Error models for reducing history match bias

摘要

Successful reservoir prediction requires an accurate estimation of parameters to be used in the reservoir model. This research focuses on developing error models for flow simulation error within the petroleum industry, enabling accurate parameter estimation. The standard approach in the oil industry to parameter estimation in a Bayesian framework includes inappropriate assumptions about the error data. This leads to the parameter estimations being biased and overconfident. An error model is designed to significantly reduce the bias effect and to estimate an accurate range of spread. A 2D viscous fingering example problem will be used to demonstrate both construction of the error model, and the benefits gained in doing so.