Enhanced prediction of plane wall jet flow features with hybrid CatBoost modelling

A wall jet is a typical flow field encountered in various engineering fields, including the deviation tunnels, the bottom outlets of dams, and wastewater treatment. Appropriate prediction of wall jet characteristics is helpful in multiple fields, such as fluid mixing optimization, flow control, and structural design. In this study, machine learning algorithms, including standalone CatBoost (CB) and four hybrid CB models including Baysian optimization algorithms (BOA-CB), Gray wolf Optimization algorithm (GWOA-CB), Coati Optimization algorithm (COA-CB), and Whale optimization algorithm (WOA-CB) are employed for the first time to predict the wall jet characteristics such as velocity profile (U), half-width(Y0.5), maximum velocity (Umax), boundary layer thickness (Ym) and skin friction (Cf) coefficient were predicted using the ML algorithms. Results showed that the hybrid models performed better than the standalone CB models in predicting wall jet characteristics. Among the hybrid CB models, the best results were obtained when the hyperparameters were optimized using the Coati optimization algorithm (COA-CB). In the cross-validation scenario, the mean absolute percentage (MAPE) of the COA-CB is 20% smaller than the MAPE of the GWOA-CB model. The feature importance analysis using SHAP values showed that the streamwise distance from the nozzle is the most critical parameter affecting the accuracy of the COA-CB model to predict Umax, Y0.5, Ym, Cf. The following essential parameters include expansion ratio, tailwater depth ratio, and Reynolds number of the jets. A comparison between the accuracy of this study and previous models showed that the accuracy of the ML models is significantly higher than that of the earlier models. In the cross-validation scenario, the MAPE of the COA-CB model for predicting maximum velocity is approximately 57% better than the top-performing model identified in prior research.