Randomization in Characterizing the Subsurface
暂无分享,去创建一个
Current methods for characterizing Earth’s subsurface, such as standard inverse techniques, are not sufficiently accurate to meet the needs of modern applications in the fields of energy exploration, environmental management, and global security. While increasing the quantity of field measurements and robustness of the applied data-/model-analysis methods can improve accuracy, such approaches can be computationally impractical for large data sets and complex site conditions. Therefore, there is a need to develop economically-feasible and robust computational methods while maintaining accuracy. For example, in-field drilling for geothermal operations may yield high failure rates, resulting in unacceptably high costs; errors and/or large uncertainties in the estimated subsurface characteristics are the main impediment to the successful siting of an in-field well. This problem is not uniquely geothermal. Accurate characterization of uncertain subsurface properties is also critical for monitoring storage of carbon dioxide, estimating pathways of subsurface contaminant transport, and supervising ground-based nuclear-explosion tests.
[1] Shusen Wang,et al. Towards real-time geologic feature detection from seismic measurements using a randomized machine-learning algorithm , 2017 .
[2] Velimir V. Vesselinov,et al. Large‐scale inverse model analyses employing fast randomized data reduction , 2017 .
[3] Shusen Wang,et al. Efficient Data-Driven Geologic Feature Detection from Pre-stack Seismic Measurements using Randomized Machine-Learning Algorithm , 2017, Geophysical Journal International.