Ensemble-based methods have been applied successfully to the history-matching of conventional oil and gas assets for many years, while few published applications exist to unconventional reservoirs.
Coal seams are naturally fractured reservoirs, where gas is adsorbed in the coal matrix, and cleats are filled with water. The Coal-bed Methane (CBM) production is achieved by reducing the fracture system's pressure that causes the gas to desorb from the coal matrix. CBM production maximization requires minimization of the reservoir pressure, which results in different dynamics compared to the conventional oil and gas reservoirs.
Typically, CBM field developments require many wells. The lower permeability coals are often developed with open-hole horizontal wells, which may have multiple laterals intersecting vertical wellbores. This introduces additional history-matching challenges as the effective contributing length of the wellbore, formation damage, and effective connection between the well and the reservoir are uncertain.
This study presents the iterative ensemble Kalman smoother application to a low permeability CBM field in Australia. All the wells in the studied reservoir are completed with an artificial lift system and permanent down hole gauges, providing a continuous stream of pressure and production data. The additional production information increases the attractiveness of using ensemble methods to condition uncertain geological properties and operations parameters on the continuously updated observed data.
We report the steps to prepare a consistent initial ensemble representing the static and dynamic uncertainties for the CBM field. We also discuss how iterative improvement of the initial ensemble with bias correction factors results in a reduced range of uncertainty on these parameters.
The final history-matched ensemble showed improved gas, water and pressure match for the three years of field production and produced a realistic uncertainty assessment of the forecast. The posterior ensemble is used to derive the representative P90, P50 and P10 models.
A forecast study was conducted to validate the history-matched ensemble. The results showed a good match of 12 months of the new production data, not used in history-matching. That highlights the robust prediction capabilities of the approach.