






















摘" 要: 發電機作為風電機組中的關鍵部件,其性能的優劣直接影響著風電場的效益和電能轉換的穩定。為監測風電機組發電機異常狀態,減少故障率并提高發電效率,通過對其SCADA歷史數據分析,提出一種基于數據驅動的風電機組發電機異常預警方法。首先針對SCADA中海量數據,用改進ReliefF特征算法(SIG?ReliefF)篩選出用于識別與目標變量(在這種情況下可能是發電機故障)具有最強關聯性的多個特征參數,這種方法的優勢在于能夠有效考慮到特征之間的相關性,最大程度地保留發電機故障相關特征與交互特征;然后建立MSET狀態參數預測模型,通過滑動窗口法對殘差的分布進行統計,從而確定故障閾值;最后通過實例驗證了所提方法的有效性與準確性,并通過與BPNN和SVM算法進行對比,驗證了其具有更優的異常預警性能。
關鍵詞: 風電機組; 發電機; ReliefF; MSET; 異常預警; 特征交互; 標準化交互增益; 滑動窗口
中圖分類號: TN911.23?34" " " " " " " " " " " " " 文獻標識碼: A" " " " " " " " " " " 文章編號: 1004?373X(2025)04?0091?06
Method of wind turbine generator abnormal warning based on
improved ReliefF?MSET algorithm
SHEN Xu1, WANG Haiyun1, DU Xin2, HUANG Xiaofang2
(1. Engineering Research Center of Education Ministry for Renewable Energy Power Generation and Grid Connection, College of Electrical Engineering, Xinjiang University, Urumqi 830047, China; 2. Beijing Gold Wind Science and Creation Wind Power Equipment Co., Ltd., Beijing 100176, China)
Abstract: As a key component in wind turbines, the performance of generators can directly affect the efficiency of wind farms and the stability of electric energy conversion. In order to monitor the abnormal states of the generator in wind turbines, reduce the failure rate, and improve power generation efficiency, a data?driven based method wind turbine generator abnormal warning is proposed by analyzing the historical SCADA data. The improved ReliefF feature algorithm (SIG ReliefF) is used to identify multiple feature parameters with the strongest correlation with the target variable (which may be a generator fault in this case) for massive data in SCADA. The advantage of this method is that it can effectively take into account the correlation between the features and preserve the generator fault correlation features and interaction features to the greatest extent. An MSET state parameter prediction model is established, and the distribution of residuals is analyzed statistically by means of sliding window approach to determine fault thresholds. The effectiveness and accuracy of the proposed method were validated by examples, and compared with BPNN and SVM algorithms, demonstrating its superior performance in anomaly warning.
Keywords: wind turbine; generator; ReliefF; MSET; abnormal warning; feature interaction; standardized interaction gain; sliding window
0" 引" 言
根據中國國家能源局最新統計數據顯示,截至2022年10月底,全國風電裝機容量約為3.5×108" kW,同比2021年增長16.6%。Wood Mackenzie于2023年1月在“2022 in review for offshore wind”中提出:到2030年,全球海上風電產業將實現5倍增長[1]。盡管裝機熱潮已經過去,但在后期機組的運維時代,如何保證機組的安全穩定運行,降低故障率,減少運維成本,并提高發電效率是需要長期關注的問題。風電機組主要包括葉片、齒輪箱、發電機、變槳系統、偏航系統等關鍵部件[2]。其中,發電機作為風電機組產生電能的關鍵部件,其較高的故障率一直是影響風電機組可靠性與穩定性的重要問題。……