Yangqiang Zhang
Abstract Coal mining activities are subjected to various hazards like roof failures, rockbursts, and gas blasts, which endanger human lives and mining structures. Roof deformation prediction is important in anticipating such tragedies and maintaining operational safety. This paper presents a dynamic forecasting model of roof deformation in coal mines using Spatiotemporal Graph Neural Networks (ST-GNN). The main goal is to break the limitations of current approaches, which do not fully take into consideration the spatiotemporal characteristics of mining terrain, by taking advantage of real-time sensor data and geology. The new model incorporates a strong preprocessing phase in order to deal with issues like missing values, normalization, and temporal aggregation of data in such a way that the dataset is clean, normalized, and ready for deep learning (DL). Missing values are dealt with via imputation methods, and normalization of data makes sure that different scale features do not have a disproportionate effect on the performance of the model. Temporal aggregation of data enables to consolidate mining activities over limited time periods, which makes more consistent. The proposed ST-GNN model achieves an accuracy of 0.9874, with a precision of 0.9869, recall of 0.9844, and an F1-score of 0.9856, demonstrating strong predictive performance for roof deformation prediction. These findings confirm that the model surpasses the conventional prediction techniques, being a very accurate, scalable, and efficient computational solution for roof deformation prediction in real-time. The above method improves the safety control of coal mining processes and presents a promising line of study for mining disaster prevention in the future.