Machine Learning Methods for Handling Parameter Space Sampling Bias in Unconventional Well Performance Prediction

Technical Abstract
Written By GATE Energy

Oliver Rojas Conde | Henry Galvis Silva | Sebastien Matringe | Viking Engineering | Texas A&M University | Hess Corporation

Abstract

Prediction accuracy for extended-reach laterals and high-intensity completions is improved by applying a Gaussian Process Regression (GPR) model with a Matérn kernel that accounts for parameter space sampling bias. The method adjusts predictions based on local data density and provides uncertainty quantification, increasing reliability in underrepresented regions of the design space. Applied to wells from the Bakken Formation, the model outperforms traditional approaches and supports confident forecasting for non-standard development designs.

  • Document ID: URTEC-4249754-MS

  • Publisher: OnePetro

  • Source: SPE/AAPG/SEG Unconventional Resources Technology Conference

  • Publication Date: June 2025

GATE Energy
Next

The CM-CX Approach: Integrated Construction Management & Commissioning