Dynamic prediction of surface movement and deformation based on machine learning for overburden bed separation grouting filling:A case study of the 3801 working face in Huoerxinhe Coal Industry
CHEN Shaojie
SHENG Shouqian
HAN Lei
YIN Dawei
WANG Rui
YANG Yingsong
Abstract：The accurate and dynamic prediction of surface movement and deformation by over-burden bed separation grouting filling is of great significance for the early warning and protec-tion of surface building deformation.Considering the direct influencing factors of surface movement and deformation caused by overburden bed separation grouting filling,a CNN conv-olutional neural network LightGBM prediction model based on GA genetic algorithm optimiza-tion(GA-CNN-LightGBM model,GCL model)for overburden separation grouting surface movement and deformation is constructed using multi-source data such as working face produc-tion data,overburden bed separation grouting filling data,and surface movement and deforma-tion data as data sources.At the same time,the reliability of the model is tested by the four indexes of the mean absolute error(EMAE),the root mean square error(ERMSE),the coefficient of determination(R2)and the residual error(ε).And the method has been successfully ap-plied to the prediction of surface movement and deformation at the 3801 working face in Huo-erxinhe coal industry.The results show that the GCL model was trained to predict surface tilt using multi-source data collected during the mining process in front of the 3801 working face of Huoerxinhe Coal Industry.After training,the maximum residual of the GCL model's surface tilt prediction was less than 0.1 mm/m,with EMAE and ERMSE of 0.029 mm/m and 0.037 mm/m,respectively.The deviation between the predicted value and the measured value was relatively small.R2 is 0.967,indicating a high degree of model fitting and meeting the require-ments of dynamic prediction.In the application of surface tilt prediction in the surface oil stor-age tank area of the 3801 working face of Huoerxinhe Coal Industry,the GCL model has a maximum residual of 0.075 mm/m for surface tilt prediction,which is less than 0.1 mm/m.The EMAE and ERMSE are 0.039 mm/m and 0.042 mm/m,respectively.The R2 is 0.941.The degree of deviation between the predicted values and the measured values,and the degree of model fit,both demonstrate that the model has good predictive performance.During and after the 3801 working face,the actual maximum tilt of the oil storage tank was less than the warn-ing value of the tank.The GCL model dynamically predicted and ensured the safe mining of the working face without shutting down the Sinopec gas station.The research results can pro-vide theoretical and technical support for related projects.
Keywords：overburden separation groutingmachine learningGCL modelsurface movement and deformationdynamic prediction
Publication Date：2025-11-30
Online Publishing Date：2026-01-09(First online date of this platform, not the publication date of the document)
Pages：15( 1261-1275 )