Abstract

In reservoir characterization studies, seismic inversion is carried out to generate elastic properties such as P-impedance, S-impedance, Vp/Vs ratio, and density. These properties, along with seismic-derived attributes, were used as inputs in linear regression or neural network methods to predict petrophysical properties such as porosity, volume of clay, and water saturation sequentially. Advanced machine learning techniques like Convolutional Neural Network (CNN) has been increasingly employed across various industries. However, CNN requires a large amount of sample data for efficient network training, which poses a significant challenge in exploration scenarios. This limitation is overcome by utilizing an Rock Physics Modeling (RPM) theoryguided approach, where geological variations are captured by simulating the thousands of synthetic wells under various scenarios, including changes in reservoir thickness, volume of clay, porosity, and water saturation