Abstract:We developed a novel reservoir characterization workflow that addresses reservoir history matching by coupling a physics-informed neural operator (PINO) forward model with a mixture of experts' approach, termed cluster classify regress (CCR). The inverse modelling is achieved via an adaptive Regularized Ensemble Kalman inversion (aREKI) method, ideal for rapid inverse uncertainty quantification during history matching. We parametrize unknown permeability and porosity fields for non-Gaussian posterior measures using a variational convolution autoencoder and a denoising diffusion implicit model (DDIM) exotic priors. The CCR works as a supervised model with the PINO surrogate to replicate nonlinear Peaceman well equations. The CCR's flexibility allows any independent machine-learning algorithm for each stage. The PINO reservoir surrogate's loss function is derived from supervised data loss and losses from the initial conditions and residual of the governing black oil PDE. The PINO-CCR surrogate outputs pressure, water, and gas saturations, along with oil, water, and gas production rates. The methodology was compared to a standard numerical black oil simulator for a waterflooding case on the Norne field, showing similar outputs. This PINO-CCR surrogate was then used in the aREKI history matching workflow, successfully recovering the unknown permeability, porosity and fault multiplier, with simulations up to 6000 times faster than conventional methods. Training the PINO-CCR surrogate on an NVIDIA H100 with 80G memory takes about 5 hours for 100 samples of the Norne field. This workflow is suitable for ensemble-based approaches, where posterior density sampling, given an expensive likelihood evaluation, is desirable for uncertainty quantification.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the **oil - field reservoir history - matching problem**. Specifically, the researchers have developed a new reservoir characterization workflow, which combines the **Physics - Informed Neural Operator (PINO) forward model** and the **Mixture of Experts (MoE)**, named **Cluster Classify Regress (CCR)**. This method aims to quickly and effectively perform the inverse uncertainty quantification task in reservoir history - matching through the **Adaptive Regularized Ensemble Kalman Inversion (aREKI)** technique. ### Main Problems 1. **Reservoir History - Matching Problem**: - Reservoir history - matching refers to the process of adjusting reservoir model parameters to make the simulation results match the actual production data (such as oil, water, and gas production). This is a typical inverse problem because the amount of available data is usually far less than the number of parameters to be determined, resulting in non - uniqueness of solutions. - This problem is particularly important in petroleum engineering because the accuracy of the reservoir model directly affects future production forecasts and resource management. 2. **Inverse Uncertainty Quantification**: - Due to the incompleteness of reservoir data and measurement errors, there is uncertainty in reservoir model parameters. Therefore, a method is needed to quantify these uncertainties in order to more accurately assess the reliability of the reservoir model. - The method proposed in the paper can efficiently perform inverse uncertainty quantification by using the aREKI technique and the PINO - CCR surrogate model. ### Solutions 1. **PINO Forward Model**: - Use the Physics - Informed Neural Operator (PINO) to construct a forward model, which can learn and predict dynamic properties such as pressure, water saturation, and gas saturation in the reservoir. - PINO improves the physical consistency and prediction ability of the model by combining physical laws (such as the black - oil equation) and data - driven methods. 2. **Mixture of Experts (MoE)**: - Adopt the CCR method as prior information and flexibly handle different types of machine - learning algorithms through three stages: clustering, classification, and regression. - These prior information help to improve the accuracy and robustness of inverse modeling. 3. **aREKI Inverse Modeling**: - Use the aREKI method for inverse modeling. This method effectively solves the over - fitting and divergence problems in traditional Kalman filtering methods through adaptive regularization techniques. - The aREKI method can efficiently infer unknown reservoir parameters (such as permeability and porosity) from observed data and perform uncertainty quantification. ### Experimental Verification - The researchers first compared the PINO - CCR surrogate model with the traditional numerical black - oil reservoir simulator, and the results showed that the outputs of the two in the water - flooding case were very close. - Subsequently, they applied the PINO - CCR surrogate model to the aREKI history - matching workflow and successfully recovered the unknown permeability and porosity fields, and the simulation speed was 6000 times faster than that of the traditional method. ### Conclusions - The PINO - CCR surrogate model and the aREKI inverse - modeling method proposed in this study provide an efficient and accurate solution to the reservoir history - matching problem. - This method can not only quickly perform inverse uncertainty quantification, but also has good physical consistency and robustness, and is suitable for large - scale reservoir simulation and optimization tasks.

Reservoir History Matching of the Norne field with generative exotic priors and a coupled Mixture of Experts -- Physics Informed Neural Operator Forward Model

A Novel A.I Enhanced Reservoir Characterization with a Combined Mixture of Experts -- NVIDIA Modulus based Physics Informed Neural Operator Forward Model

Physics Informed Neural Networks Based on a Capacitance Resistance Model for Reservoirs Under Water Flooding Conditions

History-Matching of Imbibition Flow in Multiscale Fractured Porous Media Using Physics-Informed Neural Networks (PINNs)

Field Application of a Novel Multiresolution Multiwell Unconventional Reservoir Simulation: History Matching and Parameter Identification

Inversion Framework of Reservoir Parameters Based on Deep Autoregressive Surrogate and Continual Learning Strategy

A novel hybrid recurrent convolutional network for surrogate modeling of history matching and uncertainty quantification

End-to-end neural network approach to 3D reservoir simulation and adaptation

Neural Operator-Based Proxy for Reservoir Simulations Considering Varying Well Settings, Locations, and Permeability Fields

Uncertainty Quantification for Transport in Porous media using Parameterized Physics Informed neural Networks

Assisted Upscaling of Miscible CO2-Enhanced Oil Recovery Floods Using an Artificial Neural Network-Based Optimisation Algorithm

A Physics-Informed Spatial-Temporal Neural Network for Reservoir Simulation and Uncertainty Quantification

An Efficient Spatial-Temporal Convolution Recurrent Neural Network Surrogate Model for History Matching

Ensemble-based history matching of the Edvard Grieg field using 4D seismic data

Enhancing subsurface multiphase flow simulation with Fourier neural operator

4D Seismic History Matching Incorporating Unsupervised Learning

Accelerated Bayesian inference-based history matching of petroleum reservoirs using polynomial chaos expansions

Recent Progress on Reservoir History Matching: A Review

Reinforcement learning based automated history matching for improved hydrocarbon production forecast

Physics Informed Machine Learning for Reservoir Connectivity Identification and Robust Production Forecasting

The regularizing Levenberg-Marquardt scheme for history matching of petroleum reservoirs,