ABSTRACT
Chemometric techniques like Principal Component Analysis (PCA) and Partial Least Squares Regression (PLS) are used to explore, analyze and model relationships among different water quality parameters in wastewater treatment plants (WWTP). Different data sets generated by laboratory analysis and by an automatic multi-parametric monitoring system with a new designed optical device have been investigated for temporal variations on water quality parameters measured in the water influent and effluent of a WWTP over different time scales. The obtained results allowed the discovery of the more important relationships among the monitored parameters and of their cyclic dependence on time (daily, monthly and annual cycles) and on different plant management procedures. This study intended also the modeling and prediction of concentrations of several water components and parameters, especially relevant for water quality assessment, such as Dissolved Organic Matter (DOM), Total Organic Carbon (TOC) nitrate, detergent, and phenol concentrations. PLS models were built to correlate target concentrations of these constituents with UV spectra measured in samples collected at (1) laboratory conditions (in synthetic water mixtures); and at (2) WWTP conditions (in real water samples from the plant). Using synthetic water mixtures, specific wavelengths were selected with the aim to establish simple and reliable prediction models, which gave good relative predictions with errors of around 3-4% for nitrates, detergent and phenols concentrations and of around 15% for the DOM in external validation. In the case of nitrate and TOC concentrations modeling in real water samples from the effluent of the WWTP using the reduced spectral data set, results were also promising with low prediction errors (less than 20%).
