关键词:
Australia
Surat Basin
Coal fracture porosity
Numerical modelling
Machine learning
摘要:
Porosity is one of the key reservoir properties controlling coal seam water production. An accurate interpretation of coal porosity is essential for the forecasting of coal seam gas and water production. However, unlike other conventional or unconventional reservoirs, coal fracture porosity is difficult to interpret from petrophysical logs or well tests because of the complexity and heterogeneity of coal fractures and cleats. This report presents a study to estimate the coal fracture porosity and coal seam water resource in the eastern Surat Basin, Australia, using big data analytics, dynamic modelling and a deep learning (DL) model. Given their association, gas and water production were analysed against porosity across the eastern Surat Basin. Dynamic Model No.1 was first used to simulate water production based on known porosity and Voronoi areas, establishing the relationship between reservoir properties and forecasted water output. Because estimates from Voronoi-based drainage areas alone were biased, a drainage area calibration factor was introduced. A DL model was trained to predict this drainage area calibration factor using dynamic modelling results from 378 proposed wells. The trained DL model was then applied to 1637 existing wells with up to 13 years of observed water production, enabling calculation of calibrated porosity at each well. Results show that calibration is essential for accurate coal fracture porosity estimation, with recoverable water, well online date, and on production timing variance among neighbouring wells being key influences. Dynamic Model No.2, using the locally calibrated porosities, provides the closest match to historical water production.