Realistic reservoir models are essential for efficient field management and accurate forecasting of hydrocarbon production. Such models, based on the physical description of the reservoir, need to be calibrated or conditioned to historical production data. The process of incorporating dynamic data in the generation of reservoir models, known as history matching, is traditionally done by hand and is a very tedious, time-consuming procedure that, in addition, returns only one single matched model. It has been shown that the best matched model may well not be a good predictor of future performance.
In this work, one of the first field applications of the Neighbourhood Algorithm (NA) is presented. The NA is a stochastic sampling algorithm that explores the parameter space, finds an acceptable ensemble of data fitting models and extracts robust information from this ensemble in a Bayesian framework. The aim is to forecast hydrocarbon production accurately and to assess the related uncertainty by means of multiple reservoir models.
The NA methodology was extensively applied to an offshore gas field and compared to a previously manually matched model. The Mistral field has been producing for 6 years from 7 wells. Gas and water productions and pressure data were available and the uncertainty quantification was consistently obtained. Algorithm control parameters and objective function definition effects were investigated. The posterior probability density functions of each unknown parameter, calculated taking into account the observed production data, were evaluated. The hydrocarbon production was forecast using Bayesian inference and the economic risk estimated. The overall process was carried out with a significant time reduction compared with the previous manual approach.
The results presented suggest that use of stochastic sampling techniques in a Bayesian framework may well be a valid alternative methodology to the traditional industry workflow for the uncertainty quantification in producing fields.