Abstract: The prediction and forecasting of cyclic roof pressure in fully mechanized mining faces play a crucial role in preventing roof disasters and ensuring mine safety. The current development of intelligent fully mechanized mining has generated a massive amount of pressure data from mining faces. Data-driven approaches and digital intelligence empowerment are the key technical developments in the prediction and forecasting of cyclic roof pressure. Based on the technical requirements for roof pressure forecasting and the advantages of dynamic data analysis, this study proposes a dynamic forecasting method system for spatiotemporal regional roof pressure events in intelligent fully mechanized mining faces. First, a regional block pressure discrimination method is proposed, with feature extraction and pressure discrimination algorithms designed based on regional block division. By comparing and analyzing on-site data with single-frame and regional block pressure discrimination results, it is verified that the regional block pressure discrimination method is more effective in revealing the periodic spatiotemporal patterns of pressure changes in the working face. Then, a regional block cyclic end resistance feature prediction model based on a CNN-BiLSTM-Attention fusion network was constructed. Using 200 working cycles of hydraulic support pillar pressure field data as samples for modeling analysis, the superiority of the model was validated through comparative evaluation of different models, with an MSE as low as 0.002 3, meeting application standards; Finally, a regional self-correlation aggregation model based on the local Moran index was constructed, and a regional block cyclic end resistance feature self-correlation aggregation algorithm process was designed and developed to achieve automatic and dynamic forecasting of pressure events and their regional scope. When compared with actual pressure conditions, the model's prediction accuracy reached 85%. The aforementioned model methods were applied to this working face, similar geological working faces in different mines, and different geological working faces in the same mine. The analysis results indicate: this working face can accurately dynamically predict and forecast the next phase of cyclic pressure events and their regional scope; under similar geological conditions, the model still has certain application value and good generalization capability; under different geological conditions, the model's application effectiveness significantly decreases, requiring re-modeling to improve prediction accuracy and practicality. This method provides a feasible technical approach for predicting and forecasting cyclic pressure events in working faces, offering significant application potential in mine safety monitoring and disaster warning systems.