Phase change energy storage plays an important role in the popularization of renewable energy sources (wind, solar, geothermal, etc.). However, previous studies have paid little attention to the correlation and optimal solution of phase transformation thermal performance prediction. This research combines Response Surface Methodology with the Non-dominated Sorting Genetic Algorithm II to forecast and enhance the performance of tree-shaped biomimetic latent heat storage units, targeting optimal functionality under operational conditions. Initially, the study delves into a mechanistic analysis of the heat transfer in latent heat storage units, focusing on the coupling of conductive and natural convective heat transfer during the melting phase. The synergistic enhancement of heat conduction and convection is realized through locally refined and punched fins. Subsequently, Response Surface Methodology was applied to analyze the interactions between the design variables and the objective functions, leading to the attainment of Pareto-optimal frontiers using the Non-dominated Sorting Genetic Algorithm II. The results demonstrate that the Pareto-optimal solutions substantially improved the Nusselt number of novel biomimetic latent heat storage units by 42.58%–45.58 %, relative to tree-shaped fin latent heat storage units. Concurrently, the Fourier number was reduced by 17.54%–19.22 %. Moreover, the effect of adding nanoparticles (aluminum, copper oxide, and copper) to the phase change material was studied. The results show that the proposed locally refined optimal fin structure is superior to the addition of nanoparticles.