Graph neural network is often used to depict connections within social networks, and most recently, covid pathways. Last week, Hart Energy hosted an illuminating webinar on reservoir management.

"Transform Waterflood Management Practices, using Graph Neural Network (GNN)"

https://lp.hartenergy.com...andRegistration.html

IF you can't wait to watch the entire video, here are the highlights - a hybrid approach combining physics-based and data-driven model, in which GNN nodes are injectors, producers, and can change over time! In the mean squared error sense, error is around ~8% water; 10% oil rate forecast. Once GNN deployed in the cloud, model can be trained in hours, and run in minutes to tune individual wells (e.g., shut-in / dial back) and optimize the overall field production.

More interesting is the Q&A session -

When certain categories of data: geology / subsurface model are not available, what to do?

A) 80/20 rule - use simple graph cross plot, estimate parameters form empirical field data.

B) new alternative - use GNN without physics and experiment to fit.

Can GNN work for multilaterals?

Consider representing each lateral with multiple nodes.

There is ample room for experimentation. But here is one simple use of Force Graph (no brainier, zero neural network ZNN) that may just get you thinking on the next deep learning adventure with real impact, M&A.

A picture paints 1,000 words, so with that in mind, have fun;
https://www.shaleforce.co...or-mix-visualization