Dr Gleb Shishaev graduated in 2024 with a thesis History Matching and Uncertainty Quantification of Reservoir Performance with Generative Deep Learning and Graph Convolutions
Here is the first paper in a series of papers based on his thesis:
History Matching under Uncertainty of Geological Scenarios with Implicit Geological Realism Control with Generative Deep Learning and Graph Convolutions by G. Shishaev, V. Demyanov, D. Arnold just shared on arXiv:2507.1020 introduced Graph Variational Autoencoder (GWAE) approach for assisted history matching. It demonstrates how GWAE first constructs a latent space to describe geological uncertainty and then is capable to generate new models in the latent space that match both dynamic & static data and also retain geological realism as per the assumed geological concept.
Highlights:
- Proposes generative machine learning method with graph convolutions to describe geological features of different natures in a generic parametric space to perform history matching and UQ.
- Considers uncertainty of different geological scenarios while history matching through the latent space of decreased dimensionality
- Imposes implicit control of geological realism through the latent space allows for an efficient process of geological model history matching
- Geodesic metrics allows to work effectively in hidden space, increases efficiency and reliability of search and conditioning