基于机器学习的风机叶片开裂预测研究

Prediction of Fan Blade Cracking Based on Machine Learning

为了预测风机叶片开裂的状态,使用机器学习的方法对风机叶片状态进行分类预测。首先对SCADA采集的原始数据进行预处理,然后采用逻辑回归与XGBoost集成学习算法对预处理后的数据进行建模,并通过性能度量的评价指标比较两种算法的效果与泛化能力。结果表明,XGBoost在风机叶片开裂的分类预测上有更好的效果,其预测准确率达到了97.31%,而逻辑回归预测的准确率只为69.05%,从而将XG-Boost集成学习算法用于精准预测风机叶片开裂的状态,为风电场对风机叶片状态检测提供了参考依据。另外为了提高模型训练的效率,使用嵌入式特征选择方法将430维数据降到100维,训练时间从67.06s降到13.50s,准确度从97.04%提升到97.65%。

In order to predict the condition of fan blade cracking, machine learning method is used to classify and predict the fan blade state.Firstly, the original data collected by SCADA is preprocessed, and then the data after preprocessing is modeled by using the integrated learning algorithm of logical regression and xgboost, and the effect and generalization ability of the two algorithm models are compared through the evaluation index of performance measurement.The results show that xgboost has a better effect on the classification and prediction of fan blade cracking, and the accuracy rate of logic regression is only 69.05%, while the accuracy rate of xgboost is 97.31%.Therefore, xgboost integrated learning algorithm can be used to quickly and accurately predict the state of fan blade cracking, which provides a reference for wind farm to detect the state of wind turbine blade.In addition, in order to improve the efficiency of model training, the embedded feature selection method is used to reduce 430 dimension to 100 dimension, and the training time is reduced from 67.06 s to 13.50 s, and the accuracy is also improved.