基于慢特征分析的风机多模型叶片结冰检测

A Multi-model Detection Method for Icing on Wind Turbine Blades with Slow Feature Analysis

风机在恶劣运行环境下常见叶片结冰故障,导致叶片变形,降低风力涡轮机效率,并影响电网的稳定性。由于风机的工作状态在不同的风速下非平稳动态变化,难以建立叶片结冰检测的全局模型,针对这种大范围非平稳变工况特性,我们提出了一种细粒度的状态分区算法。该方法将时间轴转变为风速轴,通过慢特征分析(Slow Feature Analysis,SFA)将整个样本划分为不同运行状态,然后为每个状态建立不同的子模型。建立子模型后,考虑到风电的动态特性,我们提出动态和静态指标相结合的监测方法。在线应用时,该方法可以通过风速快速地将每个新样本分配到不同的子状态模型进行精细监测。这种多模型划分与监测方法在风力机叶片结冰方面的实际应用中被证明快速有效。

Icing on wind turbine blade is commonly seen in bad running environment.And this may cause deformation of blade,lower wind turbine efficiency,and affect the stability of the power grid.The working state of the wind turbine changes dynamically under different wind speeds,which makes it difficult to establish a global model for blade icing detection.Taking the non-stationary feature and complicated operating state into consideration,we propose a fine-grained state partitioning algorithm.It partitions the whole samples into different states on the wind speed axis instead of on the time axis with slow feature analysis and then establishes different sub-models for each state.With the sub-models established,a monitoring method combined with the dynamic and static indicators is developed to fit the dynamic characteristics of wind power generation.For online application,the method can quickly assign each new sample to its own state by its wind speed.The multi-model detection method is proved to be fast and efficient in detecting icing on wind turbine blades.