The hydraulic support system in China cannot accurately predict disasters such as roof, coal brust, coal and gas outbursts, and the prediction of periodic weighting during support in complex environments of fully mechanized mining faces has always been a major challenge for unmanned working faces. To address this issue, research has found that the ConvLSTM model with spatiotemporal correlation analysis and Point of Intersection (POI) data has been established through methods such as theoretical analysis, data collection and preprocessing, model evaluation and optimization. To obtain the optimal solution for periodic pressure prediction. Realize real-time perception and prediction of the working environment status. The experimental results show that the mean square error of the POI ConvLSTM model prediction of working face periodic weighting is 0.159, and the R2 evaluation index is 0.999. Compared with the Seq2Seq and ConvLSTM models, the mean square error is reduced by 68.07% and 4.22%, respectively. Therefore, the POI ConvLSTM model that integrates multiple data sources has higher prediction accuracy, stronger universality, and can accurately predict periodic pressure problems in advance.