Integrating Machine Learning for Log Prediction in Seismic Inversion Workflow

Abstract The use of machine learning (ML) to predict density and sonic logs from Gamma Ray (GR) logs has demonstrated a stable performance in blocky target reservoirs. Extreme Gradient Boosting is the most appropriate machine learning (ML) algorithm to predict density and sonic logs from Gamma Ray logs. Nevertheless, several iterations are necessary to stabilise the accuracy of the prediction. Additional data like formation tops and geographical coordinates were added to find a mathematical relationship between GR and density. Unfortunately, it is not possible yet to export the mathematical relationship that links GR to density. The purpose of the use of ML to predict density and sonic logs was to increase the number of anchor points for the Deterministic Seismic Inversion. That objective was achieved with a high coefficient of correlation and the absolute impedance from the inversion proved to have a better correlation to the previous study from the vintage. The target is a blocky carbonate reservoir that has an oil-bearing zone at WELL_B and WELL_A with a tilted oil-water contact. That anticlinal trap has 5% porosity cut-offs and no water saturation observed. The seismic inversion benchmark was useful to prove the concept of additional anchor points using ML log prediction for post-stack inversion. The target reservoir can be more easily mapped using the absolute acoustic impedance. A large absolute acoustic impedance volume was delivered within a relatively short turnaround time compared to previous legacy data.