Adaptive Choke Prediction Model and Optimized Production Rate Framework for Maximum Efficient Rate and Choke Size

Abstract

Efficient choke management is essential for optimizing oil production, particularly in mature wells characterized by increasing water cut and complex multiphase flow behavior. In this work, dynamic Choke Selection Model and optimized production rate framework were developed using machine learning. Historical well test data from treated wells were utilized, covering choke sizes, flow rates, pressures, gas–oil ratio, and water cut. To address data limitations at high water cut conditions, data augmentation was performed using Generative Adversarial Networks, extending the dataset to water cut values up to 0.9. Data pre-processing and model training were implemented in a Python-based environment using numerical and machine learning libraries. The framework integrates an Adaptive Choke Prediction Model, Model Predictive Control, and reinforcement learning to dynamically determine optimal choke sizes under operational constraints. The model incorporates machine learning-based predictions for pressure drop, Maximum efficient rate (MER), and water cut, forming a unified optimization system. Results show that the developed model achieved high predictive performance for choke size and R² exceeding 0.85. The system demonstrated an improvement in MER of approximately 8–12% while reducing water cut by 3–7%. Optimal choke sizes were consistently identified within operational limits, with improved stability and over 95% compliance with constraints. In conclusion, the proposed dynamic choke selection model successfully integrates machine learning and control strategies to enhance production performance. It provides an adaptive and reliable approach for maximizing MER while minimizing water cut, thereby improving operational efficiency and enabling intelligent real-time decision-making in complex oil well systems.