Abstract: The prediction of xenon-induced power oscillations in nuclear reactors is crucial for operational safety and efficiency. However, existing methodologies are often constrained by limitations in model applicability across diverse conditions, high computational costs associated with high-fidelity codes, or a heavy reliance on extensive datasets for purely data-driven models. This study presents the development and validation of a surrogate model designed to mitigate these challenges. The model integrates a physics-informed structure with a data-driven parameterization to enable efficient and accurate prediction of xenon-induced power oscillation dynamics under various power transients, utilizing a limited number of reference simulations from the Bamboo-C code. The methodological framework was established upon a simplified physical model for the axial shape index (ASI), which was derived from the one-dimensional, two-group neutron diffusion theory and the associated xenon-iodine dynamics equations. To parameterize this model, a hybrid optimization algorithm, which combined the global search capabilities of the differential evolution (DE) algorithm with the local convergence efficiency of the sequential least squares programming (SLSQP) method, was implemented. This algorithm identified the key unknown parameters of the model by assimilating a small set of reference ASI transient data, which was generated by Bamboo-C for five training scenarios originating from a 100%FP (full power) initial condition. Following the parameter identification, functional relationships between the key model parameters and the magnitude of the power transient (ΔP) were established via regression analysis. The analysis shows that these relationships can be accurately characterized by simple linear and exponential functions, with all corresponding coefficients of determination (R²) exceeding 0.99. This enables the surrogate model to calculate the necessary parameters for new, unsimulated transient scenarios by using ΔP as the sole input. The model’s predictive performance and generalization capability were subsequently evaluated across a comprehensive range of operating conditions, with initial power levels from 100%FP down to 40%FP. The validation results indicate that the surrogate model can effectively reproduce the reference ASI dynamics. For transients originating from 40%FP to 100%FP, the maximum relative difference between the predicted and reference ASI curves is maintained below 0.1, and the mean squared error (MSE) is below 10⁻⁴. The model demonstrates higher predictive accuracy for scenarios involving large-magnitude power transients (e.g., ΔP≥60), where the oscillation characteristics are more distinct. While the proposed framework substantially reduces the computational burden of xenon-induced power oscillation analysis, its performance for small-magnitude transients indicates an area for further refinement. Future work will be directed toward improving the model’s accuracy in these specific regimes and extending its scope to include the effects of control rod perturbations.