Acoustic impedance and lithology-based reservoir porosity analysis using predictive machine learning algorithms

Porosity prediction and analysis is crucial for reservoir delineation, characterization and well placement. Basically, porosity measurements are obtained from well logs and core samples. However, methods based on core samples and well logs are sometimes challenging, time-consuming and very expensive. In this paper we attempt to predict and analyze porosity using machine learning algorithms. Porosity prediction is performed as a supervised multiple regression problem. Every data point of the training sample from the study area is made of lithology and acoustic impedance which are defined as independent variables or predictors while measured value of porosity is the dependent variable, provided as the label to be predicted. Deep neural network, Random forest and Decision tree algorithms were subjected to training and learning of rich and proper features that are important for the prediction of porosity thus, good generalization ability is crucial to the successful training of a machine learning model. Therefore, we used several optimization functions to train the deep neural network in order to choose the best performing one, on which we retrained the model using k-fold cross-validation technique. Although all the algorithms showed very good performance, deep neural network proved to be the most efficient with 0.042% of the mean squared error as the learning loss and 0.051% of the training mean absolute error. Our result was further tested for effectiveness by using new set of non-labeled lithology and acoustic impedance data and the result was compared with measured porosity and the result showed 89% of the Pearson correlation. Compared with measured porosity, prediction results displayed accurate prediction of porosity within milliseconds thus saving the working time. The result of this study shows that machine-learning algorithms can be time saving and reliable alternative for reservoir porosity prediction and analysis.