<p>Effective prediction of the logistics status for automated material handling systems (AMHS) in 12-inch semiconductor wafer fabrication systems is highly challenging, primarily due to their complex spatio-temporal characteristics, including adjacent and long-range temporal dependencies, dynamic spatial dependencies, and spatial heterogeneity. To address these challenges, we propose an AMHS logistics status prediction method integrating a Transformer and a dynamic graph attention convolutional network (ITDGATCN). Specifically, a Transformer-based framework is established to ensure complex spatio-temporal characteristics are fully incorporated into the prediction model. Within this framework, a spatio-temporal information input module is designed to enhance the model’s capacity to perceive the spatio-temporal dynamic nature of the AMHS logistics status. Furthermore, a temporal-aware multi-head self-attention mechanism is presented to effectively model adjacent and long-range temporal dependencies, while a dynamic graph attention temporal convolutional network is introduced to capture dynamic spatial dependencies and heterogeneity. Based on real logistics status data from a 12-inch wafer manufacturing facility in Shanghai, the proposed ITDGATCN method is compared with traditional logistics status prediction approaches. Experimental results demonstrate that the ITDGATCN is effective, exhibiting superior stability and adaptability compared to baseline approaches.</p>