Acta Optica Sinica, Volume. 44, Issue 24, 2401004(2024)

Application of Bidirectional Long Short‐Term Memory Network in Doppler Lidar Wind Profile Prediction

Wenchao Lian1, Xiaoquan Song1,2、*, Zhaoyang Hao1, and Ping Jiang1
Author Affiliations
  • 1College of Marine Technology, Faculty of Information Science and Engineering, Ocean University of China, Qingdao 266100, Shandong , China
  • 2Laboratory for Regional Oceanography and Numerical Modeling, Qingdao Marine Science and Technology Center, Qingdao 266237, Shandong , China
  • show less
    Figures & Tables(10)
    Doppler lidar. (a) Appearance of the lidar; (b) schematic diagram of the DBS 5 beams
    Wind vectors detection by lidar. (a) u component; (b) v component
    Loss function. (a) u component; (b) v component
    Performance evaluation metrics for predicting 24 wind profiles using the Bi-LSTM model are R2, RMSE, and MAE. (a)(b)(c) u component; (d)(e)(f) v component
    Box plots of evaluation metrics for predicting the u and v components of 24 wind profiles using multiple look backs for two models. (a) R2; (b) RMSE; (c) MAE
    Missing wind profiles prediction based on Bi-LSTM. (a) u component; (b) v component; (c) data acquisition rate
    Wind profile prediction based on Bi-LSTM model. (a) 15:30 on 21 April; (b) 12:00 on 22 April
    Box plots of evaluation metrics for predicting the u and v components of 24 wind profiles at low and high altitudes. (a) R2; (b) RMSE; (c) MAE
    • Table 1. Parameter of Doppler lidar

      View table

      Table 1. Parameter of Doppler lidar

      ParameterValue
      Laser wavelength /nm1550
      Scanning modeDoppler beam swing (DBS)
      Scanning beam elevation angle /(°)60
      ProductZonal u component and meridional v component of wind vectors at altitude layers
      Temporal resolution /min10 (derived from the averaging of second-level wind vectors)
      Height range of the data used /m51‒1013, with a total of 38 altitude layers, vertical resolution is 26 m
    • Table 2. Parameter of model training

      View table

      Table 2. Parameter of model training

      ParameterValue
      Look back3, 6, 12, 24, 48, 72
      Epochs100
      Loss functionMean squared error (MSE)
      OptimizerAdaptive moment estimation (Adam)
      Bi-LSTM

      Bi-LSTM (128)

      Dense (128), activation='relu'

      Dense(38)

      CNN

      Convolution1D (128)

      Convolution1D (128)

      MaxPooling1D (1)

      Dense (128), activation='relu'

      Dense (38)

    Tools

    Get Citation

    Copy Citation Text

    Wenchao Lian, Xiaoquan Song, Zhaoyang Hao, Ping Jiang. Application of Bidirectional Long Short‐Term Memory Network in Doppler Lidar Wind Profile Prediction[J]. Acta Optica Sinica, 2024, 44(24): 2401004

    Download Citation

    EndNote(RIS)BibTexPlain Text
    Save article for my favorites
    Paper Information

    Category: Atmospheric Optics and Oceanic Optics

    Received: Apr. 22, 2024

    Accepted: May. 13, 2024

    Published Online: Dec. 16, 2024

    The Author Email: Song Xiaoquan (songxq@ouc.edu.cn)

    DOI:10.3788/AOS240891

    CSTR:32393.14.AOS240891

    Topics