This study introduces a reduced-order modelling (ROM) and optimization framework to predict and improve unsteady flow around morphing wings. It is applied to the Intermediate-Scale A320 morphing wing prototype developed in the BEALIVE project (HORIZON-2023-2027-PATHFINDER, Grant No. 101129952), at Reynolds number 1×10⁶ and Mach 0.063 (inlet velocity 21.5 m/s) for a 10° angle of attack. The wing has a 70 cm chord and 59 cm span. Experimental TRPIV measurements in the S4 wind tunnel (IMFT) are combined with high-fidelity simulations. Flow separation occurs near x/c ≈ 0.7, and the dynamics are dominated by two shear layers: an upper shear layer over the separated region and a lower shear layer below the trailing edge and in the wake. Shear-layer instabilities generate Kelvin–Helmholtz vortices, and downstream interaction leads to von Kármán vortex shedding. Aerodynamic performance is targeted by manipulating these coherent structures using a bioinspired “live-skin” concept: a dense array of piezoceramic actuators placed over a strategic suction-side region. In simulations, actuation is represented through surface travelling or standing waves with optimized wavelength, amplitude, frequency, and actuation-zone bounds (x₀/c, x_f/c). Because exhaustive high-fidelity exploration of this design space is impractical, a ROM is built from a database of morphing cases computed with the NSMB Navier–Stokes multi-block solver using an ALE formulation for dynamic grid deformation and an Organized Eddy Simulation approach suited to coherent-structure development. For each actuation setting, a snapshot matrix combining velocity and grid-deformation information is decomposed using Proper Orthogonal Decomposition into spatial modes and temporal coefficients. Long Short-Term Memory networks learn the nonlinear time evolution of these coefficients, enabling accurate, low-cost reproduction of unsteady turbulent features and mesh motion over extended horizons. A parametric POD strategy extends the ROM across the actuation space to reconstruct flows at unseen parameters with minimal additional cost. Finally, the parametric ROM is embedded within Particle Swarm Optimization to efficiently identify actuation configurations that optimize aerodynamic performance while remaining physics-consistent.