Methods for Quantifying Data Uncertainty for Intelligent Decision Support in Computer-Integrated Oil and Gas Production Systems

Abstract

Modern computer-integrated systems for oil and gas production rely on intelligent decision support systems for monitoring, forecasting, diagnostics, and control of technological processes. The effectiveness of such systems significantly depends on the reliability and completeness of input data, which are often affected by measurement errors, missing values, noise, model inaccuracies, and variability of geological and production conditions. Under such circumstances, uncertainty quantification becomes a critical component for ensuring the reliability of predictive models and supporting robust decision-making in digital oilfield environments. The purpose of this paper is to analyze and systematize modern methods for uncertainty quantification of data used in intelligent decision support systems of computer-integrated systems for oil and gas production, as well as to determine their applicability under conditions of incomplete and uncertain information. The study considers the specific features of technological data in oil and gas production, including heterogeneity of information sources, nonlinear process dynamics, measurement uncertainty, and real-time operational constraints. The paper formalizes the uncertainty quantification problem in intelligent decision support systems and analyzes the main types of uncertainty affecting technological systems, including aleatoric, epistemic, measurement, model, and simulation parameter uncertainty. A comparative analysis of Bayesian approaches, ensemble methods, Monte Carlo sampling techniques, probabilistic neural networks, and calibration-based methods is performed from the standpoint of computational efficiency, robustness to noisy and incomplete data, interpretability, and compatibility with industrial monitoring and control systems. The obtained results demonstrate that no single uncertainty quantification method fully satisfies all requirements of computer-integrated oil and gas production systems. It is shown that ensemble methods provide the most balanced solution for operational decision support tasks, while Bayesian and Monte Carlo approaches are more suitable for reservoir modeling and long-term forecasting. Calibration techniques are identified as necessary for ensuring prediction reliability in real-time applications. A decision framework for selecting uncertainty quantification methods according to operational constraints and data characteristics is proposed. The practical significance of the study lies in the possibility of applying the obtained recommendations in the development of computer-integrated systems for monitoring, diagnostics, forecasting, and intelligent decision support for technological objects operating under uncertain and incomplete data conditions.