Artificial Neural Network to Predict the Interaction Between Natural and Hydraulic Fracture

The geometry of hydraulic fractures is a major concern, especially in highly fractured reservoirs, once it can severely affect oil productivity. Furthermore, interaction of hydraulic fractures with natural fractures can affect locally the in-situ stress state that may cause slippage of the natural fracture. In some cases, stress shadowing effects can even close the hydraulic fracture. Stress rotation around the natural fracture can make hydraulic fracture geometry complex, demanding more computational capacity and time to run numerical simulations. The neural network is a tool that can help real time daily operations when a quick decision is necessary and there is no time to run numerous simulations. This paper focuses on the development of an artificial neural network to predict the interaction between natural and hydraulic fractures. The artificial neural network predicts if hydraulic fractures cross or open natural fractures. In-situ stresses, fracture energy, friction angle of the natural fracture, and angle of approach between fractures are the required input for the artificial neural network, while mode of interaction such as opening, or crossing is the output. The database to train the neural network includes experimental data available in literature and results from numerical simulations. A considerable number of numerical simulations using interface elements to represent fractures provide a suitable database to build an accurate neural network. The neural network was trained for predicting fracture interaction with more than 90% accuracy for opening or crossing behavior. Results were compared with experimental data available in literature and numerical simulations. The results show that for higher angles of approach and higher stress differential, there is a strong tendency to fracture crossing. This is in agreement with the experimental conclusions presented by previous publications in the literature. Additionally, intermediary stress differentials, low values of cohesion and friction angle, and approach angle below 60° define conditions with a strong tendency to open the natural fracture. The scenarios predicted by the neural networks is in agreement with the rock mechanics concepts and with the expected tendencies, which gives more reliability to the neural network.