A Collapse Pressure Prediction Method Based on Virtual-Well Constraints and Deep Sequence Learning

The collapse pressure equivalent density is a key parameter for determining the safe drilling fluid density window and evaluating wellbore stability. To address the limited number of drilled wells, the lack of continuous geomechanical labels, and the complex relationship between seismic responses and collapse pressure, this study proposes a collapse pressure prediction method based on virtual-well constraints and deep sequence learning. Virtual-well samples were constructed from the acoustic impedance distributions of drilled wells, synthetic seismic records were generated through actual seismic wavelet extraction and reflection-coefficient convolution, and collapse-pressure equivalent-density labels were calculated using a rock-mechanics model, thereby forming a large-scale training dataset. Random Forest, Polynomial Regression, Support Vector Regression, and CNN-MultiLSTM models were compared. The CNN-MultiLSTM framework was further optimized in terms of input range, feature channels, and sequence structure, resulting in a whole-well sequence model combining multichannel inputs with residual dilated convolutions. The optimized model achieved an MAE of 0.00262 g/cm3, an RMSE of 0.00352 g/cm3, a MAPE of 0.241%, and an R2 of 0.9774 on the validation set. In independent validation using an actual well not involved in training, the model achieved an MAE of 0.0182 g/cm3, an RMSE of 0.0278 g/cm3, a MAPE of 1.640%, and an R2 of 0.7356 within the primary target interval of 7900.000–7994.100 m. The predicted profile reproduced the main depth-dependent variation of the calculated collapse-pressure equivalent density, although larger deviations occurred in locally abrupt intervals. Analysis outside the primary target interval further showed that the model could respond to high-collapse-pressure anomalies. Overall, integrating virtual-well constraints, rock-mechanics-based labeling, and whole-well sequence learning provides a feasible approach for collapse-pressure prediction in undrilled areas and drilling fluid density design, while further multi-well validation is required to assess cross-well generalization.