請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/53011
標題: | 運用氣象資訊於熱過載風險評估專家系統之設計:以超高壓輸電網調度為例 Design of a Thermal Overload Risk Evaluation Expert System Based on Weather Information: A Case Study of EHV Grid Dispatching |
作者: | Sheng-Kai Pan 潘聖凱 |
指導教授: | 江昭皚 |
關鍵字: | 熱過載風險,專家系統,貝氏網路, Thermal overload risk,Expert system,Bayesian network, |
出版年 : | 2015 |
學位: | 碩士 |
摘要: | 在電力系統中,輸電線安全是非常重要的議題,有許多的原因會造成電力系統損壞,除了天災之外,輸電線垂降問題會使電力系統涉及危險的情況,一旦輸電線垂降,在山區的線路有可能會觸地或樹,因而有輸電故障發生,造成嚴重的經濟損失。為了要避免與解決輸電線垂降問題,智慧電網的應用中經由架設感測器監控輸電線導體溫度,再透過無線網路即時回傳,但安裝感測器的成本很高,因此這種解決方法不實際。另一方面,台灣電力公司是經由限制輸電電流於安全的閥值內來避免輸電線垂降,然而這樣的方式無法保證輸電線完全不會垂降,所以台電派遣保線員每半年檢查輸電網的輸電線與電塔,而台電保線員要到達電塔十分困難,因此這種方法必須花費大量時間與成本,並且此種察看方法還是擁有高風險。
為了讓台電調度人員規劃超高壓輸電網調度,本研究提出一個即時熱過載風險評估專家系統,此系統不需要架設感測器於輸電線上,使用時間、天氣資訊與台電提供的歷史負載潮流來訓練專家系統的知識庫,此專家系統在沒有負載潮流資訊下可以推算出全部的熱過載機率與輸電線導體溫度機率,此研究提出的系統可讓臺電調度人員做出更全面的規劃,除此之外,此系統具有適應性的知識庫,知識庫與推理引擎的建立是使用貝氏網路(Bayesian network),為了要訓練出貝氏網路,此研究提出了一個適應性的貝氏建模方法,此方法使用十等分離散化(Ten-bin discretization)每小時的將資料離散化,並使用最大最小登山結構學習法(Max-min hill-climbing structure learning)每三十天的更新網路。建立完貝氏網路後,台電調度人員可以使用拒絕取樣法(Rejection sampling)查詢熱過載風險。此專家系統的系統架構由FTP伺服器、C++撰寫的預處理程式、包含bnlearn套件的R環境與做為圖形化使用者介面的R studio所組成。根據台電調度準則規定,當輸電線導體溫度低於80 °C是安全等級,80 °C ~ 105 °C是緊急事件等級,而超過105 °C則是危險等級。此專家系統用機率呈現熱過載風險,經過2012年2月1日至2013年2月1日資料的驗證測試結果,以最高機率的輸電線導體溫度區間來看,系統的平均命中率(Hit ratio)為0.750214,而標準差為0.022464,此結果表現出本研究提出的系統在預測輸電線導體溫度與評估熱過載風險上有著高準確度與穩定性。 For power systems, the safety of transmission lines is important. There are many factors that lead to the damage of power systems. In addition to natural disasters, the line sag problem puts power systems into a dangerous situation. If the line sag problem does occur, the lines reaching the ground or higher trees in the mountain area will cause power transmission failure, which may bring huge economic loss. To avoid or solve the line sag problem, in smart grid applications, sensors are expected to sense the conductor temperature in real time through a wireless communication network. However, this solution is impractical, because the cost of the installation of thermal sensors on line conductors is very high. On the other hand, Taipower avoids the line sag problem by setting a threshold to restrict the operation electrical current to a safe range. However such a method cannot guarantee that the transmission lines will not experience line sag. Thus, Taipower sends many people to check power grids, including line conductors and towers every half of a year. This method is time consuming and costly, because it is very difficult for Taipower maintenance worker to reach these towers. The risk of using the inspection method is high. For Taipower dispatchers to do EHV grid dispatch planning, this study provides a real-time thermal overload risk evaluation expert system which does not need to install thermal sensors on line conductors. The proposed expert system uses time information, weather information and the historical power flow provided by Taipower to train the knowledge base of the proposed system. Then, without the data of power flow, the proposed expert system can still estimate all of the thermal overload probabilities and conductor temperature probabilities. The proposed system allows Taipower dispatchers to make comprehensive plans. Moreover, the system also has an adaptive knowledge base. The knowledge base and reasoning engine is constructed by a Bayesian network. To train the Bayesian network, this study proposes an adaptive Bayesian modeling method, which uses a ten-bin discretization method to discretize the data every hour and a max-min hill-climbing (MMHC) structure learning algorithm to update the network every 30 days. After the Bayesian network is established, Taipower dispatchers can query the thermal overload risk by rejection sampling. The expert system is composed of an FTP server, a C++ preprocessing program, the R environment including the bnlearn package, and the R studio as the graphical user interface. According to the Taipower dispatch regulation, conductor temperature below 80 °C is at a safe level, between 80 °C ~ 105 °C is at an emergency level, and above 105 °C is at a dangerous level. The expert system presents the thermal overload risk by probabilities. The test results for system validation shows that the average hit ratio is 0.750214 and the standard deviation is 0.022464 for the conductor temperature interval with the highest probability, using the data from Feb. 1, 2012 to Feb. 1, 2013. The results also show that the accuracy and stability of using the proposed system are high in predicting the conductor temperature and evaluating the thermal overload risk. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/53011 |
全文授權: | 有償授權 |
顯示於系所單位: | 生物機電工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-104-1.pdf 目前未授權公開取用 | 1.89 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。