Application of convolutional neural networks for prediction of disinfection by-products

Nicolás M. Peleato

doi:10.1038/s41598-021-03881-w

Abstract

Fluorescence spectroscopy can provide high-level chemical characterization and quantification that is suitable for use in online process monitoring and control. However, the high-dimensionality of excitation–emission matrices and superposition of underlying signals is a major challenge to implementation. Herein the use of Convolutional Neural Networks (CNNs) is investigated to interpret fluorescence spectra and predict the formation of disinfection by-products during drinking water treatment. Using deep CNNs, mean absolute prediction error on a test set of data for total trihalomethanes, total haloacetic acids, and the major individual species were all < 6 µg/L and represent a significant difference improved by 39–62% compared to multi-layer perceptron type networks. Heat maps that identify spectral areas of importance for prediction showed unique humic-like and protein-like regions for individual disinfection by-product species that can be used to validate models and provide insight into precursor characteristics. The use of fluorescence spectroscopy coupled with deep CNNs shows promise to be used for rapid estimation of DBP formation potentials without the need for extensive data pre-processing or dimensionality reduction. Knowledge of DBP formation potentials in near real-time can enable tighter treatment controls and management efforts to minimize the exposure of the public to DBPs.

Highlights

Fluorescence spectroscopy can provide high-level chemical characterization and quantification that is suitable for use in online process monitoring and control
A dataset of disinfection by-product (DBP) formation potentials and associated fluorescence excitation–emission matrices (EEMs) measurements were used to assess the capabilities of deep NNs and Convolutional Neural Networks (CNNs) for water quality analysis
This study investigated the use of deep CNNs to interpret fluorescence spectra and predict the formation of regulated chlorination DBPs from a drinking water treatment plant

Summary

Introduction

Fluorescence spectroscopy can provide high-level chemical characterization and quantification that is suitable for use in online process monitoring and control. While this type of expert guided approach has been used extensively in the past, discarding the majority of collected data neglects the richness of information contained For complex systems such as those that include identifying natural organic matter (NOM) in water, organic fluorophores with similar chemical structures are not distinguished in the spectra. The use of principal component analysis (PCA) or parallel factors analysis (PARAFAC) has revealed underlying signals resembling fluorophores, which can be tied to spectral regions from which chemical properties can be inferred[15,22,23] These analysis approaches are often limited to linear dimensionality reduction, so non-linear features such as Rayleigh or Raman scattering need to be removed from the s pectra[24]. These same constraints may limit the overall accuracy of reconstruction based on the condensed r epresentation[27]

Methods

Results

Conclusion