The current challenges facing data-driven impact localisation methods primarily involve accurately localising impacts occurring outside the training impact coverage area, particularly for anisotropic composite structures. This study addresses these challenges by integrating the physical dispersion relations of impact-induced waves into the localisation process, thereby enhancing overall accuracy and reducing reliance on extensive training impact data. The dominant impact-induced flexural waves, derived from first-order shear deformation theory and classical laminate theory, exhibit significant dispersion. The phase velocity of these waves is governed by a dispersion equation dependent on structural stiffness, wave frequency, and direction. Solving this dispersion equation for a given wave frequency and known structural stiffness yields a wave velocity profile (WVP). Unlike conventional data-driven methods, which approximate the WVP with a black box and relate the time difference of arrival to the impact location, this study explicitly formulates and optimises the WVP with respect to structural stiffness and wave frequency based on the training impacts. Leveraging the partial derivatives of the dispersion equation, optimisation is performed using fast gradient-based techniques such as the Newton’s method and the trust-region reflective algorithm. The optimised WVP is then employed for impact localisation using the triangulation method. Experimental validation of the proposed hybrid physics-based and data-driven impact localisation method involves conducting drop-mass impact testing on a composite laminate flat panel and a stringer-stiffened panel. The localisation results confirm the efficiency and accuracy of the proposed hybrid method with minimal training impact data, demonstrating its capability to accurately localise impacts even for stiffened panels, including those occurring outside the coverage area of the training impacts.