Combining physics-based and data-driven modeling in well construction: Hybrid fluid dynamics modeling

This study presents a hybrid modeling approach combining physics-based and data-driven models for improved standpipe pressure prediction during well constructing. The proposed approach provides a more robust and accurate model that mitigates some of the disadvantages of using purely physics-based or data-driven models. We start with purely physics-based modeling of standpipe pressure, both steady-state and transient. The steady-state model can predict the magnitude of pressure during drilling and circulation, while the transient model can simulate dynamic behavior such as pressure spikes when flow is initiated and fluid gels in the wellbore are broken. We then evaluate several machine learning techniques including neural network, deep learning, Gaussian process, support vector machine, and random forest, which were trained with drilling time-series datasets. Finally, machine learning techniques are combined with physics-based models via a hidden Markov model, which involves a rule-based, stochastic decision-making algorithm. The results of the hybrid modeling exercise show that the proposed hybrid model can more accurately predict standpipe pressure and capture the transient events when compared to either physics-based or machine learning models used by themselves. The hybrid model also provides better interpretability of results compared to a purely data-driven black-box model. This is because purely data-driven models lack a connection to the underlying physics. Being able to accurately model and manage the pressure response during drilling operations is essential, especially for wells drilled in narrow-margin drilling environments. We demonstrate how pressure can be more accurately predicted through our proposed hybrid modeling, leading to safer, more optimized operations. The work is also presented as an example of the benefits of hybrid modeling in general, which can be applied to many areas of well construction beyond mere standpipe prediction.