Scientists use AI to improve carbon sequestration


A team of scientists has developed a new AI-based tool to include greenhouse gases such as CO2 in porous rock formations faster and more precisely than ever before.

Carbon capture technology, also known as carbon sequestration, is a climate change mitigation method that redirects CO2 emitted by underground power plants. The scientists must avoid excessive pressure build-up by injecting CO2 into the rock, which can disrupt geological formations and release carbon into aquifers above the site or even into the atmosphere.

A new neural operator architecture called U-FNO simulates pressure levels during carbon storage in fractions of a second, while doubling the accuracy on specific tasks to help scientists find optimal injection rates and locations. It was presented in a study this week Published in advances in water resourceswith co-authors from Stanford University, California Institute of Technology, Purdue University, and NVIDIA.

Carbon capture and storage is one of the few methods that industries such as refining, cement and steel could use to decarbonize and meet emission reduction targets. Over a hundred CO2 capture and storage plants are under construction worldwide.

U-FNO is used to accelerate carbon storage predictions for ExxonMobil, which funded the study.

“Reservoir simulators are intensive computer models that engineers and scientists use to study multiphase flows and other complex physical phenomena in the Earth’s subsurface geology,” said James V. White, director of subsurface carbon storage at ExxonMobil. “Machine learning techniques as used in this work provide a robust way to quantify uncertainties in large-scale subsurface flow models such as carbon capture and sequestration, ultimately facilitating better decision making.”

How carbon storage scientists use machine learning

Scientists use carbon storage simulations to select the right injection sites and rates, control pressure build-up, maximize storage efficiency, and ensure injection activity does not disrupt the rock formation. Understanding the carbon plume – the spread of CO – is also important for a successful storage project2 through the floor.

Traditional carbon sequestration simulators are time consuming and computationally intensive. Machine learning models provide similar levels of accuracy while dramatically reducing the time and cost required.

Based on the U-Net neural network and the neural Fourier operator architecture known as FNO, U-FNO provides more accurate predictions of gas saturation and pressure build-up. Compared to using a state-of-the-art convolutional neural network for the task, U-FNO is twice as accurate and requires only a third of the training data.

“Our machine learning approach to scientific modeling is fundamentally different from standard neural networks, where we typically work with fixed-resolution images,” said paper co-author Anima Anandkumar, machine learning research director at NVIDIA and Bren- Professor of Computing + Department of Mathematical Sciences at Caltech. “In scientific modelling, we have different resolutions depending on how and where we sample. Our model generalizes well across different resolutions without requiring retraining and achieves tremendous speedups.”

Trained U-FNO models are available in a internet application Providing real-time forecasts for carbon storage projects.

“Recent innovations in AI using techniques such as FNOs can speed up computations by orders of magnitude and make an important step in scaling carbon capture and storage technologies,” said Ranveer Chandra, Microsoft’s managing director of research for industry and collaborator at northern lights initiative, a comprehensive carbon capture and storage project in Norway. “Our model-parallel FNO can be scaled to realistic 3D problem sizes using the distributed memory of many NVIDIA Tensor Core GPUs.”

Novel neural operators accelerate CO2 memory predictions

U-FNO allows scientists to simulate how pressure levels build and where CO2 will spread over the 30 years of injection. GPU acceleration with U-FNO makes it possible to run these 30-year simulations in 1/100th of a second on a single NVIDIA A100 tensor core GPUinstead of 10 minutes with conventional methods.

With GPU-accelerated machine learning, researchers can now also quickly simulate many injection sites. Without this tool, selecting websites is like a shot in the dark.

The U-FNO model focuses on modeling plume migration and pressure during the injection process – when there is the highest risk of exceeding the CO level2 injected. It was developed with NVIDIA A100 GPUs in the Sherlock computing cluster at Stanford.

“To reach net-zero, we need low-emission energy sources, as well as negative-emission technologies like carbon capture and storage,” said Farah Hariri, U-FNO associate and technical lead of carbon offset projects for NVIDIA’s Earth-2, the world’s first AI supercomputer with digital twin. “By applying neural Fourier operators to carbon storage, we have shown how AI can help accelerate the process of mitigating climate change. Earth-2 will use these techniques.”

Read more about U-FNO on the NVIDIA Tech Blog.

Earth-2 will use FNO-like models to address challenges in climate science and contribute to global efforts to mitigate climate change. Learn more about Earth-2 and AI models used for climate research in NVIDIA Founder and CEO Jensen Huang’s GTC keynote:


Comments are closed.