Real-Time Reservoir Modelling and Optimization Frameworks for Enhanced Hydrocarbon Recovery in the Niger Delta Basin

Abstract The Niger Delta Basin's complex reservoirs demand adaptive strategies to overcome dynamic subsurface challenges and maximize hydrocarbon recovery. Traditional reservoir modelling techniques are hindered by computational inefficiencies and static assumptions, leading to 15–25% BHP overestimates in water-cut wells. This study introduces a novel hybrid framework integrating edge computing, symbolic regression, and physics-guided machine learning (ML) to enable real-time reservoir management. The framework derives interpretable BHP equations (e.g., BHP = 1.24THP + 0.055GasCD - 4.0ln(WaterCut + 1) + 0.0021Sand + ϵ) using IoT sensor data, achieving 93% accuracy (R²) and 45 psia RMSE—a 67% improvement over Cullender-Smith and 47% improvement over Hagedorn-Brown. Edge deployment on Raspberry Pi devices enables <15 ms inference speeds, reducing computational latency by 18% compared to desktop-based methods. Dynamic decline curves, updated with real-time THP and WaterCut, extended well life by 6–10 months in Reservoir X in the Niger Delta. Field implementations will demonstrate 12–14% gas flaring reduction through ML-optimized gas lift scheduling and 25% faster waterflood efficiency predictions using physics-informed neural networks (PINNs). By harmonizing edge-processed ML insights with reduced-order simulations, the framework is capable of reducing operational costs by an estimate of 40% and aligns with Nigeria's Petroleum Industry Act (2021) flaring mandates. This work sets a benchmark for agile, sustainable reservoir management in heterogeneous basins, prioritizing scalability and regulatory compliance.