AI-Enhanced Leak Size Estimation for Offshore Pipelines: Early Detection and Environmental Protection

Abstract Pipeline leakage remains a significant threat to operational safety, environmental sustainability, and economic performance across oil and gas infrastructure, spanning from upstream reservoirs to downstream processing facilities. Traditional detection techniques, such as pressure-based monitoring, acoustic sensing, and manual inspections, often underperform in dynamic multiphase flow regimes due to signal variability, transient effects, and limited sensitivity to minor anomalies. This paper presents a novel artificial intelligence (AI)-based framework that not only detects leaks but also estimates their severity in real time. This represents a key advancement beyond conventional binary classification systems. Due to the scarcity of labeled field data with known leak sizes, high-fidelity datasets were synthetically generated using OLGA multiphase flow simulations across a wide range of operational and leakage scenarios. The resulting time-series data were enriched with dynamic fluid properties derived from PVT analysis and subjected to a structured feature engineering workflow. This process was guided by domain expertise and supported by interpretability techniques, including SHAP values, Pearson correlation analysis, and model-specific importance metrics. Several regression models were evaluated, and XGBoost was selected as the most accurate, achieving an R² of 0.93 across diverse flow regimes and leak magnitudes. The proposed framework is designed for seamless integration with real-time data streams, such as those provided by Multiphase Flow Meters (MPFMs), enabling continuous monitoring without the need for invasive sensors or hardware-intensive systems. By converting oil rate deviations into quantitative leak size estimates, the approach offers a scalable and non-intrusive solution with wide applicability across pipeline networks. The results demonstrate strong predictive performance, combined with high model interpretability and operational feasibility, supporting the development of next-generation smart leak management systems in the energy sector.