基于DBSCAN和SDAE的风电机组异常工况预警研究

DBSCAN and SDAE-based Abnormal Condition Early Warning for a Wind Turbine Unit

提出一种基于密度的聚类方法(DBSCAN)和堆栈式降噪自编码器(SDAE)结合的风电机组性能预测及异常运行工况预警方法。首先采用DBSCAN算法对机组监控与数据采集(SCADA)系统历史运行数据进行清洗,然后利用SDAE建立风电机组的正常运行性能预测模型。基于该模型,采用时移滑动窗口方法构建能准确反映风电机组异常状态的识别指标,并根据统计学区间估计理论合理确定指标阈值,以实现异常工况预警。采用某风电机组的真实历史运行数据进行故障重演试验。结果表明:该方法能够在故障发生前及时对风电机组的异常运行工况发出预警,验证了该方法的有效性。

A combination of density-based spatial clustering of applications with noise(DBSCAN)method and stacked denoising auto-encoder(SDAE)was proposed for wind turbine normal operating performance prediction and abnormal condition early warning.Firstly,the DBSCAN algorithm was used to clean the historical operation data of the wind unit SCADA system,then the normal operating performance prediction model was established with SDAE.Based on this model,the time shift sliding window method was adopted to construct the identification index,which can accurately reflect the abnormal state of the wind turbine,and the index threshold was determined according to the statistical interval estimation theory.The fault replay test was performed by using the real historical operation data of a wind turbine.Results show that this method can timely warn the abnormal operation conditions of wind turbines before the fault occurs,which verifies the effectiveness of this method.