Twin support vector regression for characterizing uncertainty in surface reconstruction

Surface reconstruction plays a pivotal role in various fields, including reverse engineering, and oil and gas exploration. However, errors in available data and insufficient surface morphology information often introduce uncertainty into the reconstruction. It is crucial to accurately characterize and visualize the uncertainty in surface reconstruction for risk analysis and planning further data collection. To this end, this paper proposes an uncertainty characterization method based on twin support vector regression. First, various modeling data are effectively integrated and the information contained in the high-confidence sample is efficiently utilized through the uncertainty interval generated by quantiles and upper/lower bound constraints. Second, well-path points are incorporated by imposing inequality constraints on the corresponding prediction points. Finally, in order to reduce computation time, the problem of uncertainty characterization is formulated as two smaller-scale quadratic programming. The results obtained from a real fault dataset and a synthetic dataset validate the effectiveness of the proposed method. When well data are available, the generated uncertainty envelopes are constrained by well data, which can partially mitigate reconstruction uncertainties.