基于RF-LightGBM算法在风机叶片开裂故障预测中的应用

针对SCADA系统采集的数据繁杂,难以从原始数据判别工作中风机叶片开裂状态的问题,提出了一种对风机叶片状态进行分类预测的随机森林(RF)算法与LightGBM算法结合的模型。首先对SCADA数据进行预处理,特征变换,采用RF算法对特征进行重要性排序;然后利用清洗后的数据训练该分类预测模型,利用K折交叉验证法对模型进行验证调优;最后用测试数据集对叶片状态进行预测,依靠F1-score指标对模型性能进行评价。实验结果表明,数据处理后,模型性能明显提高,较XGBoost与GBDT算法分别提高了11%、16%,与传统的叶片状态识别方法相比,该算法能够更加快速精准的在线预测出风机叶片开裂状态,为风电场对风机叶片状态监测检修提供更可靠的参考依据。

Application of RF-LightGBM algorithm in early warning of fan blade cracking

Aiming at the problem that the data collected using SCADA(Supervisory Control and Data Acquisition) system is so complicated that it is difficult to distinguish the cracking state of fan blades from the original data, an algorithm model combining RF(Random Forest, RF) and LightGBM(Light Gradient Boosting Machine) for predicting fan blade state is proposed. Firstly, SCADA data are preprocessed, feature transformed, and random forest algorithm is used to rank the importance of features. Then, the classified prediction model is trained by the cleaned data, and the model is Finally, the fault status of blades is predicted by test data set, and the performance of the model is evaluated with F1-score index. The experimental results show that the performance of the model is obviously improved after data processing, which is 11% and 16% higher than XGBoost and GBDT algorithms respectively. Compared with traditional blade state recognition methods, this algorithm can identify the cracking state of fan blades more quickly and accurately on-line, which provides a more reliable reference for wind farm to monitor and repair the condition of fan blades.