Prediction and History Matching of Observed Production Rate and Bottomhole Pressure Data Sets from In-Situ Crosslinked Polymer Gel Conformance Treatments Using Machine Learning Methods

We present a computationally efficient methodology based on the use of machine learning (ML) for predicting and history matching oil rate, water rate, and injection bottomhole pressure (BHP) data recorded during an in-situ polymer gel treatment. Using this methodology, we can estimate rock/fluid parameters affecting the in-situ gel process that impact reservoir conformance. The use of two ML methods is investigated for this purpose. One of the ML methods used is the least-squares support vector regression (LS-SVR), a kernel-based ML method, and the other is the long short-term memory (LSTM) network, a deep learning method. The LS-SVR and LSTM proxy models are built on training sets of BHP and rate data generated with a commercial high-fidelity reservoir simulator (HFRS) based on compositional flow simulation using a double-permeability model. The reservoir models are calibrated using synthetically generated, random noise added BHP, oil, and/or water production rate data sets with small and large window sizes before and after polymer gel treatment. The ensemble smoother with multiple data assimilation (ES-MDA) is used for history matching the observed well outputs and assessing the uncertainty in the estimated parameters. When a compositional HFRS is replaced with any of the ML-based methods for performing history matching, we show that the LS-SVR and LSTM methods provide significant computational savings (more than an order of magnitude) in history matching including their training times over the conventional history matching based on a direct use of an HFRS. LSTM provides better predictions than LS-SVR for the same sizes of training sets and observed well outputs. However, for larger training sets, LSTM provides a significant computational gain over LS-SVR if clustering is not used for training LS-SVR surrogates for well outputs. In addition, we also identify the key parameters that have a significant impact on the performance of in-situ polymer gel treatment. The relative permeability curves of oil and water, absolute fracture permeability, polymer and crosslinked concentrations, and residual resistance factors (RRFTs) are the key parameters in the performance of in-situ polymer gel treatment. Although we have shown the application of our proposed methodology for a simple synthetic compositional reservoir model, the methodology is general in that it can be used for a real field case in a similar way for generating computationally efficient history-matched LS-SVR and LSTM surrogates that can be used to perform computationally efficient reservoir robust production optimization for in-situ gel operation.