Reconstruction and analysis of negatively buoyant jets with interpretable machine learning

Marta Alvir; Luka Grbčić; Ante Sikirica; Lado Kranjčević

doi:10.1016/j.marpolbul.2023.114881

Reconstruction and analysis of negatively buoyant jets with interpretable machine learning

Mar Pollut Bull. 2023 May:190:114881. doi: 10.1016/j.marpolbul.2023.114881. Epub 2023 Apr 1.

Authors

Marta Alvir¹, Luka Grbčić², Ante Sikirica³, Lado Kranjčević⁴

Affiliations

¹ Department of Fluid Mechanics and Computational Engineering, Faculty of Engineering, University of Rijeka, Vukovarska 58, Rijeka 51000, Croatia. Electronic address: malvir@riteh.hr.
² Department of Fluid Mechanics and Computational Engineering, Faculty of Engineering, University of Rijeka, Vukovarska 58, Rijeka 51000, Croatia; Center for Advanced Computing and Modelling, University of Rijeka, Radmile Matejčić 2, Rijeka 51000, Croatia. Electronic address: lgrbcic@riteh.hr.
³ Center for Advanced Computing and Modelling, University of Rijeka, Radmile Matejčić 2, Rijeka 51000, Croatia. Electronic address: ante.sikirica@uniri.hr.
⁴ Department of Fluid Mechanics and Computational Engineering, Faculty of Engineering, University of Rijeka, Vukovarska 58, Rijeka 51000, Croatia; Center for Advanced Computing and Modelling, University of Rijeka, Radmile Matejčić 2, Rijeka 51000, Croatia. Electronic address: lado.kranjcevic@riteh.hr.

PMID: 37011537
DOI: 10.1016/j.marpolbul.2023.114881

Abstract

In this paper, negatively inclined buoyant jets, which appear during the discharge of wastewater from processes such as desalination, are observed. A detailed numerical investigation is necessary to minimize harmful effects and assess environmental impact. Selecting appropriate geometry and working conditions for minimizing such effects often requires numerous experiments and numerical simulations. For this reason, the application of machine learning models is proposed. Several models including Support Vector Regression, Artificial Neural Networks, Random Forests, XGBoost, CatBoost and LightGBM were trained. The dataset was built with numerous OpenFOAM simulations, validated by experimental data from previous research. The average prediction of ML models has R² 0.94±0.05, RMSE 0.42±0.14 and RRSE 0.24 ± 0.09, whereas the best prediction was obtained by Artificial Neural Network with R² 0.98, RMSE 0.28 and RRSE 0.16. To understand the influence of input parameters on the geometrical characteristics of inclined buoyant jets, the SHAP feature interpretation method was used.

Keywords: Buoyant jet; Desalination; Machine learning; OpenFOAM; SHAP; Wastewater.

MeSH terms

Machine Learning*
Neural Networks, Computer*
Random Forest
Wastewater

Substances

Wastewater