Errors in History Matching

Abstract

The usual procedure in history matching is to adopt a Bayesian approach with an objective function that is assumed to have a single simple minimum at the "correct" model. In this paper, we use a simple cross-sectional model of a reservoir to show that this may not be the case. The model has three unknown parameters: high and low permeabilities and the throw of a fault. We generate a large number of realizations of the reservoir and choose one of them as a base case. Using the weighted sum of squares for the objective function, we find both the best production-and best parameter-matched models. The results show that a good fit for the production data does not necessarily have a good estimation for the parameters of the reservoir, and therefore it may lead to a bad forecast for the performance of the reservoir. We discuss the idea that the "true" model (represented here by the base case) is not necessarily the most likely to be obtained using conventional history-matching methods. Reservoir Model and Objective FunctionWe use a simple 2D cross-sectional model of a layered reservoir, as shown in Fig. 1. 57 The model is 2D, with no-flow boundary conditions. Water is injected on the left side, and a production well is located on the right. There is a single, normal fault at the midpoint between the two wells. The reservoir contains six alternating layers of good-and poor-quality sands with high and low permeabilities, respectively. The porosities of the good-and poor-quality sands are 0.30 and 0.15, respectively. We also assume that the three good-quality layers have identical properties, and the three poor-quality layers have a different set of identical properties. The thickness of the layers has an arithmetic progression, with the top layer having a thickness of 12.5 ft, the bottom layer a thickness of 7.5 ft, and with a total thickness of 60 ft. The width of the model

Controllability, observability and identifiability in single-phase porous media flow

“…28-29 are not identifiable. Often called ill-posed, this lack of identifiability in Problem 1 is mentioned in almost all publications on history matching and is problematic because a wrongly updated estimate θ up ofθ can lead to a perfect history match (i.e., V(θ up ) = 0) but incorrect long-term predictions (e.g., when the saturation front has significantly advanced)-see [34].…”

How Does Sampling Strategy Affect Uncertainty Estimations?

“…The IC Fault model is a simple 3-parameter model set up by Tavasoli et al [17] as a test example for automated history matching. It has proved extremely difficult to history match and Carter et al [25] concludes that the best history matched model obtained from an exhaustive run (159, 661 models) is a very poor predictor.…”

“…We use Genetic Algorithms (GA) and the Neighbourhood Algorithm (NA) to generate an ensemble of reservoir models for a synthetic reservoir case, the IC Fault model [17]. These algorithms also have tuning parameters that can be adjusted to change the search behaviour, e.g.…”

Implementation of Region-Based Covariance Localization Ensemble Kalman Filter in History Matching Workflow

“…The aim of history matching is to generate the most appropriate reservoir model that would provide prediction as close as possible to the acquired production data [1]. Static parameters in a reservoir model that usually need to be tuned in history matching process are skin factor, permeability thickness, vertical to horizontal permeability ratio and aquifer constant.…”