Progress Status, June 2010 (pdf, 641K)

Further project details in the report section (sponsor access only)

In Phase II, we aimed to enhance and extend the techniques developed in Phase I and apply to real field examples. Rather than summarise every achievement in Phase II (this will be done in the final report for each module), we will summarise here the successes in field applications.

In Phase II, we aimed to enhance and extend the techniques developed in Phase I and apply to real field examples. Rather than summarise every achievement in Phase II (this will be done in the final report for each module), we will summarise here the successes in field applications.

Phase I of the Uncertainty Project largely developed, and demonstrated the feasibility of, the approaches we currently use to quantify uncertainty. Application was primarily to synthetic data, although 2 real field examples were published by BP and ENI using our techniques.

We have applied our techniques to 4 fields – 3 within the framework of the Uncertainty Module, and 1 within the Geological Parameterisation Module.

The Rigel Field example – supplied by ENI – was used to history match both full field models and models generated using Top Down Modelling concepts with matching performed at well level. ENI supplied history data to only 2003. The matched models were then used to forecast uncertainty for 2003 – 2007 well-level performance, where production data was not available. After the forecasts had been made, we were supplied with the historic data, and were able to compare predictions with actual field performance. The forecast was good for most wells, but not all. The study showed the importance of two topics studies during Phase II, namely the Model Inadequacy term in the definition of misfit and the value of production data analysis for determining misfit parameters and form consistently.

The BP field example (published in Demet Erbas’s PhD thesis) compared GA & NA history matches to a compartment of a North Sea field to examine the economics of an infill well decision. Again, production data analysis was used to determine the appropriate misfit form. The results showed that different conclusions about the economic success of the infill well would be drawn from the use of GA or NA or GA with an alternate misfit definition proposed by BP for this field. The significance of this result is that it is important to check for convergence of the stochastic sampling algorithm – none of the example presented had could be considered to have converged – and also to assess the impact of the assumptions made in defining the form of the misfit and the parameters within the misfit definition.

We also have two recent applications. The first is a Chevron field with over 50 wells. Data analysis was used to determine the misfit function, and multiple models were generated using NA compared with only one found by use of a commercial package. The second example is an application of the geological parameterisation concept to a real reservoir.