Using differential evolution for compositional history-matching of a tight gas condensate well in the Montney Formation in western Canada

Abstract Production data analysis for low-permeability unconventional reservoirs is a challenging task, particularly for cases where multi-phase flow occurs within the reservoir. Analytical models developed to account for multi-phase flow typically require calculation of pseudo variables, which in turn require knowledge of relative permeability and fluid data. In the presence of sparse sampling, the analytical models often do not provide satisfactory results when there are so many unknown parameters. In such situations, numerical models are better suited, using a history matching framework to assist with reservoir and fluid characterization. In this work, we implement an assisted history-matching routine to characterize reservoir fluids and extract reservoir and hydraulic fracture properties for a hydraulically-fractured horizontal well completed in a tight gas condensate reservoir within the Montney Formation in western Alberta, Canada. The initial water distribution (e.g. movable water profile in the reservoir), in situ fluid (e.g. initial hydrocarbon composition with C 7+ properties) and reservoir properties (e.g. permeability in the matrix and around the fracture, and pressure dependent fracture permeability) are described in terms of 20 unknown parameters, which creates a high-dimensional inverse problem. We use the Differential Evolution algorithm, which is a powerful population-based optimization algorithm, and employ numerical compositional simulations to match pressure, water and hydrocarbon rates, and surface compositions of the produced fluids. Application of this optimization routine results in a good match to all measured data. The DE algorithm is repeated for an extra run to check for the existence of other non-unique solutions. The history-match results helped determine parameters for well/reservoir description and develop a compositional fluid model based on the measured separator composition data. The collected samples for both DE runs, along with one thousand extra samples from quasi-random sequence sampling design, provide a pool of data with invaluable information that are used to perform the global sensitivity analysis and to rank the contribution of each descriptive parameter on the variances of the reservoir outputs. In this way, the value of production data and surface compositions for the characterization of reservoir and fluid is quantified. This work aims to provide a practical and simple workflow for analysis of unconventional reservoirs where the direct analytical approaches cannot be applied.