基于K近邻与支持向量机协同训练的风机叶片结冰早期检测

Early Detection of Icing on Wind Turbine Blades Based on Co-training of K-Nearest Neighbor and Support Vector Machine

由于复杂的运行环境与工作机理,风力机组中普遍存在叶片结冰现象,该现象在初期不易观测,使得数据收集过程中产生了一定的误判及标签缺失样本。针对上述问题,提出一种基于K近邻与支持向量机方法的叶片结冰早期检测方法,将半监督学习中的协同训练机制引入到检测方法中,对标签不确信与缺失样本进行补充与再学习。利用某风力发电厂的风机叶片结冰进行算法验证。结果表明,与现有方法相比该方法具有更高的检测准确度与灵敏度。

Due to the complex operating environment and working mechanism,blade icing is common in wind turbines.The icing phenomenon is not easy to be observed at the initial stage,it may cause certain misjudgments and missing labels during the data collection process.To address the above issue,an early detection method for blade icing based on the K-Nearest neighbor and support vector machine method is proposed.The co-training mechanism from semi-supervised learning is introduced into the fault detection method to supply the reuse of samples with missing or untrusted labels.The algorithm is verified by the icing fault data collected from a wind power plant.Experimental results show that this method has higher detection accuracy and sensitivity compared with existing methods.