Abstract: Currently, the traction speed performance of shearers in fully mechanized mining faces exhibits high redundancy. During high-speed operations, the efficiency of hydraulic support advancement often fails to match the shearer’s pace, which easily leads to excessive lag distance and subsequent roof support safety hazards. To address the challenge of precisely controlling the lag distance in shearer-following support advancement, this study, based on a medium-thick coal seam in Inner Mongolia, analyzes the distribution characteristics of lag distance and its correlation with support pressure and shearer traction speed. Lag distance is categorized into three types: small, normal, and large. A TCN-LSTM-Attention hybrid deep learning model is proposed. Specifically, the TCN branch extracts local features from multi-dimensional time-series data to capture short-term fluctuations in support pressure; the Bi-LSTM branch focuses on mining long-term dependencies to capture the dynamic trends of lag distance under various operating conditions; and the Attention mechanism dynamically weights and fuses the features from both branches, highlighting key features while suppressing redundant information. This architecture enables precise partitioned prediction for different lag distance categories. The results demonstrate that the proposed model achieves an overall prediction accuracy of 89.86%, with a recall rate of 81.58% for high-risk "large lag distance" conditions, significantly outperforming comparable models. These findings provide technical support for the adaptive coordinated control of shearers and supports, effectively mitigating roof support safety risks.