Advanced displacement prediction of deep roadway roof with data fusion using machine-learning models: a comparative study

Purpose The machine learning tool is applied to the problem of large deformation of roof in the process of roadway excavation, and the feature extraction is realized by data fusion. Combined with machine learning tools such as a support vector machine, the dynamic model of roof settlement prediction in the process of working face mining is established, which finally serves the deep coal resource mining project. Design/methodology/approach Combining computer science with coal engineering, machine learning tools are applied to solve practical problems encountered in the mining process of deep coal resources. Findings In the five prediction models, random forest (RF) not only completed the extraction of key features, but also achieved good results in the actual prediction task, and successfully applied to the roof deformation prediction work in the working face mining site. Research limitations/implications The data set sample of the model is limited to 8,302 working face of Xinjulong Coal Mine, and the regional characteristics obviously hinder the comprehensive promotion of the model. The construction of a large database and the in-depth mining results of data are more convincing, especially for the whole Luxi mining area. Practical implications On one hand, accurately predicting advanced displacement evaluation of the roof during mining process, can provide guidance for subsequent enhanced pressure-relief and reinforcement support measures, which is of great significance for controlling deformation of surrounding rock and reducing impact threat. On the other hand, it shows the feasibility and accuracy of machine-learning tools in solving complex problems of deep rockburst mines engineering. Originality/value The prediction model of roof advance displacement under the influence of space-time effect proposed in this article has specific and universal applicability. The RF model shows excellent generalization ability and robustness, and achieves satisfactory accuracy and reliability. In addition, in the field monitoring data prediction work, the location of large-scale deformation of the roadway was successfully predicted, which ensured the safe production task of the coal mine.