Prediction of surface water pollution using wavelet transform and 1D-CNN

Abstract

Permanganate index (CODMn), total nitrogen, and ammonia nitrogen are important indicators that represent the degree of pollution of surface water. This study combined ultraviolet–visible (UV–vis) spectroscopy with a one-dimensional convolutional neural network (1D-CNN) to spectrally analyze 708 samples with different concentrations. The wavelet transform was used to preprocess the spectra to improve the model's accuracy. The results show the best prediction results using a fixed threshold (sqtwolog) of wavelets in combination with 1D-CNN, and the coefficient of determination (R2) values of the models on the test dataset all reached more than 0.98. A comparison between the backpropagation neural network model and the extreme learning machine model reveals that the 1D-CNN model has better prediction accuracy and robustness. The experimental results show the strong practical value of using 1D-CNN to predict the levels of different compounds in surface water.