Predicting Hydraulic Fracture Direction of Propagation When Intersect With Natural Fracture by Using Artificial Neural Network (ANN)

Successfulness of the hydraulic fracture treatment in unconventional reservoir
especially in shale gas reservoir was depends on the communication between hydraulic
fracture and natural fracture. Hydraulic fractures propagate across the reservoir during
the treatment and intersect with discontinuities present in the reservoir, in this case is
pre-existing natural fracture. At this point, several events might occur during
intersection. Firstly, hydraulic fracture propagation might step over the pre-existing
natural fracture. Secondly, hydraulic fracture is caught up by natural fracture and stop
the propagation. Thirdly, hydraulic fracture tip turn into the natural fracture, dilating
and opening the natural fracture as the fracture fluid infiltrate the natural fracture. In a
very low permeability reservoir, effective treatment should step over the pre-existing
natural fracture, extending the network deep into the reservoir, connecting all the
natural fracture to increase fracture conductivity and optimizing production of natural
resources especially in unconventional shale reservoir. Therefore, parameters that
characterized under which condition hydraulic fracture will step over and arrested into
natural fracture at the intersection point need to be study and fully understand for
designing the best hydraulic fracture treatment.
Parameters that affecting the course of fracture propagation, rock properties
and fluid properties, was determined and a set of input data was prepared by collecting
the data from the previous related research paper. Matlab software was used to develop
the artificial neural network (ANN) model that give prediction on the course of
hydraulic fracture propagation direction when intersecting with natural fracture by
mapping a set of input data to a set of output. The ANN model has been trained,
validated and tested by using 46 set of collected data and produced predicted output
with good accuracy. Mean squares error (MSE) and regression analysis was used to
calculate the output error to show the difference between predicted output and
observed output from the experiment. By using the same model, sensitivity analysis
was also conducted to see which parameter give the most effect and the least effect on
the fracture propagation during intersection.