本論文進行傳統法與類神經網路法應用於油浸式電力變壓初期故障診斷效能之比較性研究。傳統診斷法包括變壓器異常判斷法、數碼法及氣體模式法，其中數碼診斷法包括Dornenburg、Rogers、IEC、日本電協研、Glass及Laborelec等6種，類神經網路法方面，則選擇徑向基、倒傳遞及支撐向量機等3種。研究中所採用的分解氣體樣本來自高雄港務局、台灣電力公司及IEC資料庫。研究結果顯示3種類神經網路輸入採單獨氣體濃度效能高於IEC數碼診斷法。於測試未知輸入樣本上，支撐向量機法準確性為97％，訓練時間56秒；徑向基網路法準確性93.9％，訓練時間89分；倒傳遞網路法準確性78.8％，訓練時間19分5秒。可見支撐向量機及徑向基網路之準確性是可被接受的，但徑向基網路訓練花費時間過長，於快速變壓器故障診斷上較不適合。此外，在實際變壓器故障診斷上，5組油中氣體，分別由傳統診斷法及類神經網路診斷法辨別其故障類型。結果Rogers、IEC、日本電協研、氣體模式診斷及支撐向量機等方法皆能成功辨別故障類型；Dornenburg法雖皆診斷出故障，但每一故障均檢測出2種類型；而Glass法有1個案例診斷錯誤；Laborelec有2個診斷錯誤；徑向機法與倒傳遞網路法則只成功診斷出2個故障類型。

This thesis presents a comparative study of traditional methods and artificial neural network (ANN) for the incipient fault diagnosis of oil-filled power transformer. The traditional diagnosis criterion includes transformer’s fault identification table, gas ratio code, and gas pattern; on the other hand, radial basis function network (RBFN), back-propagation network (BPN), and support vector machine (SVM) algorithms are selected as artificial neural networks. The gas ratio methods consists of Dornenburg, Rogers, IEC, JECR, Glass, and Laborelec etc. All dissolved gas samples were collected from Taipower Power Company, Burea of Kaohsiung Port, and IEC 60599. The research results show that performances of all neural networks methods based on individual gas component concentration are better than gas code method. For the test of unknown input samples, the accuracy of SVM is 97％ and its training time is 56 seconds; the accuracy of RBFN is 93.9％ and its training time is 89 minutes; the accuracy of BPN is 78.8％ and its training time is 19 minutes and 5 seconds. It can been seen that the fault diagnosis accuracy of SVM and RBFN are satisfactory. But RBFN took too much time to obtain the results, it is unacceptable for fast diagnosis of transformer fault. Besides, five dissolved gas are used for the fault diagnosis of field transformers, and these gas are respectively analyzed by different traditional method and ANN method to identify fault type. The results illustrate that Rogers, Doenenburg, IEC, Japanese Electric Institute, Gas pattern, and SVM can successfully detect all fault types; while Doenenburg diagnosed two fault types for each case. But RBFN and BPN methods only identified two faults; one case was wrongly classified by Glass method; two unsuccessful diagnosis happened to Laborelec method.

中文摘要-----------------------------------------------------Ⅰ
英文摘要-----------------------------------------------------Ⅱ
目錄---------------------------------------------------------Ⅳ
圖目錄-------------------------------------------------------Ⅵ
表目錄-------------------------------------------------------Ⅶ
第一章 緒論----------------------------------------1
1.1 研究背景與動機------------------------------1
1.2 國內外相關研究概況--------------------------1
1.3 研究貢獻------------------------------------3
1.4 論文內容概要--------------------------------3
第二章 變壓器故障診斷------------------------------4
2.1 前言----------------------------------------4
2.2 油中溶解氣體分析----------------------------4
2.2.1 油中氣體來源----------------------------------5
2.2.2 氣體層析法----------------------------------5
2.3 故障診斷--------------------------------------6
2.3.1 變壓器異常的判定----------------------------6
2.3.2 傳統診斷法----------------------------------7
2.4 傳統診斷法於實例診斷效能之比較分析----------17
第三章 類神經網路演算法----------------------------23
3.1 簡介----------------------------------------23
3.2 類神經網路類型------------------------------23
3.3 類神經網路運作過程--------------------------24
3.4 類神經網路演算法----------------------------25
3.4.1 徑向基類神經網路----------------------------25
3.4.2 倒傳遞類神經網路----------------------------27
3.4.3 支撐向量機----------------------------------34
第四章 變壓器故障案例診斷--------------------------39
4.1 簡介----------------------------------------39
4.2 不同演算法之性能評估------------------------39
4.2.1 IEC氣體濃度比值診斷法(已知輸入樣本)--------39
4.2.2 IEC氣體濃度比值診斷法(未知輸入樣本)--------42
4.2.3 單獨氣體濃度診斷法(已知輸入樣本)-----------42
4.2.4 單獨氣體濃度診斷法(未知輸入樣本)-----------44
4.3 案例診斷-----------------------------------45
第五章 結論與未來研究方向-------------------------48
5.1 結論---------------------------------------48
5.2 未來研究方向-------------------------------48
參考文獻---------------------------------------------------50

[1]E. Doernenburg and W. Strittmatter, “Monitoring Oil Cooling Transformers by Gas Analysis,” Proceedings of Brown Boveri Review, Vol 61, May 1974, p.p.238-247.
[2]IEEE Std.C57.104-1991, ”IEEE Guide for The Interpreation of Gases Generated In Oil-Immersed Transformers,” Proceedings of Transformers Committee of the IEEE Power Engineering Society. Nov. 1991.
[3]J.J. Kelly, “Transformer Fault Diagnosis by Dissolved-Gas Analysis,” Proceedings of IEEE Transactions on Industry Applications,Vol. 16, Dec. 1980, p.p.777-782.
[4]IEC Publication 599, ”Interpreation of the Analysis of Gases in Transformers and Other Oil-Filled Electrical Equipment in Service,” First Edition 1978.
[5]童耀宗，變壓器油取樣與油中氣體分析，台電綜合研究所2005年9月。
[6]S.A. Ward, “Evaluating Transformer Condition Using DGA Oil Analysis,” Proceedings of IEEE 2003 Electrical Insulation and Dielectric Phenomena, 2003, p.p.463 – 468.
[7]J.L. Guardado, J.L. Naredo, P. Moreno, and C.R. Fuerte, “A Comparative Study of Neural Network Efficiency in Power Transformers Diagnosis Using Dissolved Gas Analysis,” IEEE Transactions on Power Delivery, Vol. 16, Issue 4, Oct. 2001, p.p.643-647.
[8]石慶男，以多層支撐向量機結合株落選擇演算法建構電力變壓器故障診斷系統，國立高雄應用科技大學電機工程系碩士班碩士論文，2006年7月。
[9]M.dual and C. Lamarre, “The Characterization of Electrical Insulating Oils by High-Performance Liquid Chromatography,” IEEE Transactions on Electrial Insulation,Vol. 5, Oct. 1977, p.p.340-348.
[10] V.G. Arakelian, “The Long Way to the Automatic Chromatographic Analysis of Gases Dissolved in Insulating Oil,” Proceedings of IEEE Electrical Insulation Magazine, Vol. 20, Issue 6, Nov.-Dec. 2004, p.p.8 – 25.
[11]黃美玲，陳幸宜，陳貴琳，李雅雯，類神經網路輔助醫療診斷分類模式之建構，中華民國品質學會第42屆年會暨第12屆全國品質管理研討會論文集，2006年4月，p.p.1-12。
[12]R. Aggrawal and Y. Song, “Artificial Neural Networks in Power System,” Proceedings of Power Engineering Journal, June 1997,p.p.7-31.
[13]王進德，類神經網路與模糊控制理論入門與應用，全華科技圖書股份有限公司，2007年1月。
[14]葉怡成，應用類神經網路，儒林圖書有限公司，2001年3月。
[15]葉怡成，類神經網路模式應用與實作，儒林圖書有限公司，2004年4月。
[16]C. Cortes and V. Vapnik, ”Support Vector Network,” Proceedings of the College of Information Sciences and Technology at Penn State, June 1995, p.p.273-295.
[17]J.W. Lu, K.N. Platanitotis, and A.N. Venetsanopoulos, “Face Recognition Using Feature Optimization and Support Vector Learning Neural Networks for Signal Processing XI,” Proceedings of the 2001 IEEE Signal Processing, Sep. 2001, p.p.372-382.
[18]H. Hu, “Support Vector Machine-Based Classification Scheme for Myoelectric Control Applied to Upper Limb” IEEE Transactions on Biomedical Engineering, Vol. 55, Issue 8, Aug. 2008, p.p.1956 – 1965.
[19]K. Jung, H. P. Se, and H. J. Kim, “Support Vector Machines for Texture Classification” IEEE Transactions on Pattern Analysis and Machine Intelligence, Vol. 24, Issue 11, Nov. 2002, p.p.1542 – 1550.
[20]朱映霖，利用支撐向量機改善最小錯誤鑑別式之語者辨識方法，國立中央大學電機工程研究所碩士論文，2007年5月。
[21]馬云潜,張學工,”支撐向量機函數擬合在分形插值中的應用,”中國清華大學學報,2000年4月, p.p.76-78.
[22]林長青，支撐向量機應用於科學探索，國立雲林科技大學電子與資訊工程研究所碩士論文，2003年6月。