證提案方法的強健性。按照不同的工業應用系統的實驗結果，所提出之混合式演算法能獲得相較於專家或其他方法如MMGA與NSGA-II而言，更好的結果。本論文同時也論證現有主流的多目標演化計算方法之修改方法，使其適用於工業界實務問題的解決方案。尤有進者，使用本論文所提出之設計方法，已成功應用於設計製作相關商業化產品。[[abstract]]In most industry applications, there is very little a priori knowledge about the fitness landscape. The solution capabilities of general-purpose multiobjective evolutionary algorithms (MOEA) are usually very limited to serve as an effective solver for widespread applications with different practical constraints and timing requirements. This dissertation thus proposes two kinds of hybrid ways to improve the method of inequalities-based multiobjective genetic algorithm (MMGA) for practical applications. The first hybrid way is to utilize an adaptive evaluation vector (AEV) to guide the evolutionary convergence and cooperates with MMGA to serve as a global solver. It efficiently balances the solution variety and performance, and is capable of solving time-sensitive airline recovery scheduling problems. On the other hand, another hybrid MMGA enhances the diversity maintenance by hybridizing Enclidean-distance threshold into the Pareto ranking. This hybrid way makes the proposed method quite suitable for solving off-line but accurate-required industry interference-free design problems. Several criteria based on practical airline and industry practices are used to build up study cases for verifying the correctness and robustness of the proposed algorithms. According to the experimental results of different industry applications, the proposed hybrid algorithms can obtain better results compared to those made by experts or other methods such as MMGA and NSGA-II. This dissertation also demonstrates the development of practical evolutionary applications which needs to modify the state-of-the-art evolutionary algorithms to suit the industry applications. Furthermore, the design work based on this dissertation has been successfully applied to commercial-use productions