← Back to Use Cases

Making the Right Decision Given Large Uncertainty

Name Affiliation Phone Number Email Address
Jonathan Carter E.ON Exploration and Production jonathan.carter@eon.com
Industrial Sectors:

Oil & Gas


The problem described below is a cartoon of a real problem from the petroleum industry. We have a reservoir that has been producing oil for 36 months, it has now started to produce water as well. Should management spend a large amount of money to shut-off those parts of the well that are producing water, and which sections should be shut. The model is a simple vertical cross section through a reservoir. On the left hand side there is an injector, and on the right hand side is a producer. Mid-way between the wells is a fault as shown in the figure. The geology is an alternating sequence of good and poor quality sand.

2.1 Process Inputs

[caption id="attachment_239" align="aligncenter" width="415"]Figure 1: Model of Reservoir Figure 1: Model of Reservoir[/caption] The data available consists of 108 measurements of oil production rate, water production rate and water injection rate, and there are three uncertain parameters that need to be determined from the data: permeability in each of the two sand types and the fault throw.

2.2 Propagation

The challenges are:

  • Build a good proxy model of the reservoir behaviour
  • Perform a probabilistic inverse problem to determine most likely intervals for the three parameters.
  • Make a recommendation for the best management action and predict the performance over the next year.

2.3 Interpretation and Communication of Results

In addition to the technical questions described above we have two human interaction issues related to this work:

  • What is the best practice to obtain realistic prior information about the uncertainty from our subject experts.
  • What are the most appropriate techniques to communicate the uncertainties to senior decision makers who have no training in probabilistic forecasting and may be hostile to non-deterministic modelling.


The model is known as the IC Fault Model and can be found at https://www.imperial.ac.uk/engineering/departments/earth-science/research/research-groups/perm/standard-models/