A Comparative analysis of neural networks and genetic algorithms to characterize wastewater from led spectrophotometry

• Published in: Journal of environmental chemical engineering Vol. 11; no. 3; p. 110219
• Main Authors: Carreres-Prieto, Daniel, Ybarra-Moreno, Javier, García, Juan T., Cerdán-Cartagena, J. Fernando
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.06.2023
• Subjects: And TSS; COD; Genetic algorithm; Neural networks; Visible led spectroscopy; Wastewater pollutant characterization; And TSS; COD; Wastewater pollutant characterization; Neural networks; Genetic algorithm; Visible led spectroscopy
• ISSN: 2213-3437; 2213-2929
• DOI: 10.1016/j.jece.2023.110219

The present research work seeks to solve the need for having real-time characterization of the pollutant load of urban wastewater using a robust and economical analytical technique such as spectrophotometry. For that reason, a comparison is presented between the use of neural networks (2 models) and genetic algorithms (4 models) to characterize COD and TSS values in raw and treated wastewater; over 500 wastewater samples were taken from the Mapocho-Trebal wastewater treatment plant in Chile. The results of this research indicate that, although both techniques provided similar fits, the neural networks generally performed better in estimating the test data than the genetic algorithms, with the exception of TSS characterization in raw wastewater, where the genetic algorithms performed better. The neural network presented an RMSE of 57.75 mg/l for COD and raw wastewater, while for TSS, the best performance was achieved with the genetic algorithm, with an RMSE of 36.7 mg/l. In treated wastewater, the neural network provided lower RMSE for COD and TSS than the genetic algorithm, 13.36 mg/l and 9.72 mg/l, respectively.The results obtained make it possible to consider using these economical, fast, and simple indirect techniques for the characterization of the pollutant load of wastewater in continuous operation, which will help to meet the objectives of Sustainable Development Goals of the UN. •LED spectrophotometry with AI techniques accurately characterizes wastewater without chemical reagents or pretreatment.•The study analyzed around 500 wastewater samples from the Mapocho-Trebal WWTP in Chile.•Two neural network and four genetic algorithm models were compared for COD and TSS characterization in wastewater.•Neural networks outperformed genetic algorithms for both raw and treated wastewater in test data.•The research has implications for improving wastewater characterization and achieving sustainable development goals.