The goal of this industry sponsored project is to develop practical methods to quantify uncertainty in reservoir performance forecasting.

The approach we are developing is to construct multiple models matching known reservoir performance data (rates, pressures etc) which are also consistent with prior geological beliefs. Because our knowledge of the reservoir properties is from a limited set of sample points, there will always be multiple reservoir models consistent with our knowledge, and our reservoir model predictions will be uncertain.

• Steering Group Meetings
  (sponsor access only)
  
  
• Computational Resources
• Confidence in Prediction
• Framework
• Generating Multiple Models
• Papers Published (pdf format)
• Results
• Sponsor Acknowledgement

We use geostatistical approaches to generate multiple reservoir descriptions which are samples from the prior distribution. This requires the model to be parameterised by a set of values. These could be input parameters to a geostatistical package, for example variance and correlation length.

We use an algorithm adapted from earthquake seismology – the Neighbourhood Approximation (NA) Algorithm – to choose which values the parameters should have.

The goal of the NA Algorithm is to improve the performance of the sampling slope by concentrating effort in the regions of good fit. An example of the performance of the NA Algorithm is shown in Figure 2. The NA algorithm finds the regions in parameter space which yield good history matched reservoir models.