An accurate prediction of fracture gradient is a necessary requirement to economically and efficiently well designing, drilling operations, stimulation treatments and many other well operations. Underestimated or overestimated fracture gradient will result in some serious problems such as lost circulation of drilling fluid followed by formation fluid kick that leads to blowout and significant increase in nonproductive time and costs of a well. Artificial neural networks are efficient tools for predicting complicated events. In this paper, a feed-forward with back-propagation neural network has been developed to predict the fracture gradient as a function of pore pressure gradient, depth and rock density in one of southern oilfields in Iran. The results indicated that the developed ANN provides adequate approximation of fracture pressure gradient in the selected field.
Fracture gradient (FG) recognized as one of the most important parameters that its accurate estimation leads to achieving optimum well design, drilling operations, casing design, stimulation treatments and economical well planning. In addition, a precise fracture gradient is so important in selective production especially when the risk of communication between a hydrocarbon-bearing formation and water-bearing formation is high. As the formation is fractured, lost circulation takes place and hydrostatic pressure of drilling mud decreases that leads to hazardous well problems such as formation fluid kick followed by blowout and losing the well . Hence, predicting precise values of FG will reduce well problems and costs, significantly. Mainly, there are two methods to measure FG. One method is based on determination of the pressure needed to fracture the rock and is generally based on leak-off test data. Another method relies on correlations to predict FG . Hubbert and Willis developed an equation to predict the FG in areas of normal faults with the assumption that the minimum matrix stress coefficient is one-third of the overburden stress. Due to the results of their studies, overburden stress gradient, pore pressure gradient and Poisson''s ratio of rocks affect FG . Matthews and Kelly proved that matrix stress coefficient mainly depends on pore pressure and depth . Eaton expanded the work of Hubbert and Willis by introducing Poisson’s ratio and a variable overburden gradient whereas overburden gradient is a function of depth and rock density . Zamora presented a FG correlation as a function of rock density and depth . Artificial neural network (ANN) has seen a great increase of interest during the past few years. ANNs are powerful and efficient tools for solving practical problems whereas they have so many advantages over conventional techniques since they have the ability to find highly nonlinear relationships. The main objective of this paper was to introduce and develop a feed-forward with backpropagation neural network approach to predict fracture gradient in one of oilfields in Iran.
2. FIELD CHARACTERIZATION
The selected field is one of the most prolific oilfields in Iran located 60-km east of Ahwaz in Southern onshore of Iran. The formation is divided into seven zones which mainly composed of limestone, dolomite, sandstone and shale.