The goal of all history matching for the reservoir models is to generate accurate predictions. There is also an assumption that a better history match should lead to a more accurate prediction, which can lead an asset team to concentrate on a single model.
The assisted history match techniques allow us to generate multiple models that satisfy the surveillance 1, and understand a range of outcomes for a set of models. What we need is to provide assurance that we are searching correctly for the range in outcomes, to understand the interaction between uncertainty, surveillance, and the business decision.
This paper describes an example for the Mahogany field in Trinidad, which is mainly gas with a thin oil rim. As the reservoir pressure is decreasing, there is a window of opportunity to drill an additional well.
The first task is to determine the correct value of a new gas well, to protect against water encroachment. The Mahogany field used BP's Top Down Reservoir Modeling (TDRMTM) to test a hypothesis - is it possible to find models that match history, and have certain key features in the future? In this case we are specifically looking for a scenario with water at the existing gas well locations. Since the system is generating discrete models, we can then test which new gas well locations would mitigate the water.
We've also been able to use an idea from our Alaskan operations, and consider water injection with a dump flood. The appraisal of the water injection has been performed using computer assisted depletion planning tools (TDDPTM 2) to assess the best location and strategy for such a well.
The key task for a reservoir simulation is providing an accurate estimate of future performance for existing and planned wells. Unfortunately, the reservoir model will always be coarser than reality due to simulation constraints, and contain approximations in the geology due to observational limitations. As a general rule, the only thing that can be said of a single model is that it will be wrong in some detail, and the use of a single model requires the subsurface team to know whether the errors are significant and whether there will be a surprise in the future.
With the development of multiple models, we can test what other reservoir descriptions also satisfy the observational constraints 4. If the range of alternatives has been set up correctly, then we should be able to determine the probability of certain outcomes, and alter our depletion plan to test whether the reservoir description is possible, and what our best strategy should be in that case.
We shall use the framework provided by an assisted history matching tool to test discrete alternative cases.