Oil production forecasting in a heterogeneous Niger Delta reservoir using ensemble machine learning and cognitive computing

Abstract

This study improves oil 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 analysed to generate a five-year forecast using a workflow that combines decline curve analysis with ensemble machine learning and DeepSeek-R1 cognitive analysis. Ensemble stacking with XGBoost delivered the most reliable performance across most wells, as indicated by MASE (MASE < 1) and RMSE, demonstrating consistent improvement over standalone models. The Random Forest and stacked ensemble models demonstrated strong predictive performance in the transformed forecasting space, while original-scale validation metrics indicated larger absolute forecasting errors due to the high variability and magnitude of field production rates. Integration with DeepSeek-R1 cognitive analysis aided the identification of reservoir heterogeneities and supported improved decision-making. This study demonstrates how combining traditional reservoir engineering with machine learning and cognitive tools in a single workflow can enhance forecast reliability and optimise development planning in mature oil fields. Ensemble machine learning improves oil production forecasting in a Niger Delta oil Field. LSTM, Prophet, and Random Forest capture complementary production trends. XGBoost Stacked ensemble modelling achieved consistently superior MASE performance. The workflow integrates DCA with machine learning and DeepSeek-R1 cognitive analysis. The framework supports data-driven reservoir management decisions.