Physics-Guided Machine Learning for Transient Reservoir Characterization: Synthetic Dataset, Models, Validation and Benchmark Results
Description
This dataset supports the study “Physics-Guided Machine Learning for Transient Reservoir Characterization: Accuracy, Out-of-Distribution Generalization, and Noise Robustness in Niger Delta Sandstones.” It contains synthetic pressure-transient reservoir data generated across representative sandstone reservoir and fluid-property ranges, together with variables used for physics-guided machine-learning model development, training, validation, testing, out-of-distribution generalization assessment, and noise-robustness evaluation. The dataset includes reservoir and fluid parameters, transient pressure-response features, and corresponding target reservoir properties. It is provided to support reproducibility, independent verification, and further research on machine-learning-assisted transient reservoir characterization.
Files
Steps to reproduce
The archive contains the 10,000-case physics-constrained synthetic transient reservoir dataset, pressure-history tables, predefined train/validation/test/out-of-distribution splits, model artifacts, predictions, performance metrics, forward-model validation outputs, quality-control results, noise-robustness analyses, and independent benchmark results. Use the supplied data splits when comparing model performance to avoid data leakage. The validation and analysis tables can be used to verify the reported permeability and skin prediction metrics, pressure-reconstruction errors, out-of-distribution performance, and noise-sensitivity results. The independent benchmark files document the blind benchmark evaluation and flow-regime-aware screening analysis. The original source scripts from the iterative computational workflow were not preserved as a standalone versioned code release; therefore, the archive supports verification and secondary analysis of the reported data, model artifacts and outputs rather than complete source-code re-execution.