How well can machine learning algorithms predict geology?
Use seismic modelling to find out!
In cooperation with RagnaRock Geo, NORSAR/NIAS developed a workflow to use ray-based synthetic pre-stack data as input to machine learning (ML) process in order to predict geological structure.
This has two main advantages one is the fact that we can easily generate a set of training data for the ML process. For that purpose, we are using a PSDM simulator that acts as a 3D pre-stack depth convolution method. The method is called simulated pre-stack local imaging (SimPLI) (Norway patent 322089, U.S. patent granted).
Furthermore, the knowledge of the geological structure shape and position allows to examine the accuracy of the predictions and investigate the factors that influence the prediction quality in a systematic and structural manner.
Combining state-of-the-art ray modelling method with deep learning techniques supporting seismic interpretation will be the next level to enhance interpretation quality and consequently minimize the risk of interpretation errors and drilling failures.