Communications of the ACM
Using Sparse Data to Predict Lab Quakes
The relative rarity of stick-slip events makes them difficult to model with machine learning.
Credit: Dreamstime
Researchers at Los Alamos National Laboratory have developed a machine learning approach that can predict fault slip in laboratory earthquakes with reliability while accounting for sparse data due to infrequent major earthquakes.
The researchers used a transfer learning approach to train a convolutional neural network on the output of numerical simulations of lab quakes and a small dataset from lab experiments. They found the neural network accurately predicted fault-slip events when fed unseen data from a different experiment.
"We can simulate a seismogenic fault in earth, then incorporate data from the actual fault during a portion of the slip cycle through the same kind of cross training," says Los Alamos geophysicist Paul Johnson. The findings could help predict fault slip and potential earthquakes in the field.
From Los Alamos National Laboratory
View Full Article
Abstracts Copyright © 2021 SmithBucklin, Washington, DC, USA
No entries found