Mitigating Phase-Redistribution Effects in Pressure Transient Analysis Using Physics-Guided Neural Networks

Summary This study introduces a new data-driven approach to mitigate the effects of complex wellbore phase-redistribution on buildup tests in vertical wells with high water-cut. Phase-redistribution occurs due to gravity-driven fluid segregation, and alters the distribution of oil and water in the wellbore and leads to unreliable bottom-hole pressure (BHP) measurements at the downhole gauge; thus, Pressure Transient Analysis (PTA) tests become unreliable or can be misinterpreted. First, we develop a coupled reservoir-wellbore black-oil fluid flow model to simulate the effects of phase-redistribution under various operational conditions (including water cut, gauge location relative to the sandface, fluid properties, and reservoir parameters). From these simulations, we assemble a labeled training set pairing distorted pressure curves (affected by phase-redistribution) with the corresponding clean signals at the sandface. Next, we train a 1D U-Net convolutional neural network on the paired data. The network receives as input the distorted pressure signal and the Bourdet derivative, along with a mask highlighting intervals where the effect of phase-redistribution is present. Once trained, the network is tested on synthetic datasets. In the cases evaluated, it successfully restores the true pressure response, although results depend on careful selection of the mask to isolate the distorted zones. By removing phase-redistribution effects upfront, corrected pressure curves can be matched directly to analytical solutions of the diffusivity equation (DE) without additional wellbore storage adjustments. The novelty of the proposed method lies in its ability to remove phase-redistribution effects using a neural network trained on high-fidelity simulations, rather than relying on analytical solutions to approximate changes in wellbore storage for pressure signal matching. This data-driven approach opens new avenues to incorporate additional physical phenomena and extend its application to inclined and horizontal wells.