應用同儕比較偵測性能低下風機及異常運作之研究

STUDY ON APPLYING THE PEER-TO-PEER COMPARISON METHOD TO DETECT THE LOW PERFORMANCE AND ABNORMAL OPERATION WIND TURBINE IN A WIND FARM

本研究採用同儕比較的方法來建立風場中性能低下的風機與操作異常之評估方法，方法分為風場分群分析與風場風機性能評估與異常偵測兩部分進行。其中風場分群分析利用每部風機量測的風速與風向資料，透過自組織映射網路（Self-Organizing Map, SOM）來建立風況模型後以信心值公式量測彼此間的相似性，再利用聚合階層式分群法分群。根據分群結果再分成兩類，第一類為有與其他風機形成同儕關係的風機，針對同群內風機的功率、轉子轉速、發電機轉速、葉片旋角與偏航誤差之資料，透過高斯混合模型（Gaussian Mixture Model）以信心值做相似性的測量，並逐日比對群內不同風機間行為是否相異，第二類為未與其他風機形成同儕關係的風機，則針對風向較特別的風機檢查其風向計是否有異常。本研究使用台電某風場23部風機進行資料分析，依風況分群結果風機被分成三群，第一群為風向正北偏東約10度的群集、第二群為風向較偏正北的群集。未與其他風機建立同儕關係的風機即為獨立群。第一群中，7、9、10及12號風機性能明顯比模範風機5號機性能差；第二群中6號風機表現最佳，13風機異常天數最多，且13號風機性能明顯比6號風機性能差，也應盡早做維護；第一群及第二群性能低下的風機9、10、12及13號機都是高風速起動，可能高風速起動的控制邏輯不佳或發電機性能有問題；獨立群方面，3、8、11、19、21及22號風機的風向有出現異常情況，其中以8號風機風向為正南風顯示風向計有問題，19號風機則是整體風向偏東北35度因此被歸類到獨立群，對獨立群風機而言可能其風向計需要檢查以確認其風向是否正確。

The Peer-to-Peer comparison method was used to construct an evaluation procedure to detect the low performance and abnormal operation wind turbine in a wind farm. The Self-Organizing Map algorithm was used to cluster the different wind turbine group based on the similarity of the wind speed and wind direction. There are two types of clustering, one is that there is more than one turbine in a group, and the other is that there is only one tur-bine which is called independent wind turbine. For a group of turbines, an exemplary wind turbine was selected. The exemplary wind turbine is defined as the root mean square error between the turbine output power and the guaranteed power curve provided by the turbine maker is minimum. Then, the Gaussian mixture model was used to model the power, rotor RPM, blade pitch angle with wind speed and the yaw angle between wind direction and nacelle angle. The confidence values of wind speed-power between the exemplary wind turbine and other wind turbines in the same group were calculated. The turbine was picked out when its power confidence value was smaller than 0.7. The other parameters were applied to make comparisons between the exemplary wind turbine and the low performance wind turbine to find out the possible causes of the low performance. The established procedure was applied to analyze the SCADA data from a wind farm of Taipower company. There were three groups after clustering the turbines of the wind farm. The first group have thirteen wind turbines which the wind direction is about 10 degrees east of north, the second group have three wind turbines which the wind direction is about north. The third group have 6 wind turbines which the wind direction has abnormal wind direction. In group 1, there are four wind turbines, 7, 9, 10, and 12 which had low performance when compared with exemplary wind turbine 5. In group 2, there is only one low performance wind turbine 13 when compared with exemplary wind turbine 6. All low performance wind turbines were started up at high wind speed which the wind speed is greater than 8m/s, except wind turbine 7. The low performance may due to the abnormal cutin control logic at high speed wind or the low performance output of the generator. The independent group have six wind turbines which are 3, 8, 11, 9, 21 and 22 wind turbines. The wind direction of wind turbine 19 is about 35 degrees east of north. The wind direction of wind turbine 8 was at south. The other wind turbines also have wind direction from south in the northeast wind season. That means the wind sensors should be checked for the independent group of wind turbines.