利用同儕比較方法進行風機故障之偵測

Abstract

本研究以台電某風場內15部風機冬季資料進行分析，基於同儕比較的概念對於風場中風機進行性能評估以及故障偵測，目地為找出造成風機性能低落的原因為何。性能評估部分透過隨機森林演算法對性能較佳的模範風機學習其風速、轉子轉速及葉片旋角對於發電功率的影響，藉此預測其他風機在不同操作條件下，理想發電功率為何，並計算實際功率與預測功率間的差值，藉由差值大小來評斷該部風機性能狀況，在資料範圍期間內，共偵測出3部風機有性能低落的情況發生。針對性能狀況較差的風機，將其各項參數透過與風場其他部風機互相比較的方法，找出導致性能狀況低落的故障原因為何。 在後續故障偵測部份對於各項參數故障與否提出了幾種偵測方法，透過時間序列的相似度分析判斷其風速計是否產生異常狀況；轉子轉速及葉片旋角部分，則透過模範風機擬和出一條參考曲線，檢視風機的轉子轉速與葉片旋角行為是否正常；偏航系統部分藉由風向、機艙角度以及風速功率做判斷。在各項參數皆為正常的狀態下，發電功率仍然較低落，則判斷為發電機發生異常情形。故障偵測結果中發現性能低落的3部風機主要故障原因為風速計異常及發電機異常，另外有6部風機偵測到風速計異常狀況產生，在整個風場內的15部風機不論風向計正常與否都有偏航系統或機艙角度計異常狀況發生。

To find out the reasons why the performance of some wind turbines was lower than others, the winter data of 15 wind turbines in a wind farm of Taipower were analyzed based on the concept of peer comparison in this study. The analysis included performance evaluation and fault detection. In the performance evaluation part, the relations between the superior wind turbines’ wind speed, rotor speed, pitch ratio and power were learned by the Random Forest Algorithm in order to predict the ideal power in different operating conditions of other wind turbines. The performance of a wind turbine was evaluated by the difference between the actual power and the predict power of the wind turbine. Three wind turbines were detected with the conditions of the low power performance. In the fault detect part, several detecting methods were used to detect whether the operating parameters of wind turbines were normal or not. The wind speed meters were analyzed by the similarity analysis of time series. The rotor speeds and pitch angles were analyzed by the fitting curve of these two of model wind turbine. The misalignment of wind turbine and yaw system were evaluated by the angle between wind direction and nacelle. If the three conditions above were normal of a low performance wind turbine, then it was attributed to the abnormal power output of the generator. The analysis shows that the main reasons of the three low power performance turbines are the abnomal of wind speed meter and generator. In addition, there were six wind turbines which had abnormal conditions of wind direction. And there were many yaw misalignments in all turbines of the wind farm.