標題: 使用SARSA( λ)決定PSO和CPSO-S最佳的切換時機達成函數值最佳化
Using SARSA(λ ) to Find the Best Switching Policy Between PSO and CPSO-S to Achieve Optimal Function Values
作者: 劉鎮榮
Liu, Zhen-Rong
林昇甫
Lin, Sheng-Fuu
電控工程研究所
關鍵字: 粒子群最佳化;增強式學習;particle swarm optimization;SARSA(λ)
公開日期: 2011
摘要: 本篇論文提出一個演算法,使用SARSA(λ)來決定基本粒子群最佳化及協同式粒子群最佳化的使用時機。經測試函數的極值搜尋結果得知,由於本論文所提出演算法結合了基本粒子群與協同粒子群最佳化,在所挑選的五個測試函數中維度間不論是否關聯以及測試函數具有單峰極值或是多峰極值的情況,在執行相同的時間之下的最佳值搜尋結果比起基本的粒子群最佳化、以及將原本的粒子群分割成維度更小的子群的協同式粒子群最佳化、基本粒子群最佳化和協同粒子群最佳化交替作用的混合式粒子群最佳化等這三種演算法的結果都能夠有最佳或是次佳的表現。即代表本論文在演算法處理各種函數最佳值搜尋的問題上比起基本粒子群最佳化和協同式粒子群最佳化僅僅處理特定函數問題上要來的好,而在能夠處理不同函數的問題上又能夠比混合式粒子群最佳化需要更少的搜尋時間此為本論文的貢獻。
This paper proposed an algorithm, using reinforcement learning mechanism to determine the best method of particle swarm optimization and cooperative particle swarm optimization. The test function extremum search showed that, the proposed algorithm in this paper combining with particle swarm optimization and cooperative particle swarm optimization can deal with a function with or without dimensional correlation and unimodal or multimodal, in the same execution time proposed algorithm in this paper compared to the basic particle swarm optimization, cooperative particle swarm optimization, hybrid particle swarm optimization. The results show that the proposed algorithm is able to have the best or second performance of other three algorithms
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT079512571
http://hdl.handle.net/11536/41089
Appears in Collections:Thesis


Files in This Item:

  1. 257101.pdf
  2. 257102.pdf

If it is a zip file, please download the file and unzip it, then open index.html in a browser to view the full text content.