作者：乔福宇, 马良玉, 马永光

作者单位：华北电力大学自动化系，河北 保定 071003

关键词：风电机组;故障预警;神经网络;状态识别;功率曲线

摘要：

为尽早发现风电机组早期故障，减少风电场的运维成本，提出一种基于功率曲线分析与神经网络的故障预警方法。首先，依据功率特性曲线，使用最小二乘与离散度分析结合的算法对SCADA数据进行筛选，以保留符合机组正常工况的数据作为建模的训练数据，从而提高模型的精度。然后，使用随机森林算法筛选模型的输入参数以简化模型结构，并通过对比隐含层的数目建立风电机组的神经网络模型。最后，结合滑动窗口模型，提出一种能反映机组实时运行状态的指标，并通过非参数估计法确定该指标的阈值，以实现状态预警及在线监测。该方法充分利用SCADA数据，且不需要对风电机组复杂的物理特性进行分析。将该方法分别用于某风电场的变桨系统和偏航系统的故障预警，实验结果是分别提前18.5 h和28.4 h出现预警信号，进一步证明方法的有效性。

Wind turbine fault early warning method based on power curve analysis and neural network
QIAO Fuyu, MA Liangyu, MA Yongguang
Department of Automation, North China Electric Power University, Baoding 071003, China
Abstract: In order to detect the early fault of wind turbine as soon as possible and reduce the operation and maintenance cost of wind farm, a fault early warning method based on power curve analysis and neural network is proposed. First of all, according to the power characteristic curve, the algorithm of least square and dispersion analysis is used to filter the SCADA data to retain the data in accordance with the normal working conditions of the unit as the training data for modeling, so as to improve the accuracy of the model. After that, the random forest algorithm is used to filter the input parameters of the model to simplify the model structure, and the neural network model of wind turbine is established by comparing the number of hidden layers. Finally, combined with the sliding window model, an index which can reflect the real-time running state of the unit is proposed, and the threshold of the index is determined by non-parametric estimation method to realize state early warning and on-line monitoring. This method makes full use of SCADA data and does not need to analyze the complex physical characteristics of wind turbines. The method is applied to the fault early warning of pitch control system and yaw system in a wind farm, and the experimental results show that the early warning signals appear 18.5 h and 28.4 h in advance, respectively, which further proves the effectiveness of the method.
Keywords: wind turbine;fault early warning;neural network;state identification;power curve
2020, 46(8):44-50  收稿日期: 2020-05-29;收到修改稿日期: 2020-06-22
基金项目:
作者简介: 乔福宇(1993-)，男，河北承德市人，硕士研究生，专业方向为风电机组状态预警及故障诊断
参考文献
[1] 龙霞飞, 杨苹, 郭红霞, 等. 大型风力发电机组故障诊断方法综述[J]. 电网技术, 2017, 41(11): 3480-3491
[2] 向健平, 凌永志, 詹俊, 等. 基于SCADA系统的风电机组主轴承故障预警方法[J]. 电力科学与技术学报, 2019, 34(3): 223-228
[3] 王爽心, 郭婷婷, 李蒙. 风电机组变工况变桨系统异常状态在线识别[J]. 中国电机工程学报, 2019, 39(17): 5144-5152, 5295
[4] 黄小光, 潘东浩, 史晓鸣, 等. 风电机组齿轮箱系统的故障预警研究[J]. 机电工程, 2019, 36(6): 574-578
[5] XIAN Y W, COURT R, JIANG J S. Wind turbine condition monitoring by the approach of SCADA data analysis[J]. Renewable Energy, 2013, 53: 365-376
[6] SINGH S, BHATTI T, KOTHARI D. Wind power estimation using artificial neural network[J] Journal of Energy Engineering 2007, 133(1): 46-52.
[7] 刘永前, 邵振州, 王铮, 等. 基于分段支持向量机的风电机组理论功率计算研究[J]. 太阳能学报, 2019, 40(3): 673-680
[8] 李航涛, 郭鹏, 杨锡运. 基于离散度分析的风电机组功率曲线绘制方法研究[J]. 太阳能学报, 2019, 40(1): 237-241
[9] ZHOU Z H. Ensemble methods: foundations and algorithms [M]. London: Chapman & Hall/CRC, 2012.
[10] 李俊卿, 李秋佳, 石天宇, 等. 基于数据挖掘的风电功率预测特征选择方法[J]. 电测与仪表, 2019, 56(10): 87-92
[11] 刘增里, 杨静, 刘亚林, 等. 基于改进BP神经网络的风电功率超短期预测[J]. 船舶工程, 2019, 41(S1): 282-287
[12] 龚妙, 李录平, 刘瑞, 等. 基于运行参数特征的风力机叶片覆冰诊断方法[J]. 动力工程学报, 2019, 39(3): 214-219
[13] 张帆, 刘德顺, 戴巨川, 等. 一种基于SCADA参数关系的风电机组运行状态识别方法[J]. 机械工程学报, 2019, 55(4): 1-9