- 學位論文
應用類神經網路於避雷器狀態診斷之研究
Study on Arrester Condition Diagnosis Using Artificial Neural Network
摘要
近年來國內高科技園區、工業區迅速擴展，致使增建多所超高壓變電所與第六、第七輸電計畫，傳輸電線路增加，但每年仍會遭遇到數十次之輸電線路事故，其中約有50%來自於落雷事故，為求提供穩定電力傳輸，避免雷擊引起電力跳電、電壓驟降或突波事故，因此大量採用避雷器保護。但近年來台電公司依據政府採購法，在競標結果均由印度廠商得標，其品質較差造成不少的避雷器事故，為使避雷器能夠達到有效安全防護及提早發現避雷器特性劣化，本研究提出有效避雷器診斷方式，經比較各種現場檢測方法，研究發現最佳診斷方式是以電阻性洩漏電流量測為最佳參考依據。 本論文主要之研究內容為應用類神經網路於避雷器狀態診斷上，為使檢測成本降至最低，採用不停電之線上量測(on-line measurement)，量測時不須拆卸避雷器，僅須在避雷器之接地線量測其洩漏電流數分鐘，即可利用軟體推估追蹤量測週期及殘餘壽命評估，為未來推動智慧型電網的一部份；而為達到有效之評估診斷，類神經網路則須有效的訓練參數資料。整合類神經網路與電阻性洩漏電流量測，可令避雷器運轉狀態與壽命估測更精準評估。本研究成功地找出最佳測量方法，並將判別規則與診斷模式透過類神經網路訓練函數建立診斷工具來評估避雷器狀態，使現場技術人員可利用簡易人機介面完成避雷器之測量並告知下次追蹤量測週期及避雷器狀態，為實用維護避雷器之工具。
並列摘要
In the recent years, science industrial parks rapidly expanded in the domestic Taiwan. Such as sixth, seventh electricity transmission line construction plans and the amount EHV substation increased. The length of transmission lines increased, but the transmission lines fault has increased proportional too. Hundreds of outage, voltage dips and flashover faults were met in our transmission system every year. Half of these faults came from the lightning fault. Therefore we use a large amount of arresters protecting power transmission line and equipment for keeping power system stability. But in recent years, Taiwan Power Company bought cheapest, poor quality Indian's arresters by follow the low of government procurement. In order to achieve system stability requirement, a good diagnosis method or tool shall be developed for pick up the deterioration arresters. By comparing a variety of methods, we found the best diagnostic detection method is the resistive leakage current measurement and it will be adopt in this study. I will using Artificial Neural Network to assist arrester diagnosis , and using on-line leakage current measurement for reduce costs and maintenance labor work, such as arrester disassemble work, outage application paper work. It just measures arrester's leakage current from ground wire for a few minutes only, then key in the measured current, using software to know arrester's residue life assessment, treatment guidance and estimating next measurement cycle time. This will be a part of smart-grid for future. The leakage current is artificial neural network's effective training parameter for important data. Combining the artificial neural networks and resistive leakage current measurement, promotes the accuracy of estimating the arrester’s status and life cycle. The report successfully developed one of the best methodologies to assess the arrester state. For helping site technicians to use this system, a simple human-machine interface was introduced, the arrester's status and treatment guide information is shown on software too, so it is really practical arresters maintenance tool.