分散式基因演算法於結構最佳化設計之應用

Abstract

基因演算法是在傳統的數值分析之外，另一種分析結構最佳化問題的 途徑。使用基因演算法，可以有效利用演算法中的演化運算子來分析最佳 化求解的問題，但由於演化過程的計算量太大，故往往需要花費較長的時 間等待結果。將基因演算法作分散式處理的目的，就是希望能藉由均分其 工作量，增加程式的執行效率。本研究係將基因演算法配合分散式處理工 具 PVM作分散式處理，利用本文中所提出的兩種程式架構，將基因演算法 中演化的工作，平均分配給每一台工作站，讓這些工作站能同時進行演化 的程序。在本研究最後，將舉 4個實例，利用高速電腦中心的八部ELC cluste工作站，來測試兩種程式架構所執行的結果。由結果可以比較出兩 種程式架構中，較適合的程式架構；而由效率提昇值圖當中，亦可以看出 基因演算法的演化效率平均最高可以達到大約5左右的效率提昇值。 Genetic algorithms (GA) have been widely used in the fields of optimization problems. Instead of calculating the gradient of traditional numerical methods, GA use the operators of evolution in algorithms can efficiently analyze optimal problems to avoid the problem of local minimum. Since the convergence of GA is highly depend on the number of chromosomes in each population, GA always need long time to reach the optimal solution. In this study, a distributed GA has been developed to increase the efficiency of GA by using PVM operation under a network computation system. Two different approaches in programming have been developed and presented in the paper. The system has been implemented using eight ELC cluste workstations in High- Performance Computing Center. Four structural optimization examples have been used to test the system under these two different approaches and the results have been compared as well as discussed. The best speed-up value of distributed GA can reach about an average of 5.