於本篇論文中，我們提出一種改良式的粒子群演算法，名稱為適應性自我學習粒子群演算法(Adaptive Self-Learning Particle Swarm Optimization, ASLPSO)，並將其應用於資料分群之問題。本文利用自我學習機制，讓粒子能夠向表現比他更好的其他粒子學習，以獲取有用的資訊，並再透過動態的模式轉換策略改進粒子的搜尋能力，使粒子能在演算法疊代過程的不同階段，轉換其搜尋模式，以提高找到全域最佳解的可能性。我們最後使用16種測試函數進行模擬，與其他已提出的不同改良式粒子群演算法做比較，實驗的結果表示，本文所提出的改良方法可以在大部分的測試函數中有著較佳的表現。最後並將本文的改良式演算法運用在資料分群的問題上，我們可以在某些性能指標上得到更好的結果，但也有較差的部分，這顯示本文的方法仍有進一步改善的可能。;This thesis proposes a new particle swarm optimization (PSO) called Adaptive Self-Learning Particle Swarm Optimization (ASLPSO), and applies it to the classification problem. A self-learning method is introduced in the ASLPSO that every particle randomly selects its learning object among the better particles to acquire useful information. We also designs a dynamic transition strategy to improve the searching approach of particles during the iterations. In the experiments, the performance of the proposed ASLPSO is compared to several improved PSO’s in the literature by testing sixteen benchmark functions. The experimental results show that the proposed algorithm performs better on most of the functions. At last, the ASLPSO is applied to a classification problem. In our experiments, many classification results are better, but not all. To be more precisely, the ASLPSO is supposed to be refined in some ways.
