Abstract
Conditioning of fault properties to production data is usually achieved by ad hoc modifications without consideration of prior probability distributions or physical realism. Here, we compare two geologically-based workflows for conditioning fault uncertainties to production data: a sensitivity analysis and optimisation scheme based on factorial experimental design and an alternative approach using the Ensemble Kalman Filter (EnKF). These methods have been tested on a producing reservoir whose combination of structure, well pattern and production history make it particularly suitable. The misfit between simulated and observed pressures was most influenced by reservoir volume (horizon depth, gas-water contact and fault positions) and inter-panel connection (fault juxtaposition and transmissibility). Using a response surface method, the best match to the observed data was obtained with a small connected volume and transmissible central fault. Application of the EnKF to fault displacement, permeability and thickness gave a similar match to the observed data; this was not improved by the inclusion of well oil production rate in the objective function. While comparable, the two optimisation methods each have different advantages and disadvantages. The method applied should be chosen in light of the particular field case and uncertainties being studied.
