<![CDATA[Oil measured volume discrepancies in custody transfer systems is becoming a persistent challenge, which is often caused by complex thermal, hydraulic, and compositional interactions. Therefore, this study aimed to introduce a data-driven framework incorporating Principal Component Analysis (PCA) and machine learning (ML) to identify as well as predict discrepancies at a representative onshore gathering station (GS) in Indonesia (Field-X). Major operational parameters, including gross volume, unallocated net oil, pressure, temperature, and Basic Sediment & Water (BS&W), were analyzed to assess the impact on volumetric imbalance. During the analysis, PCA reduced 64 correlated variables to five principal components, explaining 95% of the total variance and showing gross volume, pressure, and temperature as dominant factors. Four ML models, namely XGBoost, Random Forest, Support Vector Regression, and ElasticNet, were trained as well as validated with three-fold time series cross-validation for temporal robustness. Incorporating PCA significantly improved predictive performance, with Support Vector Regression showing the largest R² increase (from –0.0082 to 0.82). Results signified that discrepancies were primarily governed by thermodynamic shrinkage, temperature changes, and BS&W-related metering errors. In addition, the proposed PCA–ML framework offered an interpretable, reliable method for early detection and mitigation of oil volume discrepancies in complex production environments.]]>
