Optimization of Profile Control and Oil Displacement Scheme Parameters Based on Deep Deterministic Policy Gradient

Area of Science:

• Petroleum Engineering
• Artificial Intelligence in Reservoir Management
• Enhanced Oil Recovery (EOR)

Background:

• Effective parameter design for profile control and oil displacement (PCOD) is crucial for enhancing waterflooding efficiency and maximizing oil field production and recovery.
• Traditional optimization methods may not fully capture the complex dynamics of PCOD processes.
• The need for advanced computational methods to optimize PCOD parameters for improved oil recovery is evident.

Purpose of the Study:

• To develop and validate a parameter optimization model for PCOD schemes using deep deterministic policy gradient (DDPG).
• To maximize the half-year increased oil production (Q) of injection well groups.
• To optimize PCOD system type, concentration, injection volume, and injection rate under defined constraints.

Main Methods:

• Constructed a parameter optimization model and solution method based on DDPG.
• Utilized historical PCOD data and extreme gradient boosting (XGBoost) to create a proxy model of the PCOD process as the environment.
• Defined the reward function based on the change rate of Q, with system type, concentration, injection volume, and rate as actions, and employed a Gaussian strategy with noise for exploration.

Main Results:

• The DDPG-based model successfully optimized parameters for a compound slug PCOD process (pre-slug + main slug + protection slug) in an offshore oil field example.
• Achieved higher oil production PCOD schemes for well groups with varying PCOD characteristics.
• Demonstrated superior optimization and generalization capabilities compared to the particle swarm optimization (PSO) model.

Conclusions:

• The DDPG-based parameter optimization model is effective for improving PCOD scheme performance.
• The proposed method offers significant advantages in optimizing oil field production and recovery.
• The DDPG approach shows strong potential for practical application in complex reservoir management scenarios.