应用深度自编码网络和XGBoost的风电机组发电机故障诊断  

Fault Diagnosis of Wind Turbine Generator Based on Deep Autoencoder Network and XGBoost

作　　者：赵洪山[1] 闫西慧 王桂兰[1] 尹相龙 ZHAO Hongshan;YAN Xihui;WANG Guilan;YIN Xianglong(Hebei Key Laboratory of Distributed Energy Storage and Microgrid (North China Electric Power University),Baoding 071003,China)

机构地区：[1]分布式储能与微网河北省重点实验室(华北电力大学),河北省保定市071003

出　　处：《电力系统自动化》2019年第1期81-86,共6页Automation of Electric Power Systems

基　　金：国家科技支撑计划资助项目(2015BAA06B03)~~

摘　　要：针对风电机组现场故障样本难获取的问题,为实现风电机组发电机部件的故障诊断,通过分析风机监控与采集(SCADA)数据,设计了基于深度自编码(DAE)网络和XGBoost的故障诊断算法。该算法包含两部分:第一部分是DAE故障检测算法,通过DAE获取SCADA数据的重构值,分析重构误差的变化趋势与其超越阈值的情况以预测风机故障和提取故障样本;第二部分是XGBoost故障识别算法,用贝叶斯优化搜索XGBoost的最优超参数,建立XGBoost多分类故障识别模型。算例结果表明,DAE算法能够捕获风电机组发电机早期故障,XGBoost比其他算法更精确地识别不同故障类型。Aiming at the problem that wind turbine field fault samples are difficult to obtain and to realize the fault diagnosis of generator components of wind turbine generators,through the analysis of supervisory control and data acquisition(SCADA)data,a fault diagnosis algorithm based on deep autoencoder(DAE)network and XGBoost is designed.The algorithm consists of two parts.The first part is the DAE fault detection algorithm,which obtains the reconstructed values of the SCADA data through DAE and analyzes the trend of the reconstruction error and its situation beyond the threshold to predict a fault of wind turbine and to extract the fault samples.The second part is the XGBoost fault identification algorithm.By using Bayesian optimization to search the optimal hyper-parameters of XGBoost,an XGBoost multi-class fault identification model is established.The results of the example show that the DAE algorithm can capture the early fault of wind turbine generators,and XGBoost can identify different fault types more accurately than other algorithms.

关 键 词：风电场 风电机组 故障诊断 深度自编码 

分 类 号：TM315[电气工程—电机]