Produced-Water-Reinjection Design and Uncertainties Assessment

Abstract

Summary Produced-water reinjection (PWRI) is an important strategy for deriving value from waste water (Abou Sayed et al. 2005), but its implementation can face challenges related to injectivity and safety issues. Reinjection in a fracturing regime is often the only option to guarantee the sustainability of injectivity (Detienne et al. 2005), but it can present some risks to be anticipated early in the design phase. The first objective of a PWRI-design study is to supply water-quality specifications to petroleum architects (in terms of solid and oil contents) to allow for the design of water-treatment facilities. The second objective is to supply injection-pressure specifications for the design of injection pumps and an injection network. These specifications must allow for sustaining well injectivity over the life of the field, while preventing any risk of caprock failure. The water-quality specification is of prime importance because a maximum of contaminant injection is sought to minimize the cost related to water treatment, but at the same time, it must prevent any injectivity loss or excessive increase of pressure beyond which fracture confinement is no longer possible. Water quality and injection pressure are thus linked to each other. They are deduced by simulation on a case-by-case basis. The modeling approach used by Total was presented in previous publications (Detienne et al. 1998, 2005; Ochi et al. 2007). The objective of this paper is to detail the manner in which the two parameters are deduced when uncertainties on input data are considered. Indeed, a workflow for uncertainty management based on experimental design and Monte Carlo theories was implemented to combine the simultaneous effect of a relatively large number of uncertain parameters, each of them being characterized by its own probabilistic distribution. Two-thousand simulations were run systematically, and water-quality and injection-pressure specifications were supplied with a probabilistic value (P10, P50, and P90). Application of this approach to real-design examples is detailed and discussed in this paper.