In the process of ore drawing using a caving method under interburden conditions, the key to controlling ore dilution lies in the accurate prediction of boundary particle migration trajectories. To address the challenges of high computational costs and complex modeling in traditional numerical simulations, this study designs a dataset construction method. After calibrating parameters using the angle of repose, ore-drawing numerical simulation datasets with interburden (post-defined and pre-defined models) are established. Building upon this foundation, an improved Transformer model is proposed. The model enhances spatiotemporal representation through multi-layer feature fusion embedding, strengthens long-range dependency capture via a reinforced spatiotemporal attention backbone, improves local dynamic modeling capability through optimized decoding at the output stage, and integrates transfer learning to achieve continuous prediction of particle migration. Validation results demonstrate that the model accurately predicts the spatial distribution patterns and collective motion trends of particles, with prediction errors at critical nodes confined to within a single stage and an average estimation error of approximately 4% in interburden regions. The proposed approach effectively overcomes the timeliness bottleneck of traditional interburden ore-drawing simulations, enabling rapid and accurate prediction of boundary particle migration under interburden conditions.
Ma et al. (Wed,) studied this question.