- 學位論文
西元2028年台灣電力系統極小化低頻卸載研究
Research on Minimizing Under Frequency Load Shedding for Taipower System in 2028
摘要
近年來，科技產業成長迅速，伴隨的是每年的電力需求上升，必須透過謹慎的機組調度才能安全度過夏季的用電尖峰。假使在供電吃緊的狀態下，系統發生不可預期的事故，使得機組基於安全切離系統，此時電力系統瞬間失去一部分的發電量，導致系統發生頻率驟降並引發停電事故。電力系統中的頻率代表著系統的供需平衡，若系統頻率太低會使系統上的機組發生連鎖反應連續跳脫，最終導致電網崩潰。若一套適當的電力系統防衛系統，能在發生事故時，提供正確的參數給電力調度員，能快速制定應對策略，例如:啟動低頻卸載、輔助服務與啟動快速反應的備轉機組，能夠有機會避免大規模的停電事故甚至發生系統全黑。台灣於815大潭停電事故與513興達停電事故例子，但這些都是觸動低頻卸載保護電驛才啟動的保護策略。本論文使用PSS/E模擬台灣電網的頻率資訊，偵測發電機跳機事故，預測頻率最低點與發電機組的跳機量，評估事故的嚴重性。 本論文採用基因演算法結合基於功率注入預測頻率最低點與跳機功率演算法，改善在濾波器與使用頻率資料區間上的調整困難，並減少演算法預測上的誤差。本論文使用西元2028年台灣尖峰電力系統進行模擬與討論。
並列摘要
The technology industry has grown rapidly in recent years. With Taiwan's annual demand load increasing, it needs to be more cautious in dispatching generating units to get through the peak load in summer. If the generator units fail and trip unexpectedly at the low spinning reserve, the power system frequency will decline quickly and cause a severe event. The frequency represents the supply and demand balance. If the frequency is too low, it will lead to the machine disconnecting from the grid. Finally, it led to a power system blackout. However, suppose a proper power system defense strategy can provide the system operator with the correct parameters. Then the system operator can develop an emergency strategy, like load shedding strategy, ancillary service, and fast response reserve service. It will be possible to prevent a blackout event, like 815 Datan trippings and 513 Shinda power plant tripping events in Taiwan. These cases trigger the under-frequency load shedding relay, which causes the interruption of service. The thesis uses PSS/E to simulate the Taipower system frequency data and use frequency data to detect the machine event, predict frequency nadir, and evaluate the event seriousness. In this thesis, the genetic algorithm is combined with the frequency-based power injection estimation via nadir prediction algorithm to improve the adjustment difficulty of the filter and the frequency data interval and reduce the prediction error. This thesis uses the peak load in the Taipower system in 2028 for simulation and discussion.