A HIERARCHICAL ASSISTED HISTORY MATCHING APPROACH WITH GLOBAL AND LOCAL PARAMETER UPDATES

Traditional manual history matching commonly follows a structured approach with a sequence of adjustments from global to regional parameters followed by local changes in model properties. In contrast, much of the automatic history matching methods utilize parameter sensitivities or gradients to directly update the fine‑scale reservoir properties. We present a hierarchical assisted history matching approach that combines elements of both manual and automatic history matching in a structured framework. First, a probabilistic approach is used to estimate uncertainty the large‑scale static and dynamic parameters. This is followed by a sensitivity‑based deterministic model calibration for local property changes. In the probabilistic global calibration, design of experiments and response surface methodologies with evolutionary algorithms are used to calibrate global parameters, for example regional pore volumes, vertical communications, fault transmissibilities and aquifer strength. Key global parameters are first identified via sensitivity analysis and followed by proxy model construction using experimental design and response surface analysis. An improved genetic algorithm with heat‑bath sampling strategy is used to generate updated ensemble of models conditioned to static pressures (MDT/RFT) and total liquid rates at the wells. Next, each ensemble member is updated using water‑cut, GOR and flowing BHP via sensitivity‑based local calibration. We utilize streamline‑derived analytic sensitivities to determine the spatial distribution and magnitude of the local changes. The proposed approach was tested by a 3D synthetic case and a field application. The synthetic example is the benchmark PUNQ‑S3 reservoir model consisting of 6 producing wells and a strong aquifer support. The field example is an offshore turbidite reservoir with highly detailed production and pressure information. The static model contains complex sand depositional distribution combined with fault structures, four pairs of injector, deviated producing wells and more than 8 years of production history.