Integrating inflow control valve control with LSTM networks for oil production forecasting in horizontal intelligent well application

Abstract

Abstract Inflow Control Valves (ICVs) are pivotal components in intelligent horizontal well systems, enabling precise fluid flow regulation and enhancing reservoir management. Despite their advantages, ICV-equipped wells pose a significant challenge for accurate oil production forecasting. Traditional single-well production assessment methods, such as productivity index (PI) estimations and reservoir simulations, are fundamentally limited in evaluating the impact of ICV setting. This study addresses this challenge by developing an Enhanced Long Short-Term Memory (LSTM) model for intelligent horizontal wells. Unlike traditional methods, this model dynamically integrates real-time operational data, such as pump pressure and ICV configurations, to capture the intricate, time-dependent dynamics of oil production. The Enhanced LSTM was evaluated using a 5-fold cross-validation with data from 10 wells, achieving a Mean Absolute Error (MAE) of 4.333 and an R² value of 0.958. This performance surpasses that of both the Back Propagation (BP) model (MAE: 7.325, R²: 0.789) and the basic LSTM (MAE: 5.749, R²: 0.718). Validation on two independent wells further confirmed its robustness, with MAE values of 4.873 and 1.508 and R² values of 0.908 and 0.897, respectively. These results demonstrate the Enhanced LSTM’s superior adaptability and performance in real-time production forecasting, making it an effective tool for optimizing well operations in complex environments.

Keywords

• Oil production forecasting
• Inflow control valves
• Long short-term memory networks
• Machine learning in oil and gas