Optimization of Indirect Wastewater Characterization: a Hybrid Approach Based on Decision Trees, Genetic Algorithms and Spectroscopy

The spectral response of wastewater samples allows, through the use of correlation models, to estimate the pollutant load of the samples in a simple, fast and economical way. However, the accuracy of these models can be affected by alterations in the spectral by external agents such as vibrations or temperature changes. In these cases, approximating the spectral response to trend lines can sometimes provide better estimates, while in other, it is better to work with the original spectral response. This research work proposes a methodology to accurately estimate the pollutant load of wastewater using a hybrid characterization model based on decision trees, which allows, in all cases, to obtain the best possible characterization. This model, based on the analysis of the spectral response, determines which genetic algorithm-based estimation model to make use of: the original spectral response or to the approximation of this to global or individual trend lines for each colour group, to estimate the following parameters: chemical oxygen demand (COD), biochemical oxygen demand at 5 days (BOD5), total suspended solids (TSS), total nitrogen (TN) and total phosphorus (TP) in raw and treated wastewater respectively. The study was conducted on 650 wastewater samples from 43 WWTPs. The results show that the hybrid characterization model provides the best possible fit, achieving an improvement up to 5% in raw wastewater samples, and up to 26.32% in treated wastewater with respect to the use of models that employ point values of the original spectral response, being much more significant in the case of TN. The spectral response of wastewater samples allows, through the use of correlation models, to estimate the pollutant load of the samples in a simple, fast and economical way.