The filtrate contamination cleanup time on a complex carbonate well using a traditional wireline formation tester (WFT) tool can vary from a couple of hours to over half a day. The method proposed aims at reducing operational time to collect a low-contamination formation fluid sample by determining regions with a smaller depth of invasion using a forward model simulation that considers static and dynamic formation properties to predict the radial profile of invasion. The mud filtrate invasion process was modeled considering the static and dynamic properties of the near-wellbore region in an industry reference reservoir simulator, and it integrates three mechanisms for fluid flow: Darcy’s law, material balance, and capillary pressure. The physical robustness of the reservoir simulator was united to a data-driven model to reduce the computational cost. This proxy model is based on a trained neural network with a broad range of scenarios to predict the numerical simulation results with high accuracy. The invasion estimation from the model is then used to predict the filtrate cleanup time using an industry consolidated numerical modeling. One of the variables influencing most of the cleanup time is the depth of mud filtrate invasion. Thus, reducing this time is a determinant for the WFT operational efficiency. The model for mud invasion has been successfully tested on a complex carbonate well, and the results for the depth of mud invasion were comparable to the results obtained with a commercial data-driven inversion using multiple resistivity channels. The estimated cleanup time using the results of depth of invasion predicted by the forward model has been compared and matched with real carbonate sampling stations, and there was a high correlation indicating that zones with lower depth of invasion required less cleanup time. Besides, using the history-matched cases, different WFT technologies such as single and radial probes, focused, unfocused, and dual-packer WFT inlets were evaluated, showing a high potential for reduction of operational time when properly planned and selected for the specific type of reservoir. The proposed methodology is a viable method for understanding the clean-up behavior in different reservoir scenarios using different WFT technologies. The innovation of this method relies on the data calibration using basic and advanced petrophysical properties through a data-driven model based on a trained neural network to reduce the uncertainty in the predicted invasion radial profile and the WFT cleanup time. The reliability on the theoretical results was increased using real data calibration, and this calibrated theoretical model has been used to guide the sampling depth selection, saving operational time.
Data-Driven Stability Assessment of Multilayer Long Short-Term Memory Networks
