The computational efficiency and accuracy of unit commitment (UC) models directly affect the secure and economic operation of power systems. Efficient models can significantly improve UC solving efficiency without compromising optimality and have attracted attention. However

current studies mostly focus on specific boundary conditions and lack a systematic analysis of the performance differences of various modeling approaches under varying boundary conditions

which limits the practical applicability of acceleration algorithms. To address this gap

this study proposes a data-driven adaptive framework for selecting efficient UC modeling strategies based on case-specific boundary conditions. First

four efficient UC models are constructed and evaluated across varying system scales and load conditions in terms of efficiency and formulation quality. A decision tree-based method is developed to extract key boundary features. Finally

a multi-head residual neural network is built to predict the most efficient formulation based on boundary features and solving time. The simulation analysis shows the proposed method improves computational efficiency by an average of 41.20% in large-scale systems

with a prediction accuracy of 82.61% for the optimal formulation. This study provides a reference for the intelligent selection of efficient UC modeling approaches.