應用類神經網路風速模型估測超高壓輸電線溫:以峽谷地形為例

Abstract

超高壓輸電線電塔扮演著民生用電與工業用電的重要角色，超高壓輸電線的安全性被視為一重要的課題，其中，超高壓輸電線線溫是安全性一重要的指標，目前，多由IEEE Std. 738 輔以氣象資料估測超高壓輸電線線溫，經由文獻，已知風速為影響輸電線線溫最重要的因素。 由於估算超高壓輸電線路之風速資料為中央氣象局使用全台各地氣象觀測站之觀測資料代入STMAS-2D經由內插法產生2.5公里 2.5公里之氣象格點資料，此套方法可有效保留大範圍之氣象資訊，但並無考慮到台灣地形之高低起伏大，氣象格點之風速與真實風速容易產生誤差，透過估算過去五年之全台超高壓輸電線路線溫，發現在台灣山谷地區容易產生熱點，顯示此氣象格點風速資料無法有效代表山谷地形之真實風速，因此，本研究欲建立一適合台灣山谷地形知風速模型。近年來，風速模型研究蓬勃發展，目前大部分研究多使用類神經網路於風速模型中，其擁有優異處理非線性關係且可學習和自適應不確定系統的優點，因此，本研究使用人工類神經網路針對台灣之山谷地形設計ㄧ風速模型，利用中央氣象局提供之數十年風速資料，期望透過此模型能得到更真實的風速。 本研究將氣象格點資料代入本研究之風速模型產生風速與真實風速之RMSE為0.9880優於原氣象格點風速之RMSE值2.0318，使用本研究之風速模型風速估算超高壓輸電線溫，RMSE為8.8542優於使用氣象格點資料之RMSE值10.6003，期望在特定的地形（例如山谷地形），動態熱容量預測模型使用本研究之風速模型能得到更真實的線溫，提供台灣電力公司調度人員可靠參考依據。

EHV transmission lines play an important role in supplying industrial and household electricity. The safety of EHV transmission lines is seen as an important issue. The line temperature of EHV transmission lines is an important indicator of the safety of power transmission. In this study, the IEEE Std. 738 is combined with meteorological data to estimate EHV transmission line temperature. Based on previous studies, wind speed is found to be most important factor that affects the transmission line temperature. In this study, the wind speed data used to estimate the EHV transmission line temperature are provided by the Central Weather Bureau in Taiwan. The data observed by the weather stations are put into the STMAS-2D to generate meteorological grid data by using an interpolation method. This method does not take the impacts brought by topographic features into account, causing error between the meteorological grid of wind speed and true wind speed. Based on the temperature data of the EHV transmission lines in the past 5 years, the meteorological wind grid data cannot effectively represent the true wind speed in valleys. Therefore, this study proposes to develop a wind speed model for valleys in Taiwan. In recent years, research on wind speeds has used artificial neural networks as wind speed models, because artificial neural networks are able to deal with nonlinear and uncertain factors and do adaptive learning. Therefore, in this study an artificial neural network is used to develop a wind speed model for valleys in Taiwan. In this study, the meteorological grid data are incorporated with the developed wind speed model, and the RMSE for the new wind speed model is 0.9880, which is small than the RMSE for the original meteorological grid model (the RMSE is 2.0318). When using the new wind speed model to estimate the temperature of EHV transmission lines, the RMSE is 8.8542, which is also smaller than the RMSE when the original meteorological grid data are used for line temperature estimation (the RMSE is 10.6003). These results suggest that after taking terrain features into account, such as the wind speeds in valleys, the dynamic thermal rating forecasting model can generate more accurate line temperature estimation, and the temperature estimated by such a method can provide dispatchers in Taiwan power company reliable reference.