In the horizontal sublevel top coal caving of steeply inclined extra-thick coal seam groups,the segmented mining is prone to induce roof collapses with time delayed characteristics.The understanding of the collapse height and characteristics of the overlying strata is crucial for monitoring roof disasters,crack development,and surface collapses.To address this,a combination of similar simulation experiments,machine learning,and grey theory methods were utilized to explore the collapse characteristics of overlying strata and the laws governing surface collapses.Data indicators associated with collapse height were obtained,and the dimensionality of input data was reduced using Principal Component Analysis(PCA).A collapse height prediction model(PCA-GA-LSSVR)based on the least squares support vector regression(LSSVR)network was established through the optimization of training using genetic algorithms(GA)and leave-one-out cross-validation.Various combinations of mining parameters and monitoring indicators were employed for training and testing,facilitating the evaluation and selection of the optimal prediction model.The selected model was further validated through engineering inspections.Results indicate that with increasing mining depth in the experimental mining of steeply inclined extra-thick coal seam groups,continuous collapse of rock strata generates and expands V-shaped subsidence.There are high correlations between mining depth,single caving height,model section height,and collapse height.Additionally,the collapse of the roof in experimental mining exhibits a time delayed characteristic.Dimensionality reduction of data effectively addresses the issue of high dimensionality and information overlap,enhancing the accuracy of prediction models.The PCA-GA-LSS VR model demonstrates superior accuracy in predicting test data,with lower error rates,average absolute errors,and average absolute percentage errors(5.146,4.819,and 0.087 respectively)compared to other models.The established grey time delayed OBGM model exhibits good fitting and predictive capabilities for roof collapse height in the #45 coal seam scenario.During engineering inspection,the maximum error fluctuation range of the established collapse height prediction model is within 3.36 m.In scenarios with limited sample data availability,gradually increasing learning samples reduces model prediction errors.This research provides valuable insights for studying collapse height in steeply inclined extra-thick coal seam groups during horizontal sublevel top coal caving mining.
关键词
采矿工程/急倾斜煤层/水平分段开采/冒落高度/最小二乘支持向量回归/时滞模型
Key words
mining engineering/steeply inclined coal seam/horizontal sectional mining/collapse height/least squares support vector regression/time delayed model
维普
万方数据