Falsification and Uncertainty Quantification of Natural Fracture Systems

Abstract:

Many properties of natural fractures are uncertain, such as the spatial distribution of fractures, petrophysical properties, etc. The Bayesian theorem is a commonly used framework to quantify the uncertainty in geological modeling and flow simulation. It uses the definition of a likelihood and prior to state a posterior from which samples are generated. This method is widely investigated in synthetic cases. In real cases, however, it is challenging to apply Bayesian theorem due to the falsification problem of the prior: often the stated prior model cannot predict the actual observations. This may be due to too small uncertainties or some missing physics. We use global sensitivity analysis to identify the problem when the prior model is falsified. The computational complexity of flow simulation among fractures is another challenge for real cases. We employ an approximate Bayesian computation method, in combination with a random-forest based surrogate model trained on the non-falsified prior to match the production history. We apply these two approaches to a complex real fracture oil and gas reservoir where all uncertainties are considered jointly, including petrophysical properties, rock physics properties, fluid properties, discrete fracture parameters, and dynamics of pressure and transmissibility. We successfully identified several aspects of the falsification. The uncertainties of the parameters are quantified and reduced based on the Bayesian theorem.