Abstract: This study proposes a mine pressure prediction method based on Transformer-BiLSTM as a solution to the limited accuracy of single prediction models in capturing composite features, thus enhancing the accuracy of mine pressure prediction. First, abnormal mine pressure data were eliminated based on the Shawville criterion, and the missing values were complemented by Akima spline interpolation to normalize and optimize the data quality. Then, the Transformer-BiLSTM prediction model was constructed incorporating the global attention mechanism of Transformer, the bidirectional time series feature extraction of BiLSTM, and the ensemble learning voting algorithm to improve its robustness. Results show that the proposed Transformer-BiLSTM model showed higher prediction accuracy for the monitoring data of periodic weighting in coal seam mining compared to the BiLSTM and Transformer models. This verifies its effectiveness in capturing the dynamic patterns of mine pressure and improving the prediction accuracy, thus offering theoretical references and practical methods for the prediction of mine pressure behavior in coal mine working faces.