基于改进KNN回归算法的风电机组齿轮箱状态监测

针对风电机组齿轮箱的状态监测问题,提出使用改进KNN回归算法建立齿轮箱的正常行为模型。首先,对经典KNN回归算法的距离度量公式进行改进,实验证明预测精度提高约60%;其次,基于改进的离群点和相似点剪辑算法优化KNN回归算法的训练集以提升运算效率,优化后计算时间缩短约20%,预测精度基本保持不变。最后,针对某风电场一台2 MW风电机组的齿轮箱实际故障数据,应用提出的改进KNN回归算法并结合统计过程控制相关理论,实现对齿轮箱故障的预警。结果表明:较经典KNN算法,提出的改进算法故障预警能力显著增强。

Condition monitoring of wind turbine gearbox based on improved KNN regression algorithm

Aiming at wind turbine gearbox condition monitoring, the normal behavior model of gearbox is established by using improved KNN regression algorithm. Firstly, the distance measurement formula of the classical KNN regression algorithm is improved, and the experiment proves that the prediction accuracy is improved by about 60%. Secondly, the training set of KNN regression algorithm is optimized based on the improved outlier and similarity clipping algorithms to improve the operational efficiency. The experiment proves that the optimized calculation time is reduced by about 20% and the prediction accuracy remained basically unchanged. Finally, based on the actual fault data of a gearbox of a 2 MW wind turbine, the improved KNN regression algorithm proposed and the statistical process control theory are applied to realize the early warning of gearbox faults. The result shows that compared with the classical KNN algorithm, the improved algorithm proposed has enhanced fault warning ability.