Sweet Spot Identification of Carbonate Reservoirs Using Seismic Inversion and Three Machine Learning Algorithms

This study examined acoustic impedance derived from post-stack seismic inversion and machine learning for sweet spot identification in carbonate reservoirs. Sweet spots are critical for optimizing hydrocarbon exploration and production, particularly in challenging unconventional reservoirs.

Unconventional resources, such as tight gas, shale oil, and gas are becoming increasingly important for meeting global energy demands. Identifying these resources is vital for addressing the growing need for alternative sources.

This study applied two methods: post-stack seismic inversion using band-limited impedance inversion (BLIMP) and machine learning approaches using multilayer perceptron regression (MLPR), random forest regression (RFR), and extra tree regression (ETR) algorithms. It explores relationships between acoustic impedance and key parameters, including porosity (9), permeability(K), water saturation (Sw), total organic carbon (TOC), brittleness index (BI), reservoir quality index (RQI), and fracture zones (FZ).

By creating an objective function of square error between threshold and modeled parameters, and minimizing the error using L-BFGS-B optimization to determine the optimized empirical relations, this study aimed to identify promising sweet spots where water saturation fell below its thresholds, whereas porosity, permeability, TOC, RQI, and FZ exceeded their respective thresholds. This approach provides new insights into the role of acoustic impedance in unconventional characterization supporting resource optimization strategies.