• Journal of Atmospheric and Environmental Optics
  • Vol. 19, Issue 1, 73 (2024)
MIAO Junfeng1, TANG Bin1,*, CHEN Qing1, LONG Zourong1..., YE Binqiang1, ZHOU Yan2, ZHANG Jinfu1, ZHAO Mingfu1 and ZHOU Mi1,**|Show fewer author(s)
Author Affiliations
  • 1School of Electrical and Electronic Engineering, Chongqing University of Technology, Chongqing 400054, China
  • 2Chongqing Tongliang District Ecological Environment Monitoring Station, Chongqing 402560, China
  • show less
    DOI: 10.3969/j.issn.1673-6141.2024.01.006 Cite this Article
    Junfeng MIAO, Bin TANG, Qing CHEN, Zourong LONG, Binqiang YE, Yan ZHOU, Jinfu ZHANG, Mingfu ZHAO, Mi ZHOU. Research on CNN-GRU industrial wastewater classification model based on UV-Vis spectroscopy[J]. Journal of Atmospheric and Environmental Optics, 2024, 19(1): 73 Copy Citation Text show less
    Structure diagram of experimental platform based on ultraviolet-visible spectroscopy
    Fig. 1. Structure diagram of experimental platform based on ultraviolet-visible spectroscopy
    UV-Vis spectral data of 0-500 mg/L COD solution
    Fig. 2. UV-Vis spectral data of 0-500 mg/L COD solution
    Gated recurrent unit structure
    Fig. 3. Gated recurrent unit structure
    Flow chart of CNN-GRU network model
    Fig. 4. Flow chart of CNN-GRU network model
    Raw spectral data
    Fig. 5. Raw spectral data
    Noised spectral data. (a) 50 dB; (b) 40 dB; (c) 30 dB; (d) 20 dB
    Fig. 6. Noised spectral data. (a) 50 dB; (b) 40 dB; (c) 30 dB; (d) 20 dB
    Gaussian filter denoising with 30 dB noise added
    Fig. 7. Gaussian filter denoising with 30 dB noise added
    Gaussian filter denoising of raw data
    Fig. 8. Gaussian filter denoising of raw data
    CNN-GRU model COD classification training result diagram
    Fig. 9. CNN-GRU model COD classification training result diagram
    Industrial wastewater classificationConcentration/(mg·L-1)
    Ⅰ类0-100
    Ⅱ类100-150
    Ⅲ类150-500
    Table 1. Industrial wastewater COD classification standard
    LayerNetworkOutput shape
    1Conv1D(2067, 1)
    2Conv1D(2065, 1)
    3MaxPooling1D(688, 1)
    4Conv1D(686, 1)
    5Conv1D(684, 1)
    6MaxPooling1D(228, 1)
    7Conv1D(226, 1)
    8Conv1D(224, 1)
    9MaxPooling1D(74, 1)
    10GRU(64)
    11GRU(64)
    12Flatten(4736)
    13Dence(3)
    Table 2. Output dimension of each layer of CNN-GRU network
    Denoising methodRSNRSME
    20 dB30 dB40 dB50 dB20 dB30 dB40 dB50 dB
    Savitzky-Golay17.9619.0919.1919.220.060.050.050.05
    Wavelet17.2518.2718.3818.400.070.060.060.06
    Gauss17.9321.9622.7822.910.060.040.040.04
    Table 3. The result of spectral denosing
    Predictive modelPrediction accuracy/%
    CNN Model96.0
    LSTM Model97.5
    GRU Model97.5
    CNN-LSTM Model98.0
    CNN-GRU Model99.5
    Table 4. Model prediction results comparison
    Junfeng MIAO, Bin TANG, Qing CHEN, Zourong LONG, Binqiang YE, Yan ZHOU, Jinfu ZHANG, Mingfu ZHAO, Mi ZHOU. Research on CNN-GRU industrial wastewater classification model based on UV-Vis spectroscopy[J]. Journal of Atmospheric and Environmental Optics, 2024, 19(1): 73
    Download Citation