Smart Identification of Petroleum Reservoir Well Testing Models Using Deep Convolutional Neural Networks (GoogleNet)

Abstract

Abstract Identification of reservoir interpretation model from pressure transient signals is a well-established technique in petroleum engineering. This technique aims to detect wellbore, reservoir, and boundary models employing an efficient matching process. The matching was first done manually; it then tried to be automated using artificial intelligence techniques. The level of uncertainty of matching outputs sharply increases, especially for noisy and incomplete signals. In this study, the pretrained GoogleNet (a novel combination of continuous wavelet transforms and deep convolutional neural networks) is used to decrease the uncertainty of matching results. Based on our best knowledge, it is the first application of GoogleNet to analyze transient signals in petroleum engineering. This technique is used to classify a relatively huge database, including synthetic, noisy, incomplete, and real-field signals. The GoogleNet can correctly discriminate among different reservoir interpretation classes with an overall classification accuracy of 98.36%. Moreover, it can successfully handle noisy, incomplete, and real-field pressure transient signals.