透過您的圖書館登入
IP:18.220.187.178
  • 學位論文

結合分群及分類決策樹於模鑄型變壓器之異常偵測

Abnormality Detection of Cast-Resin Transformers Using The Combination of Clustering Decision Tree

指導教授 : 洪士程
若您是本文的作者,可授權文章由華藝線上圖書館中協助推廣。

摘要


在本論文中,我們提出模鑄式變壓器異常局部放電的偵測與分類方法, 要偵測一個複雜系統中的異常,由於缺少適當的模型,所以是一個困難的 任務;正因如此,這也是讓資料探勘技術具有吸引力的地方。對於模鑄型 變壓器,我們提出一個以分類為基礎的監控分析。在分類部分,我們提出 一個群集決策樹,應用於模鑄式變壓器裡的類別和連續性質屬性的資料分 類問題上。結合了群集決策樹以及CART與平均值、標準差,與數據離散化 以上幾種不同的方式作混和。因屬性的高度混雜使資料無法輕易的分割, 本篇論文使用了分離矩陣為基礎的群集演算法來提取各種輸出類別的特徵 區塊與屬性,利用這些特徵作為判斷輸出類別的依據,來進行分類。在錯 誤偵測與隔離部分,則提出一個基於分類結果準確性,決定是否產生警報 信號的標準,而異常隔離的機制則是依據群集決策樹來隔離局部放電故障。 我們將所提出的分類器應用於模鑄型變壓器所量測到的局部放電訊號進行 分類測試,並和現有的商業軟體See5.0、Weka、Windows SQL 2008 R2來以 實例進行比較分類結果。

並列摘要


In this thesis, we propose a method to detect and classify the abnormal partial discharge of cast-resin transformer. Due to the lack of appropriate model, it''s difficult to detect the abnormality in a complex system. To owe come this drawback, the proposed monitoring analysis method is based on classification. In classification part, we propose a clustering decision tree that utilizes the classification for continuous data in cast-resin transformer. The clustering decision tree, CART, average value, standard deviation, and data discretization. Because the attributes are highly mixed, the data can’t be separated easily. Thus, we used clustering algorithm to extract the characteristic blocks and attributes of outputs, and use these characteristics to classify the outputs. For fault detection and isolation, we propose a technique to make a decision of the alarm. It’s based on the accuracy of classification result. The mechanism of abnormal isolation is according to clustering decision tree to isolate the partial discharge fault. The proposed clustering decision tree is applied in the aboard partial discharge of cast-resin transformer. We also compare the classification result with the existing software See5.0, Weka, and Windows SQL 2008 R2.

參考文獻


立陽明大學醫學工程研究所,九十二年七月。
國立成功大學電機工程學系,九十三年六月。
[1] T. Denoeux, “A neural network classifier based on Dempster-Shafer
theory,” IEEE Transactions on Systems, Man and Cybernetics, Part A, vol.
[2] G. P. Zhang, “Neural networks for classification: a survey,” IEEE

延伸閱讀