A Decision Tree Framework for Tracking CO2 Fronts in Carbonate Reservoirs from Deep Measurements Data

Abstract

Abstract One of the main challenges in tracking CO2 fronts in carbonate reservoirs is the geological and petrophysical heterogeneity and complexity. This obstacle led to uncertainties in the decision-making and tracking processes. In current practices, empirical petrophysical formulas are utilized to model CO2 fronts in carbonate formations. These formulas has shown tremendous results and advancements in addressing such complexity and they are the stepping-stone to petrophysical modeling. However, the challenge is continuously augmented as the data grow. We present an innovative artificial intelligence framework to estimate CO2 fronts in carbonate reservoirs based on deep measurement data such as electromagnetics surveys and acoustic impedance, together with porosity profiles. The framework processes the deep data and then sets it as training data. Afterward, the framework uses a decision tree model, specifically a regression tree, to estimate the CO2 saturation front map based on the training data. The framework tests four types of regression trees in which each tree has a different minimum leaf size. The framework was tested on synthetic deep measurement data such as electromagnetic surveys, acoustic impedance, and porosity profiles. Expert processing was also included in the processing stage to remove illogical measurements in the training data. The proposed artificial intelligence framework is applied to the training data set. The decision tree framework produces four types of regression trees with different minimum leaf sizes. These different trees are classified as ultra-fine, fine, medium, and coarse. Overall, all regression trees showed a high resemblance to the original CO2 front. However, the tree with the smallest minimum leaf size, which is the ultra-fine tree, has the highest accuracy and certainty. As for the computational time, the coarse tree is considered to be the fastest. This framework represents an innovative approach to track CO2 fronts by combining deep measurement data such as electromagnetic surveys, acoustic impedance, and porosity profiles with an artificial intelligence framework. Also, the framework is aimed to improve inter-well saturation mapping for CO2 sequestration application. The approach is data-driven, and it takes into account the existing data-scarce petrophysical model challenges.