Reservoir Fluid Typing From Standard Mud Gas – A Machine-Learning Approach

Standard mud gas data is part of the basic mud-logging service, and it is mainly used for safety monitoring. Despite its availability for all wells, it is not used for reservoir fluid typing due to poor prediction accuracy. We recently developed a new manual method that significantly improved the reservoir fluid typing accuracy from standard mud gas data. However, there is a strong business for an automatic method to enable reservoir fluid interpretation while drilling. A machine-learning method has been developed based on a well-established standard mud gas database. The standard mud gas compositions contain methane, ethane, and propane components with reasonable quality measurements. Due to the low compositions of butane and pentane in standard mud gas (sometimes approaching the detection limit), we only use methane to propane compositions in the method. Such usage impedes an accurate prediction of reservoir fluid type. Therefore, we introduce additional data sources: a large in-house reservoir fluid database and petrophysical logs. The machine-learning algorithm extracts critical reservoir fluid information for a known field by utilizing the geospatial location and the existing reservoir fluid database. Their combination with the standard mud gas database enhanced the reservoir fluid typing accuracy from 50 to 60 to nearly 80%. Petrophysical logs are also essential to reservoir fluid type identification. Their combination with the machine-learning model (already with satisfactory performance) contributed to the prediction accuracy of about 80%. Given the difficulties of distinguishing oil or gas for near-critical fluids or volatile oil, the current prediction accuracy is sufficient for industry applications. Standard mud gas has been regarded as inapplicable data for accurate reservoir fluid typing for many decades. Nevertheless, the proposed innovation created significant business opportunities and unlocked the possibility of accurate reservoir fluid typing for real-time well decisions like well placement, completion, and sidetracking. It can also add significant value for well integrity, maturating production targets, and cost-efficient plug and abandonment (P&A) in the overburden.