Data-Driven Discovery of Physics for Reservoir Surveillance

Nagao, Masahiro; Sun, Wenjian; Sankaran, Sathish

doi:10.2118/209300-ms

Cited by 6 publications

References 32 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

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

Nagao

Datta‐Gupta

Onishi

et al. 2023

SPE Reservoir Simulation Conference

View full text Add to dashboard Cite

Routine well-wise injection/production data contain significant information which can be used for closed-loop reservoir management and rapid field decisions. Traditional physics-based numerical reservoir simulation can be computationally prohibitive for short-term decision cycles, and also requires detailed geologic model. 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 can have limited predictive power. We propose a hybrid machine learning and physics-based approach for rapid production forecasting and reservoir connectivity characterization using routine injection/production and pressure data. Our framework takes routine measurements such as injection rate and pressure data as input and multiphase production rates as output. 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 it into a neural network as input. This physics-based input feature can reduce the model complexity and provide significant improvement in prediction performance. The second approach augments the residual terms in the neural network loss function with 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. The regularization allows the model to avoid overfitting and provides better predictive performance. Our proposed hybrid models are first validated using a 2D benchmark reservoir simulation case and then applied to a field-scale reservoir case to show the robustness and efficiency of the method. The hybrid models are shown to provide superior prediction performance than pure machine learning models and reduced physics models in terms of multiphase production rates. Specifically, in the second method, the trained hybrid neural network model satisfies the reduced physics model, making it physically interpretable, and provides 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 excellent agreement was obtained. By combining the reduced physics model with the efficacy of deep learning, model calibration can be done very efficiently without constructing a geologic model. The proposed hybrid models with physics-based regularization and preprocessing provide novel approaches to augment data-driven models with underlying physics to build interpretable models for understanding reservoir connectivity between wells and robust future production forecasting.

show abstract

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

Nagao

Datta‐Gupta

Onishi

et al. 2023

SPE Reservoir Simulation Conference

View full text Add to dashboard Cite

show abstract

Discovery of Unconventional Reservoir Flow Physics for Production Forecasting Through Hybrid Data-Driven and Physics Models

Zalavadia¹,

Sinha²,

Singh³

et al. 2023

SPE Western Regional Meeting

View full text Add to dashboard Cite

Routinely analyzing producing well performance in unconventional field is critical to maintain their profitability. In addition to continuous analysis, there is an increasing need to develop models that are scalable across entire field. Pure data-driven approaches, such as DCA, are prevalent but fail to capture essential physical elements, compounded by lack of key operational parameters such as pressures and fluid property changes across large number of wells. Traditional models such as numerical simulations face a scalability challenge to extend to large well counts with rapid pace of operations. Other widely used method is rate transient analysis (RTA), which requires identification of flow regimes and mechanistic model assumptions, making it interpretive and non-conducive to field-scale applications. The objective in this study is to build data-driven and physics-constrained reservoir models from routine data (rates and pressures) for pressure-aware production forecasting. We propose a hybrid data-driven and physics informed model based on sparse nonlinear regression (SNR) for identifying rate-pressure relationships in unconventionals. Hybrid SNR is a novel framework to discover governing equations underlying fluid flow in unconventionals, simply from production and pressure data, leveraging advances in sparsity techniques and machine learning. The method utilizes a library of data-driven functions along with information from standard flow-regime equations that form the basis for traditional RTA. However, the model is not limited to fixed known relationships of pressure and rates that are applicable only under certain assumptions (e.g. planar fractures, single-phase flowing conditions etc.). Complex, non-uniform fractures, and multi-phase flow of fluids do not follow the same diagnostics behavior but exhibits more complex behavior not explained by analytical equations. The hybrid SNR approach identifies these complexities from combination of the most relevant pressure and time features that explain the phase rates behavior for a given well, thus enables forecasting the well for different flowing pressure/operating conditions. In addition, the method allows identification of dominant flow regimes through highest contributing terms without performing typical line fitting procedure. The method has been validated against synthetic model with constant and varying bottom hole pressures. The results indicate good model accuracies to identify relevant set of features that dictate rate-pressure behavior and perform production forecasts for new bottom-hole pressure profiles. The method is robust since it can be applied to any well with different fluid types, flowing conditions and does not require any mechanistic fracture or simulation model assumptions and hence applicable to any reservoir complexity. The novelty of the method is that the hybrid SNR can resolve several modes that govern the flow process simultaneously that can provide physical insights on the prevailing multiple complex flow regimes.

show abstract

Hybrid Virtual Flow and Pressure Metering Approach Integrating Physics and Machine Learning Models for Accurate and Efficient Rates and Bottom Hole Pressure Prediction

Abudeeb,

Idris,

Abdullah

2024

Day 1 Mon, April 22, 2024

View full text Add to dashboard Cite

Summary Data collection is crucial in the Oil and Gas business. Reliable production and pressure measurements for oil and gas operations can be acquired from Multiphase Flow metering and bottom hole pressure metering. Accurate and continuous production and pressure measurements are crucial not only for field surveillance, but also for effective production optimization. Virtual Flow & Pressure Metering (VFPM) is an alternative to the conventional physical meters and can exist in two forms, physics-based VFPM and Machine learning-based VFPM. The physics-based VFPM relies on multiphase flow and pressure simulations including thermodynamics, fluid dynamics, fluid modeling, and optimization techniques. Machine learning VFPM however, uncovers the relationship between target variables and sensor data. Both methods have strengths but also have some limitations. The proposed algorithm is the hybrid VFPM that merges both types, the physics-based VFPM and the Machine Learning VFPM. The approach first tests each VFPM method independently and checks for time consumed and accuracy. Then, it tests for the hybrid model option. In the hybrid model case, the framework starts with the basic well data going into the Physics-based Model to simulate the well behavior and create sensitivity analysis. The generated data is then fed to the Machine Learning model. The process then considers correlation between input features and removes highly correlated features. Data is split into train and test and then, the regression model is built using multiple model options to choose the most suitable model based on accuracy metrics and elapsed time. The best model is picked and the target variables are generated. The model here was decision tree-based Machine Learning Model as it represented the data best and needed few hyper-parameters to tune. The effectiveness of the algorithm was tested and validated, achieving a high accuracy of 90% in predicting Multiphase Flow Rate and 98% accuracy in predicting bottomhole pressure, and a reduction of 45% in time consumed to generate the data. The algorithm's prediction reduces the time needed to generate data using solely physics-based VFPM while increasing the accuracy of the solely Machine Learning-based VFPM. In addition, this approach can translate into a huge cost saving since it eliminates the need for a physical meter in each well.

show abstract

Data-Driven Discovery of Physics for Reservoir Surveillance

Cited by 6 publications

References 32 publications

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

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

Discovery of Unconventional Reservoir Flow Physics for Production Forecasting Through Hybrid Data-Driven and Physics Models

Hybrid Virtual Flow and Pressure Metering Approach Integrating Physics and Machine Learning Models for Accurate and Efficient Rates and Bottom Hole Pressure Prediction

Contact Info

Product

Resources

About