Oil Well Productivity Prediction by Machine Learning: A Case Study of an Iraqi Oil Field

Abstract

Accurate prediction of oil well productivity is essential for reservoir management, production optimization, and field development planning. This study evaluates the applicability of machine learning techniques for forecasting oil production and compares their performance with conventional Decline Curve Analysis (DCA) using production data from an Iraqi oil field. Historical production data from three representative wells covering the period 1976–2020 were collected and preprocessed through filtering and outlier removal. A Temporal Convolutional Network (TCN) model was developed using lookback windows of 3, 4, and 5 time steps to capture temporal dependencies in production behavior. Model performance was evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the coefficient of determination (R²), and the results were compared with those obtained from conventional DCA forecasting. The TCN model demonstrated strong predictive capability, achieving R² values ranging from 0.885 to 0.956 and RMSE values between 96 and 204 across the studied wells. In contrast, the DCA model produced an R² value of 0.55 and an RMSE of 665, indicating significantly lower predictive accuracy. The results confirm that the TCN model outperforms conventional forecasting methods by effectively capturing nonlinear production trends and long-term temporal dependencies. This study demonstrates the potential of deep learning approaches, particularly Temporal Convolutional Networks, to enhance production forecasting accuracy and support decision-making in complex reservoir environments using long-term production data.