Science-informed Machine Learning for Subsurface Applications

Stored: Science-informed Machine Learning for Subsurface Applications

Science-informed Machine Learning for Subsurface Applications

Type	Program
Sponsor Organization	Office of Fossil Energy and Carbon Management
Top Organization	Department of Energy
Creation Legislation	None
Website	Website
Purpose	The Science-informed Machine Learning for Subsurface Applications program develops AI and machine learning tools to enhance real-time decision-making for subsurface energy systems like carbon storage and geothermal energy. It aims to integrate physics-based models with data-driven approaches to improve forecasting, monitoring, and management of subsurface resources.
Program Start	2020
Initial Funding	$10 million
Duration	Ongoing
Historic	No

The Science-informed Machine Learning for Subsurface Applications (SMART) initiative is a Department of Energy program launched in 2020 under the Office of Fossil Energy and Carbon Management (FECM) to harness artificial intelligence (AI) and machine learning (ML) for real-time subsurface energy management.

Led by the National Energy Technology Laboratory (NETL) with partners like Battelle and multiple universities, SMART integrates physics-informed ML (PIML) with field data from DOE-supported sites—e.g., carbon storage pilots—handling over 1 petabyte of data across its phases to advance tools for reservoir forecasting and virtual learning.

Phase 2, completed by 2023, emphasized deep learning for 3D spatiotemporal models, with Phase 3 (2024 onward) focusing on uncertainty quantification, building on over 20 years of DOE subsurface research.

Official Site

Goals

• Develop real-time forecasting tools to optimize subsurface operational decisions.
• Create virtual learning environments for improved field development and monitoring.
• Enhance subsurface science with scalable, interpretable AI/ML models.

Organization

SMART is managed by FECM through NETL, with Srikanta Mishra (Battelle) as Technical Lead overseeing a consortium of national labs (e.g., LANL, PNNL), universities (e.g., Penn State), and industry partners. Funding from FECM supports cross-disciplinary teams, leveraging NETL’s Energy Data eXchange (EDX) for data curation and high-performance computing resources like Joule for ML development.

History

SMART began in October 2020, driven by the need for faster subsurface decision-making amid growing data from DOE’s carbon sequestration and fossil energy projects. Phase 1 (2020-2022) built foundational ML models using historical data, Phase 2 (2022-2023) advanced deep learning for real-time applications, and Phase 3 (2024-2025) targets uncertainty and scalability per NETL’s January 31, 2024, update. It continues to evolve, supporting DOE’s Energy Earthshots with plans for broader subsurface applications.

Funding

SMART launched with $10 million in 2020 from FECM, with subsequent phases funded through FECM’s R&D budget—e.g., $13 million in FY 2023 for Phase 2. Ongoing support, including the 2024 Phase 3 kickoff, totals over $26 million across teams, funding data processing, tool development, and partnerships with no set end date.

Implementation

SMART deploys its tools via EDX, using PIML techniques like physics-informed neural networks (PINNs) to process petabyte-scale data from field labs. It progresses in phases: Phase 1 curated data, Phase 2 deployed convolutional neural networks for forecasting (web:2), and Phase 3 enhances uncertainty tools, with ongoing adaptation to subsurface challenges like carbon storage and geothermal systems.

Related

• Energy Data eXchange
• Feedstock-Conversion Interface Consortium

External links

• https://edx.netl.doe.gov/smart/
• https://www.energy.gov/fecm/smart-science-informed-machine-learning-accelerating-real-time-decisions-subsurface
• wikipedia:Machine learning

Social media

• https://twitter.com/NETL_DOE
• https://www.linkedin.com/company/netl-doe