oa Choosing Objective Function for Conditioning on Production Data

In this paper we consider the problem of choosing an objective function, or likelihood function, when conditioning a stochastic reservoir model on production data. We argue for a pragmatic approach where the objective function also should be judged upon its predictive value. To find candidates for objective functions, we propose that the correlation structure of production data under the prior distribution should be taken into consideration. Three different objective functions are compared; two of them are based on prior correlations while the third is a weighted sum of squares. We show by a realistic example that taking advantage of the correlation structure may increase the predictive value.