Intelligent prediction of Darcy–Forchheimer hybrid nanofluid transport using recurrent neural networks

The flow of hybrid nanofluids through Darcy–Forchheimer structures is influenced by several factors, primarily turbulent and viscous damping from strong nonlinearities, the interaction of multiple properties, and how the system remembers its previous history, which can greatly obscure results that might otherwise be obtained using traditional numerical methods. This article presents an elaborate intelligent computational framework based on Recurrent Neural Networks (RNNs) for learning and predicting the nonlinear behaviour of hybrid ₂ ₃ ₂ Al ₂ O ₃ – TiO ₂ / water nanofluids in response to the factors of velocity slip, radiative heating effects, changes to permeability, and temperature stratification. The dataset utilised for training and validating the RNN architecture (having 35 hidden neurons and time delay feedback) was developed using a shooting based numerical solution to the transformed governing equations. The RNN model provides accurate predictions across all parametric cases tested, producing absolute prediction errors close to zero and highly correlated to those obtained from numerical simulations, as verified by MSE, regression accuracy, error histograms, and correlation analysis, as well as evaluation of a time series response curve. Overall, the study shows that the RNN framework provides excellent representations of the interacting influences of porous media characteristics, slip velocity, nonlinear influences of heat transfer, and stratification on both the velocity and temperature fields. Additionally, the findings support the argument for the use of RNNs as powerful alternatives to classical numerical techniques for more efficient convergence and stable operation with reduced computational costs associated with the numerical modelling of complex porous media nanofluid systems.