A steam injection distribution optimization method for SAGD oil field using LSTM and dynamic programming

Area of Science:

• Petroleum Engineering
• Artificial Intelligence in Energy
• Optimization Techniques

Background:

• Steam Assisted Gravity Drainage (SAGD) is a crucial enhanced oil recovery technique.
• Optimizing steam injection distribution is vital for maximizing oil production in SAGD operations.
• Existing methods for steam injection optimization have limitations in predictive accuracy and efficiency.

Purpose of the Study:

• To present a novel method for steam injection distribution optimization in SAGD oil fields.
• To integrate Long Short-Term Memory (LSTM) neural networks with dynamic programming for enhanced oil production prediction and optimization.
• To develop a practical tool for real-world SAGD project management.

Main Methods:

• Utilizing LSTM neural networks for accurate prediction of oil production based on historical data.
• Applying dynamic programming to optimize steam injection distribution using LSTM predictions.
• Analyzing the convergence stability and computational complexity of the dynamic programming approach.
• Developing a web-based Geographical Information System (GIS) named Petroleum Explorer.

Main Results:

• The proposed LSTM model demonstrated superior prediction accuracy compared to five other existing models.
• The integrated LSTM and dynamic programming method effectively optimizes steam injection distribution.
• The developed Petroleum Explorer system provides a practical application of the proposed method.
• Production improvement is significantly influenced by the parameter settings in the optimization process.

Conclusions:

• The novel LSTM-based dynamic programming approach offers a significant advancement in SAGD steam injection optimization.
• Accurate oil production prediction using LSTM is key to successful optimization.
• The developed system and methodology have practical implications for enhancing oil recovery in SAGD projects.