Prediction of hydraulic support pressure in a coal longwall face by integrating multi-head attention and temporal feature learning

Hydraulic support pressure in a coal longwall face is a key indicator for characterizing roof activity and strata weighting intensity, exhibiting pronounced nonlinearity and strong temporal dependence. To achieve accurate pressure prediction and early warning of roof weighting risks, this study takes the 39,103-longwall face of a coal mine as a case study and develops a time-series prediction model for hydraulic support pressure based on data acquired from an online mine pressure monitoring system. The proposed model integrates a convolutional neural network (CNN) for feature extraction, a long short-term memory (LSTM) network for temporal modeling, a multi-head attention mechanism for adaptive feature weighting, and an AdaBoost-based ensemble strategy to enhance prediction robustness. The results demonstrate that the proposed approach achieves consistently high prediction accuracy under different time-step settings, with the optimal overall performance obtained at a time step of four. Compared with the conventional LSTM model and its improved variants, the proposed method exhibits clear advantages in terms of RMSE, MAE, and R2. The findings of this study provide reliable technical support for roof weighting trend prediction and hydraulic support safety control in coal longwall mining, showing promising potential for practical engineering applications.