International Journal of
eISSN: 2475-5559
Submit manuscript...
Research Article Volume 8 Issue 1
Application of gated recurrent units for time series prediction of gas production rates
Vugar M. Fataliyev, Kanan F. Aliyev, Sonakhanim A. Salimova
ASOIU, Oil and gas engineering department, Doctor of Technical Sciences, Azerbaijan
ASOIU, Oil and gas engineering, PhD candidate, Azerbaijan
OGI (Oil and Gas institute), Hydrogasdynamics of Reservoir Systems, Azerbaijan
Received: December 17, 2025 | Published: January 2, 2026
Citation: Fataliyev VM, Aliyev KF, Salimova SA. Application of gated recurrent units for time series prediction of gas production rates. Int J Petrochem Sci Eng. 2026;8(1):1-6. DOI: 10.15406/ipcse.2026.08.00144
Abstract
Accurate forecasting of gas production rates is essential for effective reservoir management and early detection of flow assurance issues such as liquid loading. This study explores the application of a Gated Recurrent Unit (GRU)-based deep learning model to predict daily gas production rates in a gas condensate well using a univariate time series approach. To improve prediction accuracy and minimize the impact on the results, a dataset spanning 1,500 days was utilized. The dataset was divided into training and testing subsets, comprising 1,200 days and 300 days, respectively. Gas rate data were normalized prior to model training. The GRU model was trained using a sliding window of 10-day sequences and evaluated based on standard regression metrics. The model achieved a Mean Squared Error (MSE) of 78.58, a Mean Absolute Error (MAE) of 3.35, and a Mean Absolute Percentage Error (MAPE) of 3.98%, indicating strong predictive accuracy and generalization capability. In addition to accurately tracking stable production behavior, the model successfully captured sudden changes in gas rate trends with a short one-day delay. This characteristic is particularly valuable for the early identification of production anomalies, including the onset of liquid loading, a condition that typically develops over a period of several days. The findings highlight the potential of GRU-based models not only for high-resolution production forecasting but also as intelligent monitoring tools for real-time anomaly detection. The study concludes by recommending the integration of additional operational parameters such as wellhead pressure and temperature to further enhance predictive performance and extend applicability in reservoir surveillance.
Keywords: gas condensate wells, liquid loading, production rate prediction, gated recurrent unit (gru), time series forecasting
©2026 Fataliyev, et al. This is an open access article distributed under the terms of the, which permits unrestricted use, distribution, and build upon your work non-commercially.