SPE Workshop: Merging Data-Driven and Physics-Based Models for Enhanced Reservoir Insights and Predictions 19 - 20 Nov 2019 Hyatt Regency Hill Country Resort San Antonio, Texas, USA

Schedule |19-20 Nov 2019 | San Antonio, Texas

Tuesday, November 19

07:00 - 08:00
08:00 - 08:10
08:10 - 09:40
Session 1: Data-Driven Models: State of the Art
Session Chairpersons Jose Villa, Total E&P; Akhil Datta Gupta, Texas A&M University

There is a growing trend toward the use of statistical modeling and data analytics for oil and gas (and related subsurface domain) applications. This session will focus on the state of the art applications of converting data into information—particularly the actionable kind that lead to better decisions.

Presentation 1: Reservoir Flow Simulation, Data Analytics and the Gantner Hype Curve

Presenter: Larry Lake, University of Texas

Presentation 2: Robust Machine Learning for E&P Problems – Are We There Yet?
Presenter: Jared Schuetter, Battelle

Presentation 3: Changing Subsurface Engineering through Science-Informed Machine Learning
Presenter: Grant Bromhal, U.S. Deparment of Energy

09:40 - 10:00
10:00 - 11:30
Session 2: Combining Data-Driven and Physics-Driven Models
Session Chairpersons Pallav Sarma, Tachyus; Hector Klie, DeepCast.ai

Embedding physics into data driven models and /or extracing physical insights from data are becoming an integral part of machine learning approches in the Oil and Gas industry. Such approaches aim to allow better extraction of signals from noisy data, generate models with long term predcitive capacity,  enable models to be more interpretable and generalizable, and help in discovery of unexposed reservoir dynamics. This session explores new approaches to physics-based and data-driven models and combinations thereof that effectively embeds physical constraints, conservation laws and constitutive relations with data. The discussion will be showcasing several key challenges aimed at finding the best compromise between accuracy, efficiency and interpretability of a new generation of models.



Presentation 1: Deep-learning-based Surrogate Models for History Matching and Production Optimization
Presenter: Lou Durlofsky, Stanford University

Presentation 2: Deep Learning of Subsurface Flow via Theory-guided Neural Network.
Presenter: Haibin Chang, Peking University

Presentation 3: Fast, Deep Learning Based, Reservoir Simulation
Presenter: Jean-Marie Laigle, Belmont

11:30 - 13:00
13:00 - 14:30
Session 3: Delivering Innovative and Actionable Results
Session Chairpersons Xian Huan Wen, Chevron; Sebastien Matringe, Hess

This session presents various practical applications where machine-learning and physics-based modeling have been combined to offer an innovative technical solution to an existing technical challenge. Machine learning provides a fast and flexible framework to solve applied technical problems but physics-based modeling has the added advantage to contain additional information about the nature of the problem being solved. The presentations will demonstrate how combining the approach can offer superior insights that are timely and actionable for field development, reservoir management or operation. 


Presentation 1: Hybrid Modeling: Challenges and Opportunities in the Subsurface
Presenter: Shahram Farhadi , Beyond Limits

Presentation 2: Augmented AI in Reservoir Management Decision-Making
Presenter: Hamed Darabi, QRI

Presentation 3:  Physics Embedded Machine Learning for Modeling and Optimization of Oil and Gas Assets
Presenter: Pallav Sarma, Tachyus

14:30 - 15:00
15:00 - 16:30
Session 4: Data Quality and Emerging Data Types
Session Chairpersons Iraj Ershaghi, University of Southern California; Sathish Sankaran, Anadarko

This session focuses on new and emerging data types such as DTS/DAS, time-lapse seismic, novel tracers and others that go beyond traditional approaches for reservoir characterization, flow profiling, and production management. The technology implementation, data processing, analysis, interpretation and challenges of these new approaches are the focus of this session.


Presentation 1: Integrating DNA Diagnostic with Traditional Reservoir Workflows
Presenter: Hasan Shojaei, Biota

Presentation 2: Use of Surface Drilling Data to Inform Real-Time Geosteering/Lithology Identification
Presenter: Deepak Devegowda

Presentation 3: Measuring Inflow Distribution and Identifying Location of Water Influx Utilizing Intelligent Tracers Embedded in Completion Components
Presenter: Brock Williams


16:30 - 18:00

Wednesday, November 20

07:00 - 08:00
08:00 - 09:30
Session 5: Unconventional Reservoirs
Session Chairpersons Sathish Sankaran, Anadarko; Reza Rastegar, Oxy

Developing unconventional plays requires analysis of production data and understanding reservoir characteristics to forecast well performance and make better decisions about acreage selection, targeting, completions design, well spacing and operational strategy. This session covers state-of-the-art techniques and workflows for modeling unconventional reservoirs using a blend of physics-based and data-driven methods to derive meaningful insights. Discussions will include applications of these methods to operational fields along with their strengths and weaknesses and illustrate the importance of interpretability to support business decisions.


Presentation 1: AI-Based Field Development and Optimization in Unconventional Reservoirs  
Presenter: Hector Klie, Deepcast.ai

Presentation 2: Unconventional Data-Driven Well Performance Analysis
Presenter: Diego Molinari, Anadarko

Presentation 3: A Physics-Based Data-Driven Model for History Matching, Prediction and Characterization of Shale and Tight Reservoirs
Presenter: Yanbin Zhang , Chevron

09:30 - 10:00
10:00 - 11:30
Session 6: Understanding Reservoir Dynamics
Session Chairpersons Harun Ates, Devon

Understanding reservoir dynamics is paramount in order to successfully predict production behavior and optimize field development which has traditionally relied on integrated simulation modelling. With recent advances in data analytics and artificial intelligence, data-driven models have become more widely-used in analysis, predictive modeling, and optimization. This session will explore different ways, data driven and physics based models can be brought together for an enhanced understanding on factors that impacts dynamic reservoir behavior.  


Presentation 1: Physics Informed Deep Learning and its Application to Flow and Transport in Porous Media
Presenter: Cedric Fraces, Stanford University

Presentation 2: A Physical, Flow Regimes-Based Decline Curve for Unconventional Reservoirs
Presenter: Vincent Artus, Kappa

Presentation 3: Applications of the Diffusive Time of Flight to a Data Driven Approach to Decline Curve Analysis
Presenter: Mike King, Texas A&M University

11:30 - 13:00
13:00 - 14:30
Session 7: Geologic Insights and Reservoir Characterization
Session Chairpersons Omer Alpak, Shell; Harun Ates, Devon

Topics covered in this session encompasses the advances in developing predictive, data-driven, and portable static/dynamic reservoir models. The focus is on taking advantage of cutting-edge highly-scalable machine learning algorithms to rapidly generate or contribute to the construction of high-quality geologically-consistent reservoir models/model ensembles for reservoir forecasting. Envisioned applications include both exploration/appraisal/early-production-stage reservoirs and brownfield/late-production-stage reservoirs.   


Presentation 1: Getting the Best from Both Worlds: Combining Science with Data-Science
Presenter: Joe Lynch, Rock Flow Dynamics

Presentation 2: Machine Learning for Reservoir Characterisation Under Uncertainty: Explore, Describe, Update.
Presenter: Vasily Demyanov, Heriot-Watt University

Presentation 3: Uncertainty Centric Reservoir Management to Continuously Improve our Subsurface Understanding.
Presenter: Jon Saetrom, Resoptima

14:30 - 15:00
15:00 - 16:30
Session 8: Panel Discussion
Session Chairpersons Jose Villa, Total E&P; Hector Klie, DeepCast.ai

This panel discussion will address the promissing opportunities presented by combining the data driven models with physics-based models in reservoir management and field development optimization, and ideas of what the mid and long term future will hold.



Nathan Meehan, Gaffney, Cline & Associates / 2016 SPE President
Michael Davidson, Southwestern Energy
Eldad Haber, University of British Columbia
Graziella Caputo, IBM
Peter Wang, Anaconda