欢迎访问《太阳能学报》官方网站,今天是 分享到:
ISSN 0254-0096 CN 11-2082/K

太阳能学报 ›› 2022, Vol. 43 ›› Issue (8): 399-408.DOI: 10.19912/j.0254-0096.tynxb.2021-0505

• • 上一篇    下一篇

基于WD-LSTM的风电机组叶片结冰状态评测

刘杰, 杨娜, 谭玉涛, 孙兴伟   

  1. 沈阳工业大学机械工程学院,沈阳 110870
  • 收稿日期:2021-05-11 出版日期:2022-08-28 发布日期:2023-02-28
  • 通讯作者: 刘 杰(1980—),女,博士、副教授,主要从事风电机组智能运行与维护方面的研究。starliujie@126.com
  • 基金资助:
    辽宁省教育厅科学研究经费项目(LQGD2020016); 辽宁省“兴辽英才计划”(XLYC1905003)

ASSESSMENT OF ICING STATE OF WIND TURBINE BLADES BASED ON WD-LSTM

Liu Jie, Yang Na, Tan Yutao, Sun Xingwei   

  1. School of Mechanical Engineering, Shenyang University of Technology, Shenyang 110870, China
  • Received:2021-05-11 Online:2022-08-28 Published:2023-02-28

摘要: 为有效识别叶片结冰状态,尽早采取除冰措施,提出基于小波去噪的长短期记忆神经网络(WD-LSTM)的评测方法。首先基于过采样与欠采样相结合的方法解决SCADA系统数据中的类别不平衡问题,通过对叶片结冰相关的26项指标进行分析,并从结冰机理和数据探索的角度筛选特征量,小波去噪处理后建立WD-LSTM模型,进一步完成模型的训练和测试。分别以15号和21号风电机组为例进行模型验证,通过与LSTM、概率神经网络(PNN)模型和BP神经网络模型进行对比。结果表明,WD-LSTM方法在风电机组叶片结冰评测中的准确率可达98%,优于其他方法。

关键词: 风电机组叶片, 长短期记忆, 状态评测, 特征量筛选, 小波去噪, 结冰状态

Abstract: An assessment method based on wavelet denoising long short term memory(WD-LSTM) was proposed in the paper to effectively identify the icing state of blades and take deicing measures as soon as possible. The problem of category imbalance in the SCADA system data was solved based on the combination of over-sampling and under-sampling. The 26 indicators related to blade icing were analyzed, and characteristic quantities were selected from the perspective of icing mechanism and data exploration. The WD-LSTM model was established after wavelet denoising to further complete the training and testing of the model. The No. 15 wind turbine and No. 21 wind turbine were taken as examples respectively for model verification compared with LSTM, Probabilistic Neural Network (PNN) model and BP neural network model. The results show that the accuracy rate of the WD-LSTM method reaches 98% in the assessment process of the wind turbine blades, which is better than other methods. It provides new ideas for the prediction of blade icing.

Key words: wind turbine blades, long short-term memory, state assessment, feature selection, wavelet denoising, icing state

中图分类号: