4D Seismic History Matching of the Nome Field Model Using Ensemble-based Methods with Distance Parameterization

摘要

A distance parameterization of flood fronts derived from time-lapse seismic anomalies was recently developed to facilitate incorporation of time-lapse seismic data into history matching workflows based on ensemble methods such as the ensemble Kalman filter (EnKF) and the ensemble smoother (ES). A number of advantages were demonstrated on synthetic data including a significant reduction in the number of data points and flexibility in the type of attribute from which the front information can be extracted. In order to enable the use of the proposed method in real-field history-matching cases, we first extended the applicability of the algorithm computing the distance between observed and simulated fronts from regular Cartesian grids to generic corner-pint grids. Secondly, we used concepts from image analysis to generalize the innovations from the distance parameterization used in the EnKF as a directed local Hausdorff distance (from simulated to observed fronts) whereby a further improvement was achieved by taking into account the reverse measure (from observed to simulated fronts) as well. The workflow was subsequently applied to a series of numerical experiments on synthetic realistically complex test cases with promising results. The next step is a comprehensive examination on real field data as an objective of the study supported by National IOR Centre Norway. In this paper, we apply the history matching workflow to the Norne field where multiple high-quality seismic surveys were conducted. An iterative ES is used for history matching. The estimated model parameters include permeability, porosity, net-to-gross ratio, vertical transmissibility multipliers, fault transmissibility multipliers and saturation endpoints of relative permeability curves. The observations of front positions are acquired from an inversion of the Norne AVO seismic data set. Special attention is paid to the generation of the initial ensemble of reservoir models and the interpretation of inverted seismic data to ensure a proper estimation of the uncertainties for both model variables and data. The results show that additional benefits are received by matching to both production and 4D seismic data which contributes a better understanding of the reservoir and some new insights are gained regarding the performance of the proposed method. The outcomes of the application to the Nome field cases also suggest a couple of topics that are worth of further investigation. The order in which production and seismic data are incorporated, the localization approach, and for example parameterization of production data could all potentially improve the results.