AI-Driven Production Forecasting: Integrating LSTM, Prophet, and Random Forest

Abstract This study improves production forecasting for a heterogeneous Niger Delta reservoir using an ensemble of Long Short-Term Memory (LSTM), Prophet, and Random Forest (RF) models. Thirty-two years (1992–2024) of Gabo Field, production data were analyzed, and a five-year forecast was generated through a workflow combining Decline Curve Analysis with advanced machine learning. Ensemble stacking with XGBoost delivered the most reliable performance (MASE < 1), demonstrating consistent improvement over standalone models. Random Forest showed high predictive strength (R² = 0.98–0.99), while LSTM and Prophet captured temporal and seasonal patterns that enhanced ensemble robustness. Integration with Deepseek-R1 cognitive analysis aided identification of reservoir heterogeneities and supported improved decision-making. Results highlight the value of hybrid physics-informed and AI-driven workflows for production forecasting and reservoir management. The study demonstrates how combining traditional reservoir engineering with machine learning and cognitive tools can enhance forecast reliability and optimize development planning in mature oil fields.