中国沙漠 ›› 2021, Vol. 41 ›› Issue (2): 38-50.DOI: 10.7522/j.issn.1000-694X.2020.00124

• • 上一篇    下一篇


王同亮1,3(), 马绍休1(), 高扬1,3, 宫毓来1,3, 安志山2   

  1. 1.中国科学院西北生态环境资源研究院,沙漠与沙漠化重点实验室,甘肃 兰州 730000
    2.中国科学院西北生态环境资源研究院,公共技术服务中心,甘肃 兰州 730000
    3.中国科学院大学,北京 100049
  • 收稿日期:2020-10-08 修回日期:2020-12-04 出版日期:2021-03-20 发布日期:2021-03-26
  • 通讯作者: 马绍休
  • 作者简介:马绍休(E-mail:
  • 基金资助:

The hybrid of wavelet packet decomposition and machine learning models in wind speed forecasting

Tongliang Wang1,3(), Shaoxiu Ma1(), Yang Gao1,3, Yulai Gong1,3, Zhishan An2   

  1. 1.Key Laboratory of Desert and Desertification /, Northwest Institute of Eco-Environment and Resources,Chinese Academy of Sciences,Lanzhou 730000,China
    2.Public Technical Service Center, Northwest Institute of Eco-Environment and Resources,Chinese Academy of Sciences,Lanzhou 730000,China
    3.University of Chinese Academy of Sciences,Beijing 100049,China
  • Received:2020-10-08 Revised:2020-12-04 Online:2021-03-20 Published:2021-03-26
  • Contact: Shaoxiu Ma



关键词: 风速预报, 小波包分解, 深度学习模型, 干旱半干旱区, 耦合模型


Wind speed forecasting is an effective method to improve wind power utilization and power system stability. Previous studies proposed a amount of wind speed forecast models, but there are few comparative studies on wind speed forecast models for different underlying land cover. This study mainly explores the ability of wavelet packet decomposition, 12 machine learning models and their coupled models to forecast the wind speed for three different underlying land cover and landforms (gobi, oasis and desert), and explores the optimized coupled model for wind speed forecast. Three sets of model experiments were set up for comparison in the article: single XXX model, WPD-XXX hybrid model and WPD-XXX-CNN hybrid model. We found that the deep learning models with feature selection, memory functions (such as convolutional long short-term memory networks) and extreme learning machine have better forecast capabilities for wind speed forecast. The wavelet packet decomposition significantly improves model accuracy. The coupled modesl of wavelet packet decomposition and convolutional long short-term memory network, convolutional gated recurrent unit network has better performance. This shows that the use of signal decomposition and the coupling of deep learning models can effectively improve the forecast accuracy of the models. It could be widely used in industry practices.

Key words: wind speed forecast, wavelet packet decomposition, deep learning model, arid and semi-arid area, hybrid model