A Formation Pressure Calculation Method Based on the GWO-CNN-LSTM Neural Network

Formation pore pressure is a key parameter in the process of oil and gas drilling engineering design and construction. Its accurate prediction is of great significance for ensuring drilling safety and improving operational efficiency. Traditional prediction methods often suffer from limited accuracy and insufficient adaptability in complex geological conditions. To address this, this paper proposes a hybrid prediction model based on the GWO-CNN-LSTM Neural Network Model. This model first cleanses multi-source data from drilling, logging, and mud logging. Combining Pearson correlation analysis and engineering experience, 19 feature parameters are optimally selected as inputs. The CNN is used to extract local spatial features from the data, followed by the LSTM to capture temporal dependencies. Finally, the GWO is introduced to optimize the number of LSTM hidden units, initial learning rate, and L2 regularization parameters. Experimental results show that, under both single-well and adjacent-well prediction scenarios, the GWO-CNN-LSTM model achieves a mean absolute error (MAE) of 4.7168 MPa and 0.69053 MPa, a root mean square error (RMSE) of 5.5807 MPa and 0.94595 MPa, and a mean relative error (MRE) of 9.5109% and 2.4974%, respectively. All these indicators are superior to those of the comparison models (LSTM and SVM). This study validates the effectiveness and superiority of the GWO-CNN-LSTM model in formation pore pressure prediction, providing reliable technical support for intelligent drilling.