Аннотация
Bubble-point pressure and solution gas-oil ratio are critical PVT properties that govern phase behavior, reservoir performance, and the accuracy of material balance, production forecasting, and numerical simulation studies. Existing empirical correlations have historically been calibrated using geographically limited datasets, resulting in significant prediction errors when applied to crude oils with different compositional and geological characteristics, such as those found in Kazakhstan. This study investigates the performance of widely used global correlations in an extensive set of Kazakhstan oilfield data and demonstrates their systematic limitations. To address these shortcomings, new regional correlations were developed for bubble-point pressure and solution gas-oil ratio using a machine-learning-assisted workflow. Symbolic regression was employed to derive an explicit analytical expression for the solution gas-oil ratio, while multivariate log-linear regression was used to formulate a physically interpretable correlation for bubble-point pressure. Both models were evaluated using industry-standard error metrics and compared with established global correlations. The newly developed regional correlations exhibit substantially lower prediction errors and markedly improved consistency across the full range of fluid properties. These findings highlight the importance of region-specific PVT model development and provide more reliable tools for reservoir engineering applications in settings where laboratory data are limited or incomplete.