摘要
风力发电机叶片出现结冰现象时若照常工作,不仅会影响经济效益,严重时还会直接损坏叶片等设备引发安全事故。为此提出一种使用KmeansSMOTE的数据平衡方法与应用结冰相关的机理构建新特征和RFECV-DT特征筛选算法相结合的特征工程互补的数据处理方式,之后采用卷积神经网络模型进行训练与预测。实验结果表明,在卷积神经网络模型中采用KmeansSMOTE算法比SMOTE算法准确率提升2.78%。模型采用特征工程时比不采用特征工程相比准确率高出4.77%。与KNN、SVM、LR这些传统模型相比,所有衡量指标均有提升且不存在过拟合现象。所提出的方法,可解决应用SMOTE插值机制所带来的不足并且对特征工程进行精细化设计,也为风机叶片结冰故障诊断问题提供一种新的解决思路。
If the blades of wind turbines remains working in icing condition,they may damage directly,cause safety accident and affect the economic efficiency of power generation.In this paper,a data balancing method using K-means clustering algorithm synthetic minority oversampling technique(KmeansSMOTE)is proposed.The icing related mechanism is applied to construct the new features,and the recursive feature elimination and cross validation-Decision Tree(RFECVDT)feature screening algorithm are adopted for feature engineering complementary and data processing.Then,the convolutional neural network model is used for training and prediction.The experimental results show that the accuracy of using KmeansSMOTE algorithm in the convolutional neural network model is improved by 2.78%over SMOTE algorithm.The model is 4.77%more accurate than the same model without using the above-mentioned feature engineering.Compared with traditional models such as K-NearestNeighbor,Support vector machine,Logistic regression,all the measurement indexes are improved and there is no overfitting phenomenon.The proposed method solves the deficiency caused by applying SMOTE interpolation mechanism and refines the design of feature engineering.It provides a new solution to the troubleshooting of wind turbine blade icing.
作者
汤占军
史小兵
肖遥
李英娜
TANG Zhanjun;SHI Xiaobing;XIAO Yao;LI Yingna(School of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650504,China)
出处
《噪声与振动控制》
CSCD
北大核心
2023年第4期96-103,共8页
Noise and Vibration Control
基金
国家自然科学基金资助项目(61962031)。
