Refined Adaptive Gaussian Mixture Filter - Application on a Real Field Case (SPE 154479)

Over the last decade the ensemble Kalman filter (EnKF) has attracted attention as a promising method for solving the reservoir history matching problem: Updating model parameters so that the model output matches the measured production data. The method possesses unique qualities such as; it provides real time update and uncertainty quantification of the estimate, it can estimate any physical property at hand, and it is easy to implement. The method does, however, have its limitations; in particular it is derived based on an assumption of a Gaussian distribution of variables and measurement errors. A recent method proposed to improve upon the original EnKF method is the Adaptive Gaussian mixture filter (AGM). The AGM loosens up the requirements of a linear and Gaussian model by making a smaller linear update than the EnKF and by including importance weights associated with each ensemble member at computational costs as low as EnKF. In this paper we present a refined AGM algorithm where the importance weights are included in the calculation of the apriori and the aposteriori covariance matrix and we also present results where this algorithm is combined with distance based localization. Moreover, in this paper the AGM algorithm is for the first time applied to a real field study. To validate the performance of AGM the result is compared with the EnKF, with and without distance based localization. Several statistical measures are used to validate the performance of the filters, and we are able to distinguish the performance of the different filters. In particular all the methods provide good history match, but we see that the AGM stands out by better honoring the original geostatistics.