Abstract: The classification of industrial wastewater is a prerequisite and foundation for water pollution prevention and water resources management. However, compared to domestic sewage, research on industrial wastewater classification is relatively lagging behind. Chemical Oxygen Demand (COD) of water is a core indicator for measuring water quality. To address the problem of low prediction accuracy in existing industrial wastewater COD classification algorithms, a convolutional neural network (CNN) hybrid model based on gated recurrent units (GRU) is proposed. According to the hybrid model, the COD data of
industrial wastewater measured by UV-Vis spectroscopy is subjected to Gaussian filtering and denoising at
the first, then the denoised spectral data is input into the CNN model for feature extraction, and finally,
COD classification of industrial wastewater is achieved using GRU neural network. The experimental
results show that the CNN-GRU classification model converges after 200 times of training, with a
classification accuracy of 99.5%. Compared with the long short-term memory method, the GRU method,
and the CNN-LSTM method, the classification accuracy of CNN-GRU method has a significant advantage.

Key words: industrial wastewater classification, ultraviolet-visible spectroscopy, Gaussian filter denoising, convolutional neural network-gated recurrent unit model