<![CDATA[This study addresses the critical need for accurate oil and gas production forecasting to support strategic decision-making in Indonesia’s energy sector. PT Pertamina Hulu Rokan Zone 4 (PHR Zona 4), a key player in national energy production, frequently encounters technical and external operational challenges. To tackle these issues, this research proposes a deep learning-based predictive model using the Long Short-Term Memory (LSTM) architecture, structured in an encoder-decoder format and enhanced with an attention mechanism. The model was trained and tested on historical oil and gas production data from PHR Zona 4, evaluated under two data-splitting scenarios: 80:20 and 90:10. Model performance was assessed using Mean Absolute Percentage Error (MAPE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), and the coefficient of determination (R²). Results from the 80:20 scenario showed RMSE of 5.83, MAE of 5.54, MAPE of 1.71%, and R² of -1.97, suggesting difficulties in capturing extreme data fluctuations. However, the 90:10 scenario demonstrated significantly improved performance with RMSE of 0.42, MAE of 0.36, MAPE of 0.11%, and R² of 0.00, indicating better trend prediction stability. The novelty of this study lies in the integration of attention mechanisms within the LSTM encoder-decoder framework for oil and gas time series forecasting, offering enhanced accuracy and robustness. This research provides a valuable foundation for future improvements in predictive analytics and operational efficiency in the oil and gas industry.]]>
Copyrights © 2025
