基于LightGBM的风电机组齿轮箱油温故障预警研究

Research on fault warning of wind turbine gearbox based on lightgbm algorithm model

基于风电机组运维历史大数据,探索齿轮箱油温异常预测性预警及异常原因分析的方法。首先,基于齿轮箱油温正常状态的SCADA运维大数据,在特征工程中采用方差排序、Pearson相关系数和递归特征消除进行降维,产生三个不同特征组合的数据集,分别建立LightGBM模型并选出表现最优的数据集;其次,选取boosting另外两个流行算法XGBoost与CatBoost作为对照算法,从均方误差(MSE),拟合优度(R Squared)等多个评价指标进行综合评价;最后,通过比较齿轮箱的预测油温与真实油温的偏离程度,在偏离较大时启动预警。结果显示,LightGBM算法模型保证时效性的同时,误差值明显低于其他算法模型,预测齿轮箱油温的效果最佳。本文提出的模型系统能够提前对齿轮箱油温异常进行预警,提升风电场能源持续输出效率。

Based on the historical big data of wind turbine operation and maintenance,the method of gearbox oil temperature anomaly prediction and abnormal cause analysis was explored.Firstly,based on the SCADA operation and maintenance data of the normal state of gearbox oil temperature,three data sets with different feature combinations were generated by variance sorting,pearson correlation coefficient and recursive feature elimination for dimensionality reduction in feature engineering.The LightGBM model was established respectively and the best data set was selected.Secondly,the other two popular boosting algorithms XGBoost and CatBoost were selected as control algorithms to carry out comprehensive evaluation from MSE,R Squared and other evaluation indexes.Finally,by comparing the deviation between the predicted oil temperature and the real oil temperature of the gear box,the early warning is started when the deviation is large.The results show that the error value of LightGBM algorithm model is significantly lower than other algorithm models while ensuring timeliness,and the effect of oil temperature prediction of gearbox is the best.The model system proposed in this paper can give early warning to gearbox oil temperature anomalies in advance,and improve the sustainable energy output efficiency of wind farm.