- 學位論文
基於深度學習之超高壓輸電線安全裕度預測及調度策略擬定
Safety Margin Prediction and Ampacity Dispatch Planning for EHV Transmission Line Based on Deep Learning
摘要
隨著電力需求與日俱增，超高壓電網所傳輸的負載也越來越重，準確的負載預測能使電力調度者做出關鍵的決策。此外，由於再生能源的興起其電力併網時可能會衝擊目前的電源調度模式，隨著發電不確定性的增加，其重要性進一步提高。目前台灣電力是根據靜態熱容量(Static thermal rating, STR)為參考做調度，但這會顯得過於保守且不夠靈活有效益。近年來被認為可能解決這個問題的技術為動態熱容量(Dynamic thermal rating , DTR)。DTR利用天氣資訊來估算架空輸電線的載流容量，是協助智慧電網進行規劃與決策的有效工具，如果能夠預測出未來數小時的載流變化，不僅能在不犧牲輸電安全的情況下提升輸電效益並且能夠提早處理異常載流的情況發生。因此，本研究提出了三種不同的混合預測模型比較，分別為Recurrent neural network (RNN)、Long Short-Term Memory (LSTM) 與 Gated Recurrent Unit (GRU)，並搭配Extreme Learning Machine(ELM)選出其準確度較佳的混合模型作為安全裕度的預測模型，最後將其預測結果搭配本研究提出的電力調度策略應用於五種特別挑選出來的案例，並以輸電線垂度和敏感度分析來評估其可靠性。本研究結果證實，藉由提出的策略限制下能在安全無疑的增加輸電線的負載，更重要的是能夠應付多種可能會遇到的負載變動，並且藉由敏感度分析能夠找出敏感度相對較高的跨距，降低決策者陷入調度盲點使調整負載時發生不可逆的風險。
並列摘要
With the increasing power demand, the load transmitted by an ultra-high voltage grid is getting heavier. Accurate load forecasting enables power dispatchers to make critical decisions. In addition, the rise of renewable energy uses and the integration of renewable energy into the grid will affect the current power dispatch mode. As the uncertainty of power generation increases, the importance of renewable energy generation will also increase. At present, Taiwan Power Company uses static thermal rating (STR) results as reference for power dispatch, but such a dispatch strategy appears to be too conservative and not flexible enough for effective power dispatch. In recent years, dynamic thermal rating (DTR) has been considered as one of the technique that may solve this problem. DTR uses weather information to estimate the current carrying capacity of overhead transmission lines. It is an effective tool to assist smart grids in power dispatch planning and decision-making. If the current-carrying changes in the next few hours can be predicted, not only will the transmission be improved without sacrificing transmission safety, but also the anomalies can be detected and excluded early. Therefore, this research examines the performances of an Extreme Learning Machine (ELM) model and three different combined models that include both the ELM model and one of the three models (Recurrent neural network (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU)) in safety margin prediction. The combined model with higher accuracy is used as the prediction model of safety margin, and finally, based on the prediction results, power dispatch strategies are proposed by this research. The proposed strategies take into account the impact of weather conditions on the transmission line and the change in the forecast range after the load changes. The main purpose of the strategies is to increase the load of the transmission line under safe circumstances. More importantly, these strategies can overcome a variety of unpredictable load changes. Five specially selected cases are used to address these strategies, and the reliability of transmission line sag detection is examined, and a sensitivity analysis is conducted to verified the proposed strategies.