Perpustakaan Digital ITB

To identify unproduced or bypassed oil in conventionally mature water-flooded oilfields, Engineer must rely on history-matched dynamic reservoir simulation. This method not only incurs high computational costs, but also requires full re-calibration every time new data becomes available. This study poses a set of entirely new questions: How much of the displacement state can be reconstructed using only universal data namely well completion records, static rock properties, and measurements from special core analysis (SCAL) laboratory tests? What is the quantified value of the simulation information that cannot be captured by these universal datasets? This study develops a hybrid "rule score × machine learning" framework, which is validated using the Volve field dataset publicly released by Equinor (formerly Statoil). The deterministic rule score integrates two physically interpretable features: a proxy for injector travel time, and a fault-aware Dijkstra distance to production well perforations. Both features are mapped to hierarchical percentile rankings, and all parameters are set according to displacement physics and recovery mechanisms, with no values tuned to match the evaluation ground truth. The machine learning layer then predicts changes in these continuous scores over the course of well completion events. Additionally, reservoir units are labeled as un-swept based on two criteria: a unit mobility threshold inverted from the SOF3 relative permeability table, and a water-phase relative permeability krow > 0.001. All evaluations are conducted across 60,270 reservoir column units. Under the verification target set for August 2016, the prediction framework proposed in this paper achieves a median AUC of 0.7277 for each layer, a rank-corrected combined AUC of 0.6873, and an improvement of 1.397 for the undrained category. From 2011 to 2016, its AUC remained stable within the range of 0.714 to 0.730. An exhaustive search of all features only yielded an AUC of 0.7311, while the persistence baseline AUC for 2014 was 0.9227. The difference of 0.1950 verifies the diagnostic value of dynamic states. The assessment map developed in this study only adjusts its scores when perforations are opened, and it can adapt to the reservoir conditions of Norway's Volve oilfield in 2014, 2015, and 2016. The matching rates from comparative verification are 0.7183, 0.7256, and 0.7277 in sequence, which show a steady increase over time. This map describes the final coverage range of the fracturing network. Because it uses only absolute permeability to characterize displacement rate, it is only applicable to scenarios with a narrow range of fluid mobility. This condition is satisfied under the light oil conditions of the Volve oilfield; for heavy oil oilfields, relevant prerequisites must be met. Ultimately, this study produces a set of implementable, auditable screening tools and a description of their applicable boundaries..