To address the increasing regional pressure on charging facilities caused by the rapid growth of electric vehicles in cities, this study aims to realize short-term prediction of charging pile occupancy across multiple urban regions, thereby supporting urban charging resource scheduling and optimization. A traffic-informed charging facility utilization prediction model (TI-CFU) is developed. First, a multi-type node feature generation module is designed to extract temporal traffic features and charging facility states for each grid cell. Then, a channel attention mechanism is applied to refine key traffic patterns and highlight regional hotspot features. Finally, a gated fusion module dynamically integrates traffic features with charging status information, and a prediction layer is employed to estimate short-term charging facility utilization. The proposed model is evaluated using real traffic and charging facility datasets from Shanghai. Experimental results show that TI-CFU achieves the best performance among all comparison methods, reducing overall prediction error by approximately 7%–15%, and maintains high accuracy and stability across different urban regions.