Simulation of improved recovery processes from gas condensate and volatile oil reservoirs is significantly affected by the match of experimental PVT data. Therefore, a proper characterization of the mixture composition and tuning of the equation of state used are crucial for the accuracy of the reservoir model. This paper presents an efficient method for phase behaviour matching by non-linear regression. The constraints introduced by the boundaries of the regression variables are eliminated by a suitable change of variables. This leads to an unconstrained optimization problem, solved by an original implementation of the Levenberg–Marquardt algorithm. The C7+ fraction is described by a continuous (semi-infinite or truncated) distribution function. A new, rigorous generalized quadrature method is developed for the characterization of the continuous portion of the mixture, which can be applied to any distribution function.
The regression was applied to PVT data for several gas-condensate, including near-critical, and volatile oil systems subjected to different types of expansion and swelling tests, as well as multistage separation processes. The calculated results indicate very good agreement with experimental values. The regression algorithm proves to be robust and efficient.