CO2 Prediction of the Undrilled Prospects of the Arthit Field via Geological-Informed Machine Learning

Abstract

High Carbon dioxide (CO2) content presents a serious challenge in the development of Arthit Field. Accurate resource estimation, especially in the deep reservoir sections (Lower Miocene - Oligocene), depends on the accuracy of CO2 prediction. Formulated as a manual clustering approximation, conventional CO2 prediction requires intensive labor and fails to re-calibrate the model once the latest information is acquired. This paper introduces the application of machine learning concepts to the prediction of CO2. The proposed CO2 prediction methodology leverages machine learning techniques to enhance the understanding of a known field with existing CO2 concerns. Compared to the conventional manual clustering method, the machine learning model improves accuracy and reduces time and cost in the process of CO2 prediction and, in turn, resource estimation. Although our methodology is demonstrated specifically for CO2 in Arthit Field, it is equally applicable to other parameters and fields.