Enhancing Short-Term Oil Forecasting in SAGD Operations Using ARIMAX-GARCH and Bi-directional LSTM Models

Abstract This study focuses on enhancing short-term oil forecasting in Steam-Assisted Gravity Drainage (SAGD) operations by leveraging advanced ARIMAX-GARCH and Bi-directional LSTM models. SAGD, a vital method for bitumen extraction, involves injecting steam into the reservoir to facilitate oil production through gravity drainage. Conventional analytical models often fail to account for critical operational parameters, such as subcool, and lack the sensitivity required to capture short-term variations and distinct SAGD phases (Ramp-up, Plateau, and Decline). To address these limitations, this work introduces time series models—ARIMAX-GARCH and Bi-directional LSTM—designed to manage temporal dependencies, autocorrelation, and non-stationarity inherent in SAGD production data. Initially, the limitations of traditional machine learning techniques, such as regression and tree-based models, for forecasting challenging cases are highlighted. The performance of ARIMAX-GARCH and Bi-directional LSTM models is then evaluated, showcasing their respective strengths. Bi-directional LSTM is one of the most common time series modeling approaches, but the focus of this paper is on the ARIMAX-GARCH approach. For some test cases, the two methods are compared, demonstrating that for short-term forecasting applications, ARIMAX-GARCH consistently produces better results with significantly shorter training times. The ARIMAX-GARCH model effectively forecasts both the mean and variance of the time series, addressing volatility and uncertainty, while the Bi-directional LSTM excels at capturing complex temporal relationships and uncovering hidden patterns. In this case, ARIMAX-GARCH provided sufficient accuracy for short-term forecasting. In this paper, short-term forecasting refers to a maximum period of 3 years. Finally, the study introduces a novel hybrid methodology that combines ARIMAX-GARCH and Bi-directional LSTM, where ARIMAX-GARCH predicts the linear trends and variance, while Bi-directional LSTM models the residual non-linear patterns. The forecasts are aggregated to improve accuracy and robustness, enhancing the prediction of production rates during different SAGD operational phases.

Récupéré en direct depuis OpenAlex et désinversé. Les résumés ne sont pas conservés dans cette base de données : les index inversés représentent 8,6 Go des 9,3 Go de texte de la base, et le serveur dispose de 13 Go libres.