Publications

Kujung Formation in the East Java Basin has a lithology of carbonate rocks with three reservoirs which are dominated by secondary porosity. Kujung 1 reservoir mainly produces gas while Kujung 2 and Kujung 3 reservoirs produce a mixture of oil and water. Identification of fluid type in this formation has been carried out using several methods but has resulted in uncertainties. Resistivity logging from this formation faces challenging interpretation because of the presence of a mixture of oil and the fresh water. The objective of this paper is to highlight machine learning prediction for better fluid interpretation.

The analysis was based on logging, sampling, and historical outcrop data. Operational experiences during drilling are also collected to understand more about wellbore stability and its contribution to error, as the formation is carbonate rock but the underlying bottom layer is shale. The Nuclear Magnetic Resonance (NMR) logging carried out while drilling with the main output of T2 distribution, which allows separation between free fluid and irreducible fluid inside pores, is also used. The Wells B and C pressure data while drilling are used to identify the fluid type through differences in pressure gradients. Finally, historical data similar to Kujung is collected to improve the reservoir fluid interpretation. A special machine learning workflow has been developed with this set of data.

Initially, well results showed less accuracy due to low fluid mobility and the carbonate vuggy formation, therefore formation pressure was still influenced by hydrostatic pressure from drilling mud. After machine learning modelling using algorithms, the existing NMR output was processed to new imaging indicating the fluid as depths went deeper. This study has produced better visualization to predict fluid type, especially in difficult reservoir sections.