相似分群方法在風場風機故障檢測的應用研究

Study on the Application of Similar Clustering Approach in Wind Turbine Fault Detection

10.6342/NTU201703624

謝佩鈞

風力發電機 ； 相似分群方法 ； 高斯混合模型 ； Wind Turbines ； Similar Clustering Approach ； Gaussian Mixture Model

國立臺灣大學工程科學及海洋工程學系學位論文

2017年

碩士

蔡進發

繁體中文

本研究以相似分群方法建立風場中風機故障檢測與性能評估之流程，並分為風場集群分析與風場風機性能評估與異常偵測兩部分進行。其中風場集群分析利用每部風機量測的風速與風向資料，透過高斯混合模型建立風況模型後以信心值量測彼此間的相似性，並利用聚合階層式分群法分群。再根據分群結果針對同群内風機的功率、轉子轉速、葉片旋角與偏航誤差之資料，同樣透過高斯混合模型與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。 　　本研究利用台電彰濱風場23部風機資料進行分析，結果顯示風場在這季中的風況可分為風向偏北方的群集一以及風向偏東北方的群集二。並且在群集一中22號風機異常發生次數最多而7號風機無偵測出異常；群集二則是4號風機最多異常而12號風機無偵測出異常。同時比較各項參數之信心值發現，同群中有風機葉片旋角信心值在0.6以下者大多功率信心值也低於0.5；而轉子轉速與葉片旋角無異常發生，但偏航誤差信心值較低者功率信心值也低於0.5。

A method of fault detection based on similar clustering approach for the wind turbines within a wind farm is proposed in this study. The SCADA datas were used for the clustering and fault detection. The wind speed and wind direction were used for the clustering approach. The Gaussain distribution was derived for each turbine with the parameters of wind speed and wind direction. Then the confidence values (CV) were calculated between turbines based on the Gaussian distribution of wind speed and direction. The turbines were clustered by the hierarchical clustering method with the value of 1-CV. Then the Gaussian distributions of power, rotor speed, pitch angle of balde and yawing misalightment were calculated within the same cluster. The confidence value of these parameters between turbines were used to figure out the abnormal operation turbine. The SCADA datas of 23 wind turbines of Taipower which is located in Chang Hua Costal Industry Park were used for the analysis in this study. The turbines can be clustered into two groups. The first group has the wind direction from the north and the second group has the wid direction from northeast based on the wind datas from SACDA. The fault analysis shows that the No. 22 wind turbine of group 1 had most abnormal operations and the No.7 wind turbine of group 1 had no abnormal operation. It also shows that the No. 4 wind turbine of group 2 had most abnormal operations and the No. 12 wind turbine of group 2 operated smoothly without any abnormal operation. It is found that the confidence value of power is smaller than 0.5 when the confidence value of pitch angle is smaller than 0.6. And the confidemce value of power is low when the confidence value of the yawing misalightment is smaller than 0.5.

1. [2] "IEA 中期再生能源市場報告 2016," 工業技術研究院, 2016
   連結：
2. [5] Hernandez, Wilmar, José Luis López-Presa, and Jorge L. Maldonado-Correa. "Power Performance Verification of a Wind Farm Using the Friedman’s Test." Sensors 16.6 (2016): 816..
   連結：
3. [7] Jain, Anil K., M. Narasimha Murty, and Patrick J. Flynn. "Data clustering: a review." ACM computing surveys (CSUR) 31.3 (1999): 264-323.
   連結：
4. [8] Lapira, E., Brisset, D., Ardakani, H. D., Siegel, D., & Lee, J. "Wind turbine performance assessment using multi-regime modeling approach." Renewable Energy 45 (2012): 86-95.
   連結：
5. [9] 楊其昌. "高斯混合模型在風機預兆式健康管理上的應用研究." 臺灣大學工程科學及海洋工程學研究所學位論文 (2016): 1-95.
   連結：
6. [10] Peng, Xuan, Wang Xu, and Bingxi Wang. "Speaker clustering via novel pseudo-divergence of Gaussian mixture models." Natural Language Processing and Knowledge Engineering, 2005. IEEE NLP-KE'05. Proceedings of 2005 IEEE International Conference on. IEEE, 2005.
   連結：
7. [12] 詹勳智. "類神經網路在風機預兆式健康管理上的應用研究." 臺灣大學工程科學及海洋工程學研究所學位論文 (2016): 1-71.
   連結：
8. [13] 葉柏廷. "自組織映射圖在風機預兆式健康管理上的應用研究." 臺灣大學工程科學及海洋工程學研究所學位論文 (2016): 1-67.
   連結：
9. [15] Reynolds, Douglas. "Gaussian mixture models." Encyclopedia of biometrics(2015): 827-832.
   連結：
10. [19] Aho, J., Buckspan, A., Laks, J., Fleming, P., Jeong, Y., Dunne, F., ... & Johnson, K. (2012, June). "A tutorial of wind turbine control for supporting grid frequency through active power control." In American Control Conference (ACC), 2012 (pp. 3120-3131). IEEE.
    連結：
11. [20] Cha, Sung-Hyuk. "Comprehensive survey on distance/similarity measures between probability density functions." City 1.2 (2007): 1.
    連結：
12. [21] 李維平, & 張加憲. (2013). "使用 N 組連結平均法的階層式自動分群." 電子商務學報, 15(1), 35-56.
    連結：
13. [1] "Global Wind Report," GWEC, 2013
14. [3] "2016年能源產業技術白皮書第肆章　新及再生能源," 經濟部能源局, 2016
15. [4] Lapira, Edzel R." Fault detection in a network of similar machines using clustering approach." Diss. University of Cincinnati, 2012.
16. [6] Andu, Ashok Kumar Thavani, and Antony Selvdoss Thanamani. "Multidimensional Clustering Methods of Data Mining for Industrial Applications."
17. [11] Uluyol, O., Parthasarathy, G., Foslien, W., & Kim, K."Power curve analytic for wind turbine performance monitoring and prognostics." Annual conference of the prognostics and health management society. Vol. 2. 2011.
18. [14] 葉秉豐. (2012). "基於高斯混合模型之特徵補償應用於噪音強健性語音辨識." 國立中山大學
19. [16] Bhat, Harish S., and Nitesh Kumar. "On the derivation of the Bayesian Information Criterion." School of Natural Sciences, University of California(2010).
20. [17] Ester, M., Kriegel, H. P., Sander, J., & Xu, X. (1996, August). A density-based algorithm for discovering clusters in large spatial databases with noise. In Kdd(Vol. 96, No. 34, pp. 226-231).
21. [18] Pao, Lucy Y., and Kathryn E. Johnson. "Control of wind turbines." IEEE Control Systems 31.2 (2011): 44-62.
22. [22] Petersson, Andreas, and Torbjörn Thiringer. "Control of a Variable-Speed Pitch-Regulated Wind Turbine." Chalmers University of Technology, 2005..
23. [23] S. Pedersen, Gil Marin(2016, April). "Yaw Misalignment and Power Cuve Analysis." EWEA analysis of operating wind farms 2016 . .
24. [24] VESTAS. 'VESTAS V80-2.0MW 2000 80.0."' VESTAS 2000