改良式非同步並行處理之粒子群聚最佳化法
No Thumbnail Available
Date
2008-06-07
Authors
許陳鑑
林耕宇
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
本文提出ㄧ種 改良式非同步並行處理之粒子群聚最佳化法 ，以提升粒子群聚最佳化法在不同質(heterogeneous)的計算環境中之計算效率。作法上係綜合傳統的同步與非同步並行處理計算法，以僕工作端(slave)之性能為基準，分配適當的粒子數量，以減少工作站等待時間的浪費，使計算效能得以提升。為評估本文所提出方法之有效性，我們將以minimax 最佳化問題及系統模型降階的問題作為標的，分別使用傳統的同步並行處理、非同步並行處理、ㄧ台獨立電腦、以及本文所提出之方法做比較。實驗結果指出，我們所提出的方法在兩個範例都有較好的性能展現。
An enhanced asynchronous parallel computation scheme for particle swarm optimization (PSO) is proposed in this paper to improve computational efficiency for heterogeneous workstations. Taking advantages of the conventional parallel computation methods of synchronous and asynchronous approaches, the proposed method distributes appropriate number of particles to slave workstations depending on performance of the individual workstations. As a result, problems of idle time and extra communications between master and slaves associated with synchronous and asynchronous parallel computation methods are accordingly avoided. To validate the effectiveness of the proposed method, we adopt a minimax optimization and model reduction problem as target problems for optimization by synchronous, asynchronous, a single workstation, and the proposed method, respectively. Simulation results indicate that the proposed method has a good computational performance for these two examples, with a significant improvement on the computation efficiency.
An enhanced asynchronous parallel computation scheme for particle swarm optimization (PSO) is proposed in this paper to improve computational efficiency for heterogeneous workstations. Taking advantages of the conventional parallel computation methods of synchronous and asynchronous approaches, the proposed method distributes appropriate number of particles to slave workstations depending on performance of the individual workstations. As a result, problems of idle time and extra communications between master and slaves associated with synchronous and asynchronous parallel computation methods are accordingly avoided. To validate the effectiveness of the proposed method, we adopt a minimax optimization and model reduction problem as target problems for optimization by synchronous, asynchronous, a single workstation, and the proposed method, respectively. Simulation results indicate that the proposed method has a good computational performance for these two examples, with a significant improvement on the computation efficiency.