Machine Learning Workflows - Seismic Inversion using AI - Machine Driven Seismic Inversion Workflow

Today, in our series on OpendTect Machine Learning workflows, we show a workflow for rock property prediction using real wells.
This workflow has many variations. You can train on real or synthetic seismic data to predict well log properties of interest.
In this case, we predict absolute Acoustic Impedance (AI) from synthetic seismic and low frequency AI. We compare the Machine Learning result with model-driven inversion (Bayesian Linear Inversion) and with stochastic inversion (HitCube trace matching inversion).
The model for Machine Learning inversion is XGBoost of Random Forest models (Scikit Learn). We have 200 Random Forests in our ensemble with a depth of 10. We have 4 wells in our study. We start with 4 runs in which we train on 3 wells and use the one left out for blind testing. This enables us to tune model parameters and to gain confidence in model predictions. When satisfied, we train on all 4 wells and apply the trained model on real seismic and low frequency model inputs to obtain
the desired full bandwidth AI.
The workflow shown in the video is recorded in version 7, which is due for release on 5 June 2023 (at the EAGE conference in Vienna - booth 4314).
To replicate this workflow in version 6, please be aware
of user interface differences.
We look forward to hearing your thoughts on this.
dgbes.com/blog...
/ marieke-van-hout-de-gr...
#OpendTect #geoscience #seismic #geophysics #geology #ai #machinelearning