Prediction of the Maximum Horizontal Principal Stress from Breakout Data Using Generative Adversarial Networks and Backpropagation Neural Network

Abstract

Summary A good understanding of the magnitude and direction of in-situ stresses is very important for oil and gas exploration. The conventional wellbore breakout method directly uses information about rock strength and wellbore shape (i.e., depth and width of breakout) to predict the in-situ stresses, but it is difficult to accurately describe the relationship between the breakout shape and the in-situ stresses. This paper presents a deep learning model, combining the generative adversarial networks (GAN) and backpropagation neural network (BPNN) to predict the maximum horizontal principal stress (MHPS) from breakout data. First, a GAN is used to effectively improve the quantity and quality of training data by generating a certain number of new training data that approximate the original data. Second, the training data enhanced by the GAN are used to train the BPNN, which predicts the MHPS based on wellbore breakout geometries. The two independent modules, the GAN and BPNN, use the training data to train themselves, respectively. This dual deep learning pattern ensures that the potential relationship between the in-situ stresses and wellbore breakout shape can be found. To examine the reliability of this technique, 86 sets of laboratory data from published literature are used to train the model, and 19 sets of laboratory data from other published literature are used to test the prediction performance of the trained model. The results show that the proposed model has good accuracy with an average relative error of 4.76% when predicting the MHPS. In addition, this deep learning model combining the GAN and BPNN requires only a few seconds to run on a laptop computer, thus providing an effective and efficient tool for predicting the MHPS.