一种基于lightGBM框架改进的GBDT风力发电机叶片开裂预测方法

风力发电机叶片开裂直接影响风力发电机运行,采用梯度提升决策树算法与基于lightGBM框架改进的梯度提升决策树算法对风力发电机叶片开裂进行预测。对比分析了预测准确度与可行性。基于lightGBM改进的梯度提升决策树算法分析的风力发电机运行数据得出的预测结果优于梯度提升决策树算法,且对于风力发电机叶片开裂预测准确度较高,并具有实用价值。同时该算法能够大幅降低样本中的无效数据,减少计算量。其独立特征合并能够使得划分点特征数量降低,提高风力发电机叶片开裂预测的准确性。最后,风力发电机叶片开裂预测实验结果表明,基于lightGBM改进的梯度提升决策树算法取得了更好的预测结果,计算量更小且能够准确预测风力发电机叶片开裂故障。

A Method for Predicting Blade Cracking of GBDT Wind Turbine Based on Improved LightGBM Framework

The blade cracking of wind turbine directly affects the operation of wind turbine. The GBDT(gradient boosting decision tree) algorithm and the improved GBDT algorithm based on LightGBM (light gradient boosting machine framework) were used to predict the blade cracking of wind turbine. A comparative analysis of the accuracy and feasibility of prediction was conducted. The results of wind turbine operation data analyzed by the improved GBDT algorithm based on lightGBM were better than those of the GBDT algorithm, which were characterized by higher accuracy and practical value for the prediction of wind turbine blade cracking. Meanwhile, the algorithm could greatly reduce the invalid data in the sample and the amount of calculation. The combination of independent features could reduce the number of features at dividing points and improve the accuracy of the prediction of wind turbine blade cracking. Finally, the experimental results of wind turbine blade cracking prediction showed that the improved GBDT algorithm based on lightGBM could achieve better prediction results with less computation and accurate prediction of wind turbine blade cracking fault.