Abolfazl Dehghan Monfarad

Production forecasting of oil wells plays a crucial role in field development planning, economic decision-making, and efficient management of hydrocarbon reservoirs. Conventional approaches, such as decline curve analysis based on the Arps model, despite their simplicity and widespread use, have limitations in modeling complex reservoir behavior and production fluctuations. In this study, a combination of analytical and data-driven methods was employed to improve prediction accuracy. Initially, decline curve analysis of wells was performed using ProMate production engineering software based on the Arps model. Subsequently, a data-driven model based on a Long Short-Term Memory (LSTM) neural network was developed to reconstruct actual production trends with higher accuracy. The input data consisted of historical daily production records, which were preprocessed, normalized, and divided into training, validation, and testing datasets for model development. Comparison of the two models showed that the LSTM model achieved higher accuracy in predicting production rates. The obtained statistical metrics — Mean Squared Error (MSE) = 4×10⁴, Mean Absolute Error (MAE) = 6×10², and Root Mean Squared Error (RMSE) = 7×10¹ — demonstrate the superior performance of the data-driven model compared to the Arps model. Overall, the results indicate that integrating classical analytical methods with deep learning algorithms can effectively reduce forecasting errors and provide a reliable tool for analyzing and modeling oil well production behavior.