基于梯度提升决策树的风电机组齿轮箱故障检测

Fault Detection Based on Gradient Boosting Decision Tree for Wind Turbine Gearbox

针对传统提升算法在处理风电机组运行大数据时,存在效率和准确度低、实时性差等问题,提出基于梯度提升决策树的风电机组齿轮箱故障检测方法。首先,结合专家经验并利用最大信息系数分析风电机组海量大数据之间的相关性,在此基础上进行特征选择得到合理的特征向量;其次,利用贝叶斯超参数优化算法,优化梯度提升决策树算法中关键参数,建立基于梯度提升决策树的故障检测模型;最后,将经过最大信息系数分析处理后的特征向量作为梯度提升决策树故障检测模型的输入值,将齿轮箱不同工况下类别作为输出值。实验结果表明,该方法具有更低的漏报率和误报率,具有更强的泛化能力,能实时检测故障,降低维护成本。

Aiming at the problems of traditional boosting algorithm in dealing with wind turbines running big data, such as low efficiency, low accuracy and poor real-time performance, a fault detection method for wind turbine gearbox based on gradient boosting decision tree is proposed. Firstly, combining the expert experience and using the maximum information coefficient,the correlation between massive data and large data of wind turbines is analyzed,and feature selection is made to obtain reasonable eigenvectors based on this. Secondly, the Bayesian hyperparameter optimization algorithm is used to optimize the gradient boosting decision tree. The key parameters in the gradient boosting decision tree algorithm are used to establish a fault detection model based on the gradient boosting decision tree. Finally, the maximum information coefficient analysis is used as the input value of the fault detection model of the gradient lifting decision tree, and the type of the gearbox is used as the output value. The experimental results show that compared with other algorithms, this method has a lower missed alarm rate and false alarm rate and stronger generalization ability, which can detect faults in real time and reduce maintenance costs.