A novel approach to predict interaction between hydraulic fracture and natural fracture using artificial neural networks

Hydraulic fracturing is a common applied stimulation method to develop the tight gas resource which is widely distributed all over the world. A lot of field measurement and experimental observation have shown the hydraulic fracture (HF) complexity in the environment of tight sandstones. This complex fracture is strongly influenced by the natural fractures (NF) which are inherent in the tight gas formation. It was documented that, as a result of interaction, the hydraulic fracture will cross, dilate or slip into the natural fracture. Though some researches have been done, the interaction mechanism is still not fully understood. The lack of knowledge in this topic calls for further research in it. With this study, two neural networks methods are used to establish models to predict the interaction pattern between HF and NF. Artificial neural networks have proven to be excellent predictive tools in various petroleum-engineering applications. The excellent predictive capability of artificial neural networks comes from the fact that neural networks have large degrees of freedom that allows them to capture the non-linearity of the system being studied better than conventional regression techniques. The back-propagation neural network (BPNN) and probabilistic neural network (PNN) which display strong mapping ability are chosen to establish the prediction models. These models can be used to predict whether a HF will cross, dilate or slip into a NF under certain conditions, i.e. under different approach angles, different differential horizontal stress, net pressure and so on. Results have shown accuracy in predicting the interaction pattern. Conventional HF design is based on the assumption that the rock is homogeneous and the fracture propagates symmetrically in a plane perpendicular to the minimum stress. In naturally fractured reservoirs due to interaction with NF, the fracture may propagate asymmetrically or in multiple strands or segments. The result from this work can be applied to estimate the interaction picture thus update and optimize the fracturing design.