摘要
提出一种基于Stacking多模型融合的风电机组齿轮箱异常状态监测方法。首先,以齿轮箱油池温度为输出特征,利用随机森林算法对清洗后的数据进行特征选择;随后,利用经随机搜索寻优后的多个算法建立单一预测模型,并以Pearson相关系数为指标计算各模型的预测误差相关性选取基学习器;所选算法经5折交叉验证得到的输出作为第2层元学习器的输入进行训练,建立Stacking预测模型;最后,采用滑动窗口方法计算状态评价指标,并根据区间估计理论设置的阈值进行故障预警。以实际风场的齿轮箱历史故障为例,验证Stacking模型的预警能力,表明该方法的有效性。
An anomaly warning method for wind turbine gearbox based on Stacking fusion of multiple models is proposed.Firstly,taking the gearbox oil pool temperature as the output feature,the random forest algorithm is used to select the features of the cleaned data.Subsequently,a single prediction model is established using multiple algorithms optimized through random search,and the Pearson correlation coefficient is used as an indicator to calculate the prediction error correlation of each model and select a base learner.The output obtained from the selected algorithm after 5-fold cross validation is used as the input of the second layer meta learner for training,and a Stacking prediction model is established.Finally,the sliding window method is used to calculate the state evaluation indicators,and fault warning is carried out based on the threshold set by interval estimation theory.Taking the historical faults of gearboxes in an actual wind farm as testing examples,the early warning ability of the Stacking model is verified,indicating the effectiveness of this method.
作者
马良玉
耿妍竹
梁书源
程东炎
段新会
MA Liangyu;GENG Yanzhu;LIANG Shuyuan;CHENG Dongyan;DUAN Xinhui(Automation Department,North China Electric Power University,Baoding 071003,Hebei Province,China;Baoding SinoSimu Technology Co.,Ltd.,Baoding 071000,Hebei Province,China)
出处
《中国电机工程学报》
北大核心
2023年第S1期242-251,共10页
PROCEEDINGS OF THE CHINESE SOCIETY FOR ELECTRICAL ENGINEERING
基金
河北省中央引导地方科技发展资金项目(226Z2103G)
