Qiang Liu, Xingya Feng, Soroush Abolfathi
Porous marine structures are increasingly employed as energy-dissipating elements in coastal and offshore engineering. Their hydrodynamic behaviour can be simulated using macroscopic CFD modelling approaches, in which porous effects are represented through equivalent pressure drop models. However, the predictive accuracy of such models strongly depends on the discharge coefficient (μ), which is commonly assigned a fixed empirical value or determined through computationally expensive trial-and-error calibration. This study develops a novel surrogate-assisted numerical framework that couples support vector regression with particle swarm optimisation (SVR–PSO) to provide rapid, condition-dependent recommendations of μ for macroscopic CFD simulations. Two macroscopic CFD models were formulated using well-established pressure drop models proposed by Chen et al. and Molin. A total of 65 numerical simulations were conducted for each model to generate a dataset for the corresponding ML models. Four ML algorithms were trained and compared, revealing SVR–PSO as the superior model. Under a representative wave condition, the SVR–PSO framework predicted discharge coefficients of μ=0.4806 for the Molin model and μ=0.4874 for the Chen et al. model, both deviating by less than 4% from the prescribed target value of μ=0.5. The ML-optimised μ values were incorporated into CFD models, and the resulting simulations were validated against laboratory measurements. Experimental validation demonstrated that the surrogate-assisted CFD model reduced the mean absolute percentage error (MAPE) of pressure drops from over 36% in the fixed-μ model (μ = 0.5) to below 17%. The coefficient recommendation procedure was completed within one minute, providing an efficient alternative to conventional empirical calibration in macroscopic simulations of wave interactions with perforated structures.