Oil-source correlation analysis using neural networks: a case study from Bozhong Sag, Bohai Bay Basin, China

The Bozhong Sag, a hydrocarbon-rich depression in the Bohai Bay Basin, contains multiple sets of source rocks with similar environment of deposition, posing challenges for oil-source correlation. This study optimizes a neural network approach to address the nonlinear relationships between biomarker parameters and hydrocarbon origins. Using 28 biomarker parameters from 43 source rock samples, we built conventional discriminant analysis and neural network models. The neural network achieved 95% accuracy in distinguishing three source rock groups, outperforming the discriminant method (79%). Due to the higher accuracy the neural network method was extended to investigate oil-source correlation for 17 crude oil samples from oilfield A. Results reveal that oils above 3600 m mainly originate from Es 1 Member, while deeper oils derive from Es 3 Member, which is consistent with established charging histories (5-6 Ma for deeper reservoirs, ∼2.5 Ma to present for shallower ones). Generally, this study presents a novel oil-source correlation technique that combines traditional biomarker parameters with machine learning, enabling a significant improvement in classification accuracy. This approach is not only pivotal for hydrocarbon exploration and exploitation in the Bohai Bay Basin but also establishes a widely applicable methodology for oil–source correlation in other complex basins worldwide.