Fair train-test split in machine learning: Mitigating spatial autocorrelation for improved prediction accuracy

Article Geosciences, Multidisciplinary

A Spatial Correlation-Based Anomaly Detection Method for Subsurface Modeling

Wendi Liu et al.

Summary: The proposed spatial anomaly detection method integrates the semivariogram spatial continuity model and joint cumulative probability to identify anomalous data, and can assist in determining anomalous samples, regions, or facies transition boundaries by incorporating geological interpretations and domain expertise knowledge.

MATHEMATICAL GEOSCIENCES (2021)

添加到收藏夹

Article Energy & Fuels

Demonstration and Mitigation of Spatial Sampling Bias for Machine-Learning Predictions

Wendi Liu et al.

Summary: Spatial geological data sets are characterized by sparsity and bias, posing challenges for machine-learning predictive models; a proposed spatial-weighted tree method is effective in reducing prediction bias and improving accuracy for reservoir evaluation.

SPE RESERVOIR EVALUATION & ENGINEERING (2021)

添加到收藏夹

Article Geosciences, Multidisciplinary

Stochastic Pix2pix: A New Machine Learning Method for Geophysical and Well Conditioning of Rule-Based Channel Reservoir Models

Wen Pan et al.

Summary: Accurately reproducing geological heterogeneity in subsurface models is critical but can be time-consuming and computationally intensive. Overfitting of data and slow convergence due to high dimensionality are common problems in modeling. To address these issues, a new machine learning approach is introduced to parameterize stochastic reservoir models into low-dimensional Gaussian random variables.

NATURAL RESOURCES RESEARCH (2021)

添加到收藏夹

Article Energy & Fuels

Tuning machine learning dropout for subsurface uncertainty model accuracy

Eduardo Maldonado-Cruz et al.

Summary: This study proposes a method for improving the tuning of Dropout frequency in deep learning models by integrating model uncertainty and assessing the goodness of the entire uncertainty model based on prediction realizations. Testing on a subsurface flow surrogate model shows that using the proposed method results in accurate and precise uncertainty models. The robustness and generalization of the method are also tested with out-of-distribution testing data.

JOURNAL OF PETROLEUM SCIENCE AND ENGINEERING (2021)

添加到收藏夹

Article Energy & Fuels

Geostatistical significance of differences for spatial subsurface phenomenon

Jose J. Salazar et al.

Summary: Optimum subsurface development decision-making depends on uncertainty models that integrate all information, data, and interpretations. Failure to account for the spatial context of subsurface data may lead to naive and overconfident estimations of uncertainty. New methods are needed to assess the significance of local differences in spatial datasets and to improve decision-making processes.

JOURNAL OF PETROLEUM SCIENCE AND ENGINEERING (2021)

添加到收藏夹

Article Computer Science, Interdisciplinary Applications