限制推理型之粒子群演算法於解決分群問題之研究

本論文提出一個限制推理型之粒子群演算法用以解決資料分群問題。由於粒子群演算法主要是透過適應值函數來評估資料集中分群距離之總和，但是當問題的複雜度提高時，除了搜尋時間會相對提高外，且有可能會陷入區域最佳解之困境。基於上述之缺點，本研究根據使用者要求，加入自訂限制條件，整合限制推理及粒子過濾機制於粒子群演算法架構中，可以有效減少粒子產生的搜尋空間，使得粒子群演算法能夠更快速的找出符合限制之最佳解。根據實驗結果顯示，限制式粒子群演算法較一般傳統的粒子群演算法有更快速且較符合目標之結果。

關鍵字

限制推理；資料分群；粒子群演算法

並列摘要

This paper proposes a constraint-based particle swarm optimization approach to solve the problem of data clustering. Since the existing particle swarm optimization algorithm designed for searching the cluster centroids is mainly to evaluate via a fitness function. Major drawback of Particle Swarm Optimization for the problem of data clustering is computation inefficiency when the complexity of the clustering problem increases. In this paper, we propose a constraint-based particle swarm optimization approach for solving the clustering problems. This study integrated constraint-based reasoning mechanism to reduce the search space and produce better solutions. The proposed approach is compared with a regular Particle Swarm Optimization using UCI repository data sets. According to the experimental results show that the constraint-based particle swarm optimization have faster and more consistent with the user's preferences than a regular Particle Swarm Optimization.