Title : Integrating deep learning for accurate pore pressure prediction in fractured conglomerate formation of the junggar basin
Abstract:
Accurate pore pressure is basic for optimal mud density window determination and hence plays an essential role in oil and gas fields' safety and successful drilling process. Diverse pressure mechanisms associated with the complex geological formation of the Upper Wuerhe Formation in the Jinlong 2 block in the Junggar basin impair the accuracy of the conventional prediction methods. Most traditional methods rely on empirical relations between formation velocity and pressure extracted with insufficient consideration of unique geomechanical properties and underground structure; faults, fractures, and rapid change of burial depth.
This study proposed a new nonparametric multivariate model leveraging the DNN machine learning technique. Five logs; Neutron Porosity, Density, Shale Content, Longitudinal Acoustic Velocity, Overage, and Anisotropy logs were used to predict the effective stress. Thus, the PP prediction was achieved by using the effective stress theorem. A total of 1746 measured data sets were extracted from well logs, of which 80% were used in model training and 20% in validation.
The model shows a very good prediction performance with a coefficient of the determinant (R2 ) of 0.9821 and root mean square value (RMSE) of 0.02594 g/cm3 . By comparing with Sayer’s and Eberhart Philip’s performances (0.1256 g/cm3 , 0.0975 g/cm3 for JLHW 204 and 0.6425 g/cm3 , 0.6879 g/cm3 for JLHW 261 respectively), the DNN model outperforms and prevailing as a potential alternative multivariate prediction model capable on improving PP prediction accuracy in fracture conglomerate formation.
Audience Take Away Notes:
- Learn the applicability of advanced machine learning techniques to improve pore pressure prediction.
- Learn the advantages of integrating multiple petrophysical data for enhanced geo-pressure analysis.
- Learn the comparative performance of DNN-based models against traditional methods like Sayer’s and Eberhart Philip’s.
- Promote and encourage the creation of more robust models by integrating several log parameters.
- Optimize mud density window determination in complex formation
- Improve accuracy and reliability in drilling operations in similar formations.
- It provides a framework for incorporating machine learning in geomechanics.
- Offers a case study for advanced pore pressure prediction techniques in teaching.
- Can be a basis for further research on machine learning applications in petroleum.
- Yes, it simplifies the prediction process with a more accurate model.
- Reduces the need for extensive empirical calibration.
- Provides a robust tool for handling complex formations.
- Yes, it enhances the accuracy of pore pressure predictions that significantly enhance the precision of wellbore stability analyses.
- Provides new insights into geomechanical properties and fracture influences.
- Offers valuable data for designing effective drilling strategies.
