Enhanced Computer-Assisted Model Calibration of Mukhaizna Heavy Oil Field

Abstract

The Mukhaizna field is a giant heavy oil field under steamflood for the past four years. A sector with NW-SE elongate low relief anticlinal structure from a complex full-field geological model with 200,000,000+ cells of channelized braided-fluvial sandstone deposition was upscaled to a dynamic model with 400,000+ cells. This sector model approximately 15% of the field production area, from the crestal part of the field comprises 87 producers and 78 Limed Entry Perforated injectors with several years of production and steam injection history. A precursor for almost every reliable production forecast is a good history matched simulation model which becomes the cornerstone to optimize the decision making process. This paper describes an integrated and comprehensive simulation modeling workflow which was implemented to incorporate geological, petrophysical and dynamic uncertainties in a systematic way to calibrate the simulation model. Sensitivity analysis was performed to identify key static and dynamic input parameters. A computer assisted optimizer which applies experimental design and Tabu search techniques was used to test nonlinear and interaction effects of about 25 input parameters. This could be practically impossible using traditional sequentially modifying one-variable-at-a-time approach which always is cumbersome and may skew into an apparent solution even when none exists. During the iterative history matching process, uncertainties related to input parameters were constrained and modified from time to time to get an optimal solution. In addition to history matching 165 production/injection wells and 35 wells with pressure data, an attempt was made to match temperatures in 13 observation wells as well as heat breakthroughs in 70 patterns. The use of advanced techniques allowed us to minimize number of simulation runs and significantly reduced time to complete model calibration in less than couple of months which could traditionally take more than a year and still be limited in efficiently testing all possible combinations of input parameters.