Physics-informed machine learning for identification of preferential flow paths and performance forecasting of subsurface carbon storage and utilization

Monitoring the CO2 plume movement in the subsurface is essential for safety and storage integrity during CO2 storage and utilization. During field-scale CO2 EOR, routine well-wise injection/production data contain significant information which can be used for closed-loop reservoir management. Traditional physics-based numerical reservoir simulation can be computationally prohibitive for short-term decision cycles, and it requires detailed geologic models. As an alternative, reduced physics models provide an efficient simulator-free workflow but often have a limited range of applicability. Pure machine learning models lack physical interpretability and fail to provide process-based insights, resulting in limited predictive power. To address these challenges, we propose hybrid models, combining machine learning and physics-based approach, for rapid production forecasting and reservoir connectivity characterization using routine injection/production and pressure data collected during CO2 EOR operation. We combine reduced physics models into a neural network architecture by utilizing two different approaches. In the first approach, the reduced physics model is used for pre-processing to obtain approximate solutions that feed into a neural network as input. This physics-based input feature can reduce the model complexity and provide significant improvement in prediction performance. In the second approach, physics-informed neural network (PINN) is applied. The residual terms are augmented in the neural network loss function using a physics-based regularization that relies on the governing partial differential equations (PDE). Reduced physics models are used for the governing PDE to enable efficient neural network training. Our proposed hybrid models are first validated using a benchmark reservoir simulation case and then applied to a field case to show the robustness and efficacy of the method. The hybrid models are shown to provide superior prediction performance than pure machine learning models in terms of multiphase production rates. Specifically, the trained PINN model satisfies the reduced physics system, providing inter-well connectivity in terms of well-flux allocation. The flux allocation estimated from the hybrid model was compared with streamline-based flux allocation, and reasonable agreement was obtained for both the benchmark case and the field case. • Two novel hybrid models have been developed in this study, which infuse physical laws into neural network architecture for reservoir connectivity identification and robust multiphase production forecasting in CO2-EOR problems. • The proposed hybrid models are shown to outperform existing pure machine learning models (LSTM) in terms of long-term three-phase production forecasting. • The proposed physics-informed neural network framework provides reservoir connectivity in terms of dynamic flux allocation • The proposed hybrid models are calibrated solely based on the historical well measurements without relying on prior geological information.