Performance analysis of red mud concrete pier columns in coal pillarless mining and roadway deformation prediction based on machine learning

Abstract

To reduce concrete costs in pillarless coal mining and mitigate environmental impacts of red mud, this study integrates onsite research, numerical simulation, theoretical analysis, laboratory testing, and machine learning to develop a genetic algorithm-optimized AdaBoost prediction model for roadway deformation control using 20 % red mud concrete pier columns. Laboratory results demonstrate that the 20 % red mud concrete achieves balanced strength development, providing sufficient early strength (e.g., 16 MPa at 7 days) and high mid-to-late stage strength (e.g., 26.9 MPa at 28 days). Field implementation at Xiegou Mine’s 23,111 working face, validated by onsite monitoring, confirms that these columns effectively stabilize roadway deformation, limiting top and bottom slab displacements to 480 mm and 260 mm respectively at 1000 m mining distance—well within design tolerance. The prediction model exhibits high accuracy (MSE: 0.7830, RMSE: 0.8465, MAE: 0.4721, MAPE: 0.0342), with field data closely matching predicted deformation trends and ultimate limits. This high accuracy ensures the safe mining of the 23,111 working face.