Research on Automatic Correction Methods for Underground Coal Mine Drilling Trajectories Based on Deep Learning

This study addresses the complex geological environment in underground drilling by analyzing the influence mechanisms of geological structures, drill string configurations, and rock layer properties on trajectory deviation. A multi-source data fusion prediction model was developed. By introducing a multi-head attention mechanism into an improved Transformer architecture, the system achieves adaptive weight allocation for measurement while drilling, geological parameters, and process parameters, significantly enhancing prediction accuracy. Based on deep reinforcement learning, an Actor-Critic intelligent deviation correction algorithm was designed to establish an automatic trajectory correction system. Field tests demonstrate that the system controls trajectory deviation within ±0.5°, achieving a 42.3% improvement in accuracy and a 35.8% increase in efficiency compared to traditional methods, providing a novel technical approach for precise underground drilling.