Catalytic residues play a crucial role in understanding enzyme evolution and function at the molecular level. Sequence alignment-based approaches have been particularly effective in identifying catalytic residues, as they tend to be conserved throughout evolution. However, the increasing number of enzyme sequences poses significant challenges for alignment-based methods, as the sheer volume of data forces a trade-off between accuracy and computational efficiency, limiting their practicality on a large scale. In this work, we present an alignment-free deep learning method for high-throughput catalytic residue prediction by incorporating enzyme structural information. Experimental results on benchmark datasets show that our approach not only accelerates catalytic residue prediction but also consistently outperforms existing methods. Furthermore, we demonstrate that it generalizes well to enzymes with low similarity to the training set and accurately captures key structural and biophysical characteristics of catalytic residues. This positions our method as a competitive alternative to traditional approaches for the reliable prediction of catalytic residues, making it valuable for both industrial and academic applications.