<p>Uncertainty model updating techniques are pivotal for improving the accuracy of numerical models in the presence of uncertainties. However, current interval-based methods often fail to account for parameter correlations and suffer from computational inefficiency, particularly with complex numerical models. This study introduces a novel interval uncertainty model updating framework based on ellipsoidal convex model similarity. Initially, the ellipsoidal convex model is utilized to represent the bounded uncertainties of parameters along with their correlations, thereby addressing the shortcomings of traditional interval models that presume parameter independence. Subsequently, a novel metric combining Euclidean distance and Riemannian distance is employed to assess both the central distance and the shape similarity within the parameter space. Moreover, an efficient parameter updating strategy is developed by incorporating a dual-layer surrogate model—comprising kriging and transformer—with Riemannian gradient descent. This approach significantly reduces computational overhead while preserving the geometric constraints inherent to the ellipsoidal space. The proposed framework’s efficacy is confirmed through both numerical simulations and experimental case studies, showcasing its precision in updating parameter bounds and correlations. This research offers a robust and efficient methodology for interval model updating, suitable for practical engineering applications involving correlated uncertainties.</p>