针对风电齿轮箱轴承故障预警中阈值依赖人工经验设定且更新不灵活的问题,提出一种基于指数加权移动平均与双边漂移流峰值超阈值的自适应阈值算法（EWMA-Bi-DSPOT）。该方法分为两阶段：首先,利用指数加权移动平均对原始温度序列进行平滑,以抑制高频噪声;其次,在初始化阶段选取高位分位数作为初始阈值,对超出量进行广义帕累托分布拟合,并通过极大似然估计获得初始报警阈值;最后,在线阶段,算法持续吸收未触发报警的边缘极值,动态更新GPD参数与报警阈值,从而实现阈值随系统状态漂移的动态更新。实验结果表明,EWMA-Bi-DSPOT算法能实现阈值的动态自适应更新,在保证较低误报率与漏报率的前提下有效提升故障预警的实时性与可靠性。
Abstract
To address the problem that threshold setting in fault early warning of wind turbine gearbox bearings relies heavily on empirical experience and lacks adaptability during long-term operation, an adaptive threshold algorithm based on Exponentially Weighted Moving Average and Bilateral Drifting Stream Peaks-Over-Threshold （EWMA-Bi-DSPOT） is proposed. In the proposed method, the bearing temperature time series collected from the SCADA system is first smoothed using an Exponentially Weighted Moving Average （EWMA） to suppress high-frequency noise and short-term fluctuations, thereby improving the stability of subsequent extreme value modeling. During the initialization stage, a high-level quantile is selected as the initial threshold, and the exceedances above this threshold are modeled using a Generalized Pareto Distribution （GPD）. The initial alarm threshold is then determined through maximum likelihood estimation of the GPD parameters. In the online stage, the algorithm continuously incorporates marginal extreme values that do not trigger alarms to update the GPD parameters, enabling dynamic adjustment of the alarm threshold to track gradual changes in system operating conditions. This mechanism allows the threshold to evolve with system state drift while preventing abnormal data from contaminating the model during alarm periods. Experimental results based on real wind turbine SCADA data demonstrate that the EWMA-Bi-DSPOT algorithm can achieve adaptive threshold updating for bearing state detection. Compared with traditional fixed-threshold methods, the proposed approach improves the timeliness and reliability of fault early warning while maintaining low false alarm and missed detection rates.
关键词
风电机组 /
齿轮箱 /
状态监测 /
阈值自适应 /
故障预警
Key words
wind turbines /
gearbox /
condition monitoring /
threshold adaptive method /
fault early warning
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 陈图南, 李文萱, 李云, 等. “双碳” 背景下碳-绿证-电力市场耦合机制综述[J]. 广东电力, 2024, 37(8): 14-25.
Chen T N, Li W X, Li Y, et al.Review on coupling mechanism of carbon, green certificate and power market under the background of carbon peaking and carbon neutrality[J]. Guangdong Electric Power, 2024, 37(8): 14-25.
[2] 杨耀攀. 电力行业碳中和的实施方式和障碍[J]. 中国电力企业管理, 2021(1): 53-57.
Yang Y P.Implementation mode and obstacles of carbon neutrality in power industry[J]. China Power Enterprise Management, 2021(1): 53-57.
[3] 黄志军. 风电新能源的现状与发展策略[J]. 电子技术, 2023, 52(1): 292-293.
Huang Z J.Analysis of current situation and development strategy of wind power new energy[J]. Electronic Technology, 2023, 52(1): 292-293.
[4] 郭鹏, David Infield, 杨锡运. 风电机组齿轮箱温度趋势状态监测及分析方法[J]. 中国电机工程学报, 2011, 31(32): 129-136.
Guo P, Infield D, Yang X Y.Wind turbine gearbox condition monitoring using temperature trend analysis[J]. Proceedings of the CSEE, 2011, 31(32): 129-136.
[5] Peng X G, Li H H, YUAN F, et al.An extreme learning machine for unsupervised online anomaly detection in multivariate time series[J]. Neurocomputing, 2022, 501: 596-608.
[6] Abhaya A, Patra B K.An efficient method for autoencoder based outlier detection[J]. Expert Systems with Applications, 2023, 213: 118904.
[7] 顾冬, 郁怀光, 吴宇, 等. 基于相似测点残差信息挖掘的大坝变形异常值识别方法研究[J]. 水力发电, 2023, 49(4): 102-106.
Gu D, Yu H G, Wu Y, et al.Outlier identification based on residual information mining of similar measurement points for dam deformation[J]. Water Power, 2023, 49(4): 102-106.
[8] Luo Z H, Fang C Y, Liu C L, et al.Method for cleaning abnormal data of wind turbine power curve based on density clustering and boundary extraction[J]. IEEE Transactions on Sustainable Energy, 2022, 13(2): 1147-1159.
[9] 向玲, 高鑫, 姚青陶, 等. 基于视觉Transformer多模型融合的风电机组异常状态监测[J]. 太阳能学报, 2025, 46(4): 522-529.
Xiang L, Gao X, Yao Q T, et al.Abnormal condition monitoring of wind turbine based on vision transformer multi-model fusion[J]. Acta Energiae Solaris Sinica, 2025, 46(4): 522-529.
[10] Jonathan P, Randell D, Wadsworth J, et al.Uncertainties in return values from extreme value analysis of peaks over threshold using the generalised Pareto distribution[J]. Ocean Engineering, 2021, 220: 107725.
[11] Mishra V K, Rajagopalan A.A novel extreme value theory based approach to hyperparameter optimization[J]. Procedia Computer Science, 2023, 218: 2411-2419.
[12] WüThrich M V. Bivariate extension of the pickands-balkema-de haan theorem[J]. Annales de l’Institut Henri Poincare (B) Probability and Statistics, 2004, 40(1): 33-41.
[13] Cambron P, Masson C, Tahan A, et al.Control chart monitoring of wind turbine generators using the statistical inertia of a wind farm average[J]. Renewable Energy, 2018, 116: 88-98.
[14] Grimshaw S D.Computing maximum likelihood estimates for the generalized Pareto distribution[J]. Technometrics, 1993, 35(2): 185-191.
[15] Yu Z X, Qin L, Chen Y J, et al.Stock price forecasting based on LLE-BP neural network model[J]. Physica A: Statistical Mechanics and is Applications, 2020, 553: 124197.
[16] 许子昂. 基于多参数融合的风机故障预警及发电性能评估[D]. 吉林: 东北电力大学, 2021.
Xu Z A.Multi-parameter fusion-based wind turbine fault warning and power generation performance assessment[D]. Jilin: Northeast Dianli University, 2021.
[17] 张小田, 鄢盛腾, 周雪青, 等. 基于状态监测的风电机组主轴承早期故障预测方法[J]. 广东电力, 2012, 25(11): 6-9, 50.
Zhang X T, Yan S T, Zhou X Q, et al.Early stage failure forecast method for main bearing of wind turbine based on state monitoring[J]. Guangdong Electric Power, 2012, 25(11): 6-9, 50.
基金
吉林省自然科学基金（YDZJ202501ZYTS630）
PDF(1879 KB)
