- 學位論文
基於適應式橢圓型體之進化型模糊辨識器之設計
An Evolutionary Fuzzy Classifier with Adaptive Ellipsoids
摘要
本論文旨在運用多個橢圓型體來近似決定區域進行辨識的模糊分類器設計。為了要學習橢圓型體的大小和方向，本論文提出稱為進化型橢圓型體分類的演算法（EECA），它結合Gustafson-Kessel演算法(GKA)與基因演算法(GA)的運用。在EECA的基因演算法用來學習每個橢圓型體的大小。使用每一個橢圓型體的大小去編碼和有智慧的估計GA染色體，其中GKA被用於學習相對應的橢圓型體。 GKA能夠運用適應式的距離範數去表示依原型資料點的分佈所學習到橢圓型體大小。本論文也提出一個GA染色體的初始化估測方法，可改進提升EECA的學習效率。因為EECA是透過自動調節橢圓型體去學習適當的大小和方向進而對原型資料點分群，包含在橢圓型體的資訊被進一步運用在改善群集有效性。本論文運用每個群集內散佈距離的總和與群集間分離距離的比值提出一個群集有效性測量方法。所提出的群集有效性測量方法利用EECA的學習能力優勢，作為一種有效的指標來確定識別分類所需的橢圓型體適當數量。 其次，本論文將EECA模式識別分類算法運用在光碟機提出了一個新的光碟片識別方法。基於光碟機的反射雷射信號，測量4個特徵信號，利用EECA模式光碟片判別方法進行分類。實驗用六種不同類型的光碟片透過EECA模式識別分類算法成功的被判別分類。 最後，在本論文中運用多個橢圓型體來近似決定地區提出了一個有效率的分類方法。本論文提出了一個進化式GKA的演算法（EGKA）是由粒子群聚演算法（PSO）結合Gustafson-Kessel演算法（GKA）。在EGKA中粒子群聚演算法學習每一個橢圓型體的體積，而每一個橢圓型體的參數是由GKA學習。本論文提出了一種在不完整的問題中能估計缺失值的策略，這策略結合了EGKA和回歸插補法。由於EGKA能自我適應學習判定區域，群聚分群結果能更準確的學習非球型的群聚分佈。缺少的特徵數值是使用完整的資料標準距離模下最接近的向量估測。由這策略恢復的不完整的資料數值能更接近原始資料，以使錯誤分類可以被有效地降低。在本論文中所提出的策略與其他策略利用5個2維的人工設計含有缺失特徵數值的資料集和三個UCI資料庫含有缺失特徵數值的資料集進行比較。實驗結果證明本論文中所提出的策略性能是有效的。
並列摘要
A fuzzy classifier using multiple ellipsoids to approximate decision regions for classification is designed in this dissertation. An algorithm called the evolutionary ellipsoidal classification algorithm (EECA) that integrates the genetic algorithm (GA) with the Gustafson-Kessel algorithm (GKA) that learns the sizes and orientations of ellipsoids is proposed. The GA is employed within the EECA to learn the size of every ellipsoid. With the size of every ellipsoid encoded and intelligently estimated in the GA chromosome, GKA is utilized to learn the corresponding ellipsoid. GKA is able to adapt the distance norm to the underlying prototype data points distribution for an assigned ellipsoid size. A process called directed initialization is proposed to improve the EECA’s learning efficiency. Because the EECA learns the data point distribution in every cluster by adjusting an ellipsoid with suitable size and orientation, the information contained in that ellipsoid is further utilized to improve the cluster validity. A cluster validity measure based on the ratio of summation for each intra-cluster scatter with respect to the inter-cluster separation is defined in this dissertation. The proposed cluster validity measure takes advantage of the EECA’s learning capability and serves as an effective index for determining the adequate number of ellipsoids required for classification. This dissertation presents a novel disc discrimination method that uses a pattern recognition scheme called evolutional ellipsoid classification algorithm (EECA). Four features are measured based on reflected laser signals and utilized as disc discrimination method inputs for classification. Six different disc types were successfully discriminated using the proposed method. An efficient scheme imputing missing features from incomplete data is proposed in this paper. The missing features are imputed based on a group of nearest complete data in the residual features space of the incomplete data to be recovered. In order to find the complete data points in the residual features space, an algorithm called evolutionary Gustafson-Kessel algorithm (EGKA) is proposed that learns the ellipsoid to adaptively cluster the complete data points with the incomplete data points to be recovered. A linear regression model is utilized to impute the missing features based on the complete data clustered by the ellipsoid learned by the EGKA.