Genetic Algorithm–CFD Optimization of Multi-Obstacle Arrays for Controlling Density-Driven Fluid Flows

Gravity currents—density-driven intrusions relevant to estuarine, coastal, and environmental engineering—can be modified by reorganizing how an advancing head is routed through obstacle arrays. This study couples a genetic algorithm (GA) with three-dimensional variable-density CFD to identify a high-performing arrangement of six full-depth square-prismatic obstacles in a confined lock–exchange channel. For H0 = 50 mm and Δρ/ρ0 = 0.01, the production input corresponds to Re = 3.50 × 103, Pe = 3.90 × 106, and Sc = 1.11 × 103. The population-mean endpoint front position decreases from 5.38 to approximately 4.40 (18.2%), while the best-found configuration reaches xf/H0 = 4.28 and adopts a 3–2–1 streamwise row allocation with nonuniform transverse spacing and staggered downstream rows. Across 500 randomly generated feasible configurations, the non-streamwise resolved-energy fraction is strongly negatively associated with endpoint front position (Spearman ρ = −0.82), a stronger association than that of the best single geometric descriptor (|ρ| = 0.61). Time-resolved velocity, density, and component-energy fields show asymmetric lateral bypass, downstream deflection, transient front deformation, and reduced late-time streamwise kinetic energy in the GA-best configuration. The results therefore connect combinatorial placement directly to the dynamics of gravity-current suppression within the present confined-channel design space.