摘要
提出一种基于四分位(QM)-具有噪声的基于密度聚类法(DBSCAN)与双向长短期记忆网络(BiLSTM)的风电机组故障预警方法。首先,针对风速-功率图中限功率点难以清洗完全的问题,提出利用QM与DBSCAN联合来对建模运行数据进行预处理;其次,通过分析风电机组运行原理,并结合轻量梯度提升机(LightGBM)特征选择法确定风电机组正常工况预测模型的输入输出参数,并基于BiLSTM建立了高精度的风电机组正常性能预测模型;之后,利用滑窗算法构建了风电机组状态性能评价指标,并通过统计学区间估计法确定指标阈值;最后,采用风电机组真实故障数据,开展风电机组异常工况预警实验,验证了方法的有效性。
A wind turbine fault warning method based on quartile method(QM)-density-based spatial clustering of applications with noise(DBSCAN)and Bi-directional long and short-term memory network(BiLSTM)is proposed.Firstly,in view of the difficulty of cleaning the power limit point in the wind speed-power diagram,the combination of QM and DBSCAN is proposed to preprocess the modeling operation data.secondly,by analyzing the operation principle of wind turbine and determining the input and output parameters of the normal working condition prediction model of wind turbine combined with LightGBM feature selection method,a high-precision normal performance prediction model of wind turbine is established based on BiLSTM.Then,the state performance index of the fan is determined by the sliding window algorithm,and the index threshold is determined by statistical interval estimation method.Finally,the real fault data of the fan is used to carry out the early warning experiment of the abnormal working condition of the whole wind turbine,which verifies the effectiveness of the method.
作者
马良玉
梁书源
程东炎
耿妍竹
段新会
MA Liangyu;LIANG Shuyuan;CHENG Dongyan;GENG Yanzhu;DUAN Xinhui(Department of Automation,North China Electric Power University,Baoding,Hebei 071003,China;Baoding Key Laboratory of State Detection and Optimization for Integrated Energy System,Baoding,Hebei 071003,China;Baoding SinoSimu Technology Co.Ltd,Baoding,Hebei 071000,China)
出处
《计量学报》
CSCD
北大核心
2024年第9期1384-1393,共10页
Acta Metrologica Sinica
基金
国家自然科学基金(61973117)
河北省中央引导地方科技发展资金(226Z2103G)。
