摘要
针对风电机组叶片覆冰影响机组运行安全和降低发电量的问题,提出一种基于极端梯度提升算法和麻雀搜索算法优化卷积神经网络的风电机组叶片覆冰诊断方法。首先,利用基于极端梯度提升算法计算实际机组监控和数据采集系统(SCADA)数据的特征权重,筛除冗余特征变量,降低诊断模型的复杂度、减少诊断时间;再利用卷积神经网络模型对筛选后SCADA数据进行特征提取建立叶片覆冰诊断分类模型;最后,利用麻雀搜索算法对诊断模型中的超参数寻优,提高诊断模型的准确率。实验结果表明提出的方法对叶片覆冰的诊断准确率达到98%,相比于长短期记忆网络、K近邻算法等分类模型诊断准确率更高。
Wind turbine blade icing poses significant risks to operational safety and energy production efficiency.To address this challenge,this study proposes an ice detection framework combining XGBoost feature selection with a Sparrow Search Algorithm(SSA)-optimized convolutional neural network(CNN).First,XGBoost evaluates feature importance in SCADA system data to eliminate redundant variables,streamlining model architecture and accelerating diagnostic processes.A CNN then extracts discriminative features from this refined dataset to establish an ice accretion classification system.The SSA subsequently optimizes critical hyperparameters in the CNN model to enhance detection precision.Experimental validation demonstrates 98%diagnostic accuracy for blade icing conditions,outperforming both Long Short-Term Memory networks(97.2%)and k-Nearest Neighbors classifiers(95.1%).This integrated approach provides a reliable solution for real-time ice monitoring in wind farms.
作者
邢作霞
张玥
郭珊珊
张超
Xing Zuoxia;Zhang Yue;Guo Shanshan;Zhang Chao(School of Electrical Engineering,Shenyang University of Technology,Shenyang 110870,China;Liaoning Key Laboratory of Wind Power Generation Technology,Shenyang 110870,China)
出处
《太阳能学报》
北大核心
2025年第3期661-667,共7页
Acta Energiae Solaris Sinica
基金
辽宁省“兴辽英才计划”项目(XLYC2008005)。
