Robust Method for Reservoir Simulation History Matching Using Bayesian Inversion and Long-Short-Term Memory Network-Based Proxy

Summary History matching is a critical process used for calibrating simulation models and assessing subsurface uncertainties. This common technique aims to align the reservoir with observed data. However, achieving this goal often challenging due nonuniqueness of solution, underlying uncertainties, usually high computational cost simulations. The traditional approach based on trial error, which exhaustive labor-intensive. Some analytical numerical proxies combined Monte Carlo simulations are reduce time. these approaches suffer from low accuracy may not fully capture study proposes new robust method using Bayesian Markov chain (MCMC) perform assisted history under We propose novel three-step workflow that includes (1) multiresolution low-fidelity guarantee high-quality matching; (2) long-short-term memory (LSTM) network as model reproduce continuous time response model, optimization obtain optimum model; (3) MCMC runs inversion uncertainty parameters. sensitivity analysis LSTM’s architecture, hyperparameters, training set, number chains, length setup Bayesian-LSTM matching. also compare performance predicting recovery factor (RF) different surrogate methods, including polynomial chaos expansions (PCE), kriging, support vector machines regression (SVR). demonstrate proposed water flooding problem upper Tarbert formation 10th SPE comparative model. case represents highly heterogeneous nearshore environment. Results showed Bayesian-optimized LSTM has successfully captured physics in high-fidelity produces an accurate prediction narrow ranges posterior through ensures robustness workflow. provides efficient practical history-matching flow modeling significant