以融合新式親代選擇機制之MOEA/D求解多目標最佳化問題

Abstract

本論文提出一個融入新式親代選擇機制的MOEA/D演算法，用來求解多目標最佳化問題，新式親代選擇機制由三個機制組成：密集度評估機制、收斂評估機制、交配池選擇機制，使用此機制來改良MOEA/D演算法，增進演化效能，使用密集度跟收斂評估機制來分配計算資源，使計算資源不致浪費在無謂的演化上，充分的利用計算資源來達到更有效的演化，使用交配池選擇機制來改變交配池成員，原本MOEA/D演算法的設計是選擇固定成員當成交配池，而有少許機會可以選擇整個族群當成交配池，或許由固定成員進行交配在某些問題上很難逼近Pareto Front，所以我們使用交配池選擇機制來改變交配池成員，由此本論文提出一個新式親代選擇機制架構在MOEA/D演算法中，由實驗結果得知，此機制對於複雜的多目標最佳化問題可以得到很好的效能。This thesis presents the multiobjective optimization using MOEA/D with a new mating selection mechanism to solve multiobjective problems. The MOEA/D algorithm often selects fixed members as the mating pool. The probability for a whole population can be selected as the mating pool is low. Select fixed members to make offspring may difficult to approach Pareto Front in some problems. And the MOEA/D algorithm allocates computing resources equally. It may waste computing resources in useless evolution. Thus, this thesis presents a new mating selection mechanism in the MOEA/D algorithm to solve multiobjective problems. The new mating selection mechanism includes three mechanisms: the crowding distance mechanism, the convergence mechanism, and the mating pool selection mechanism. We use the new mating selection mechanism to improve MOEA/D’s efficacy. This thesis uses the crowding distance mechanism and the convergence mechanism to allocate computing resources, which can avoid wasting computing resources in useless evolution. This strategy makes full use of computing resources in effective evolution. We use the mating pool selection mechanism to change the mating pool members. Experimental results show that the new mechanism has good performance in these problems.