The intellectualization of the fully mechanized mining face is a very attractive field, and mine pressure prediction is an important part of realizing the intellectualization of the fully mechanized mining face, which has a wide range of application scenarios. The hydraulic support load is a direct reflection of the overlying rock movement of the fully mechanized mining face, and the end-cycle resistance of the load peak is a key indicator for judging the pressure from the roof and carrying out the prevention and control of the coal mine roof, water, and fire disasters. In this paper, the change of resistance at the end of the cycle is predicted to characterize the occurrence of rock pressure, and a set of support data quality optimization methods based on regionalization, roof pressure prediction model construction based on space-time fusion, and multi-scale deployment method systems are formed. The experimental results show that the Conv1d+Bi-LSTM temporal-spatial model combined with the attention mechanism can effectively extract the characteristic information of the working condition matrix of the hydraulic support, and the data optimization method for different situations improves the prediction accuracy of the model, and the model has good robustness With generalizability, accurate prediction of mineral pressure can be achieved. The multi-scale deployment method realizes the flexible deployment of the model, which is of great significance for practical prediction.