Distribution characterization and strength prediction of backfill in underhand drift stopes based on sparrow search algorithm-extreme learning machine and field experiments

The backfill is the safety barrier of the backfilling stopes, and the uniaxial compressive strength (UCS) of the backfill is the key indicator to keep the stability of the backfill. To investigate the actual distribution of the UCS of the backfill in the stopes, a large-scale nickel mine in Northwest China is taken as the research object, and field experiments are carried out to investigate the actual distribution characteristics of the UCS of the backfill at different locations in the underhand drift stopes with different curing times. Further, the machine learning algorithm is used to construct a spatial-temporal distribution prediction model of the backfill UCS with different locations and different curing times as variables, and with the UCS of the backfill as the prediction target, to identify the weak parts and corresponding strengths of the backfill. The results show that the UCS of backfill has the distribution characteristics of increasing and then decreasing in the horizontal direction, and gradually decreasing in the vertical direction. When the curing time of the backfill in the stopes exceeds 80 days, its UCS growth tendency slows down. Compared with the back propagation neural network (BPNN), radial basis function neural network (RBFNN), and general regression neural network (GRNN), the prediction accuracy of extreme learning machine (ELM) is improved by 4.4 %, 3.4 %, and 13.9 %, respectively. Compared with the grey wolf algorithm (GWO) and particle swarm algorithm (PSO), the prediction accuracy of ELM optimized by the sparrow search algorithm (SSA) is improved by 7.2 % and 8.4 %, respectively. The SSA-ELM spatial-temporal distribution prediction model constructed in this study achieves fast and high-precision prediction of backfill UCS, which can provide guidance for stability analysis and quality control of underhand drift stopes. • Prediction of CPB strength can be quickly realized based on machine learning. • The backfill in the mining area has significant heterogeneity. • Weaknesses in the backfill roof can be found quickly by predicting the strength of the CPB.