3D flow simulation of straight groynes using hybrid DE-based artificial intelligence methods

In this study, hybrid differential evolution algorithms were run to study the three-dimensional mean flow field around straight groynes. The three-dimensional velocity components in a refined mesh around a groyne were measured using an acoustic Doppler velocimeter. Two novel hybrid methods, namely differential evolution-based multilayer perceptron (DE-MLP) and differential evolution-based radial basis function (DE-RBF), were used to simulate the most significant flow features. It was found that the DE-MLP method modeled well the separation of the bottom boundary layer, downward deflection of the upper layers’ streamlines and horseshoe vortex (HSV) development at the upstream groyne face. However, the DE-RBF model was unable to predict the main flow features correctly, especially the HSV system. Both DE-MLP and DE-RBF models predicted the overall flow structure of the separation zone downstream of the groyne well. However, the RBF model could not predict the inflection points in the transverse velocity profiles, which are indicative of a mixing layer bounding the separation zone. The DE-RBF model underestimated the velocity amplification at the groyne head. The DE-MLP model perfectly simulated the distribution of the streamwise cross-stream Reynolds shear stress around the groyne head and also along the turbulent mixing layer.