Forward Mineral Modeling Using Regularized Least-Squares Regression With Singular Value Decomposition: Case Study From Qusaiba Shale
- Guangping Xu (Schlumberger-Doll Research and Sandia National Laboratories) | David McCormick (Schlumberger-Doll Research) | Michael Herron (Schlumberger-Doll Research) | Stephen Cheshire (EXPEC Advanced Research Center) | Ahmed Al-Salim (EXPEC Advanced Research Center) | Anas Almarzouq (EXPEC Advanced Research Center)
- Document ID
- Society of Petrophysicists and Well-Log Analysts
- Publication Date
- June 2017
- Document Type
- Journal Paper
- 242 - 269
- 2017. Society of Petrophysicists & Well Log Analysts
- 1 in the last 30 days
- 158 since 2007
- Show more detail
Mineralogy is important in the evaluation of reservoir quality and completion quality, and thus mineral modeling is of great interest to the industry. The inversion of major-element data from either XRF or nuclear spectroscopy tools to obtain mineralogy logs is a common approach to the interpretation of these data. The inversion method requires one to take into account all major minerals in the samples, which is often challenging to achieve with high accuracy and precision. An alternative approach, forward mineral modeling, uses a limited calibration dataset to derive mineral composition from elemental abundance.
This paper uses a forward-mineral-modeling method to correlate mineral content with all eight major elements (Si, Al, Ca, Mg, Na, K, Fe, and S), in which each individual mineral or a group of minerals is solved independently. The coefficients between mineral and element are obtained through the local calibration. This local calibration method can solve for most of the minerals sought even in situations where some minerals cannot be separately measured with confidence, such as illite and smectite from XRD measurements.
We present a calibrated algorithm using least-squares regression that is optimized by regularization and singular value decomposition applied to a set of samples from the Lower Silurian Qusaiba Member of the Qalibah Formation of Saudi Arabia. The optimized algorithm estimates mineral composition using elemental concentrations from either core or log measurements. With as few as 10 representative samples in the calibration dataset, the optimized algorithm has the ability to predict minerals with an accuracy of a few percent.
Mineralogy affects many petrophysical parameters including porosity, permeability, water saturation, and attributes related to rock strength, which are crucial for evaluating reservoir and completion quality of potential reservoir rocks. The mineralogy of unconventional hydrocarbon shale reservoirs is more complex, but has been less extensively characterized than conventional sandstone and carbonate reservoirs. For unconventional reservoirs, the common industry practice is to measure mineralogy on a set of selected samples to calibrate petrophysical data. The mineralogy can also be estimated from geochemical logging data.
|File Size||33 MB||Number of Pages||28|