Time series clustering and class-based machine learning in predicting elastic properties of rocks: why, how, what, and so what

Time series clustering and class-based machine learning in predicting elastic properties of rocks: why, how, what, and so what

Shuvajit Bhattacharya, Ph.D., Bureau of Economic Geology, UT Austin

Multivariate time series clustering and class-based machine learning (ML) are relatively new concepts in geosciences; they have an immense potential to improve our models and provide more geologic insights than traditional baseline ML models. Seismic and wireline logs are a form of time series or depth series that share interdependence or conditional dependence with each other, depending on the rock type. Moreover, seismic and log data are highly redundant from an ML modeling perspective. We often do not consider these fundamental features of our datasets in ML models. This results in reduced explainability and troubleshooting of ML models and our models' failure when the boundary conditions change slightly. This talk will discuss the promises and challenges of semi-supervised time series clustering and class-based ML to solve these challenges. I will show an example of accurately and consistently predicting elastic properties of mudrocks using these concepts.

Biography: Dr. Bhattacharya is a researcher at the Bureau of Economic Geology, UT Austin. He is an applied geophysicist/petrophysicist by background. Prior to joining BEG, he worked with the University of Alaska Anchorage, Battelle, and other organizations in different roles, such as an assistant professor and petroleum geoscientist. He completed multiple projects for fossil fuel and geothermal energy exploration and carbon sequestration in the US, Australia, South Africa, and India. He has published over 50 technical articles in different journals and conferences.

To register, use this link: https://us02web.zoom.us/webinar/register/WN_yKwFn9ZGR3e1ddp-G22zTQ