Analysis and experimental prediction of surface-internal gangue content differences in coal-gangue accumulations under intelligent longwall top coal caving

Intelligent top coal release is central to Longwall Top-Coal Caving (LTCC), where real-time monitoring of gangue content is essential for refined control. Recent advances in image recognition enable accurate detection of surface gangue content, but differences between surface and internal gangue remain unclear. Based on geological and mining parameters from a Shendong coal mine in China, an experimental LTCC model was developed to investigate surface–internal gangue content differences in coal–gangue accumulations on a scraper conveyor. The influences of mining height, top coal thickness, coal seam dip angle, conveyor speed, and coal gangue block size were analyzed, and the sensitivity of the difference coefficient to these factors was examined. Results show that when coal block size is not larger than gangue, internal gangue content exceeds surface content, while the opposite holds when coal block size is larger. A linear relation exists between surface and internal gangue content under single-gate release. The difference coefficient correlates positively with mining height and top coal thickness, but negatively with conveyor speed and coal seam dip angle. Sensitivity analysis highlights mining height, gangue block size, and conveyor speed as dominant factors. Experimental validation identified Gradient Boosted Regression Trees (GBRT) as the optimal predictive model, achieving high accuracy in internal gangue content prediction.