Missing well log prediction using convolutional long short-term memory network

What is it about?

I propose a deep learning workflow to predict missing sonic logs from gamma-ray, density, and neutron porosity logs based on a convolutional long short-term memory network. The method takes into account the geologic trend and local shapes of logs. The network is trained on 177 wells from UK Continental Shelf and is applied on 5 different testing wells from different geological areas. The method also estimates the model uncertainties with 95% and 5% quantiles by using dropout layers at test time.

Featured Image

Why is it important?

The method does not depend on rock types and is not performed on interval-based like rock physics models. The method takes into account the geologic trend and local shapes of logs. The method estimated the model uncertainties that are useful for further analyses.