In order to avoid the abnormal annular pressure caused by well failure and safety accidents, accurately predict the annular pressure value, and take preventive and solution measures in advance when it exceeds the control value. A hybrid model based on the training of the actual time series data of the annular pressure value and the dataset of feature capture, namely ARIMA-LSTM hybrid model, is proposed to predict the annular pressure value of a practical well, and compared with the single model and the RNN neural network model. The results show that after the training of the actual data, the model has good performance in error size, fitting accuracy and overall performance, which can provide reference for improving the prediction accuracy and efficiency of the annular pressure value, and has certain help for the well integrity design.