Accurate formation fracture gradient prediction is an essential part of well planning. Erroneous fracture gradient estimates may jeopardize the entire drilling operation and result in serious well problems, the least of which are lost circulation and kick leading to blowout. Accurate fracture gradient values play an important role in the selection of proper casing seats, prevention of lost circulation and planning of hydraulic fracturing for the purpose of increasing well productivity in zones of low permeability. Furthermore, a good knowledge of the fracture gradient is of great importance in areas where selective production and injection is practiced. In such areas the adjacent reservoirs consist of several sequences of dense and porous zones such that, if a fracture is initiated during drilling or stimulation, it can propagate and extend, establishing communication between hydrocarbon reservoirs and can extend to a nearby water-bearing formation.
Fracture gradient depends upon several factors including magnitude of overburden stress, formation stress within the area and formation pore pressure. Any prediction method should incorporate most of the above factors for a realistic prediction of the fracture gradient.
This paper presents an artificial neural network model that yields reasonably accurate values of the fracture gradient. The input training data are actual field data. The results obtained from the model are compared with those obtained from correlation. The comparison shows that the method is promising and under some circumstances it is superior to the available techniques.