Twin support vector regression for characterizing uncertainty in surface reconstruction.

Area of Science:

• Geosciences
• Data Science
• Engineering

Background:

• Surface reconstruction is vital for reverse engineering and oil/gas exploration.
• Data errors and limited surface information create uncertainty in reconstruction.
• Accurate uncertainty visualization is essential for risk assessment and data acquisition planning.

Purpose of the Study:

• To propose an uncertainty characterization method for surface reconstruction using twin support vector regression (TSVR).
• To effectively integrate diverse modeling data and leverage high-confidence samples.
• To incorporate well-path data for improved reconstruction accuracy and uncertainty reduction.

Main Methods:

• Utilized TSVR to generate uncertainty intervals via quantiles and bound constraints.
• Incorporated well-path points using inequality constraints on prediction points.
• Formulated the uncertainty characterization as two smaller-scale quadratic programming problems to reduce computation time.

Main Results:

• Validated the proposed method on real fault and synthetic datasets.
• Demonstrated effective integration of various data sources.
• Showcased how well data constrains uncertainty envelopes, partially mitigating reconstruction uncertainties.

Conclusions:

• The proposed TSVR-based method accurately characterizes and visualizes surface reconstruction uncertainty.
• The integration of well-path data significantly constrains uncertainty envelopes.
• This approach enhances risk analysis and planning in subsurface modeling and exploration.