至今越來越多研究加入群體智慧(Swarm Intelligence)概念應用於資料分群(Data Clustering)問題,協助截取資料隱含意義提供多樣資訊給管理者,文獻顯示可獲得不錯的效果。其中,粒子群演算法是模擬鳥群覓食的行為所衍生的最佳化搜尋技術,具備快速收斂、易實現和強健性(Robust)的優點,在解空間搜尋最佳解的精度和效率有不錯表現。 本研究針對分群問題,提出的四種研究架構:(1) DPSO Clustering; (2) SPSO Clustering ;(3) DSPSO Clustering;(4)SDPSO Clustering。透過實驗證明各有特色,DPSO Clustering增加粒子更多機會搜尋新空間,突破需要設定固定迭代才會擾動的模式,算法更自動和靈活。SPSO Clustering透過SWAP機制,提升最佳解的精度。DSPSO Clustering以SWAP局部搜尋,進而改變粒子的位置。SDPSO結合SWAP和DPSO,粒子在一定機率下隨機擾動粒子來跳出局部解,並以SWAP提升粒子的最佳解。 實驗結果顯示,本研究提出的概念應用於資料分群問題,確實優於原始粒子群演算法和基因演算法,此外,SPSO和SDPSO與差分演算法也不相上下。
Data Clustering can be implied the meaning of data, providing managers diverse information. Now more and more researches joining on the Swarm Intelligence concept get good results, for example, PSO algorithm. PSO is the simulation of bird populations to search for food with some characters, including rapid convergent and achieve easily, robust and so on. It is good performance in the accuracy and efficiency. In this study, four framework of research include, DPSO Clustering, SPSO Clustering, DSPSO Clustering and SDPSO Clustering. Some characteristics were found though the experiment. DPSO Clustering being more automatic and flexible increase more opportunities to search new space for particle. SPSO use SWAP mechanism to enhance the accuracy of cluster’s center. DSPSO changes the particles’ location with disturbance strategy of SWAP. SDPSO is combined with SWAP and DPSO and helps particle jump out of local solution and enhance final result.