Development of a Deep Learning Model to Estimate Porosity of a Fractured Granite Basement Reservoir

Deep learning (DL) can be seen as the latest development of ANN for various applications in almost all scientific, technological and social fields. The question whether or not DL can be useful in petrophysical analysis is therefore worth studying. In this study, a DL model was successfully developed using five well log data types to predict fracture porosity of a reservoir in the fractured granite basement of the Cuu long basin, Southern Vietnam. A two-step analysis was conducted, i.e., in the first step a single hidden layer ANN model was run; and in the following step a DL model was developed, using a new transfer function that is the rectified linear unit (ReLU), typical for a deep learning model to overcome the vanishing gradient problem that is one of the challenges in any ANN analysis, especially when dealing with a huge data amount. The final DL model employed in this study consisted of five input parameters, 12 hidden layers with 4 neurons per each layer. The predicted fracture porosity was found in the range from 0 – 0.26 %, which matches quite well with that estimated by the conventional procedure based on the average value of density and neutron porosities.