Publications

The acquisition of Gas While Drilling (GWD) data for the needs of geological surveillance and safety operation during drilling is a common practice in Mutiara Field, Kutai Basin. However, it still rarely used for reservoir fluid identification because the presumptions of wide uncertainties coming from mud characteristics used in the drilling and tempered by the complex geological reservoir compartmentalization. The previous in-house conventional GWD analysis conducted by Battu et al. (2023) had 60% success ratio in correctly identifying the reservoir fluid. This condition attracts attention of the current study to leverage the success ratio of fluid identification using GWD data. Therefore, an innovative machine learning-based workflow which comprised of three different methodologies, such as: regression, neural network, and random forest were performed.

The results of implementing the workflow in 36 wells (32 wells used for building a model and 4 wells used for real time blind testing) drilled in 2014-2023 showed an increasing fluid identification accuracy with success ratio up to 80%. The results have been validated with wireline fluid analysis testing and perforation results hence it added confidence to the workflow utilization. However, the workflow is open for further development. Particularly for the effect of dual-fluid, pressure depletion, and water-rise that still made the model less accurate thus also impacted on the success ratio.