A novel method for calculating bottom hole pressure in underground gas storage wells promises to greatly improve operational efficiency and precision, marking a significant advancement for the industry. Researchers have introduced the Theory and Data-Driven Neural Network (TDDNN) approach, which integrates theoretical wellbore flow equations with artificial intelligence techniques. This innovative methodology aims to address the challenges faced by traditional calculation techniques used during various stages of gas storage operations, such as injection, withdrawal, and shut-in.
The accurate assessment of bottom hole pressure is fundamental for the effective management of underground gas storage (UGS) facilities. Historically, two primary methods for measuring bottom hole pressure were applied: one relying on downhole pressure gauges and fiber optic technologies, and the other utilizing conventional reservoir engineering methods. While the first method provides real-time data, it is costly and requires high reliability under extreme conditions. The latter method, more cost-effective, suffers from inaccuracies and complex calculations.
By leveraging the advantages of machine learning, the new TDDNN method aims to provide both speed and accuracy. It constructs its models by first analyzing the key variables affecting bottom hole pressure calculations. These include factors such as gas injection rates, wellhead pressure, and temperature variations throughout the wellbore. This new approach promises rapid calculations, reportedly cutting the average processing time for predictions from several seconds down to milliseconds.
The researchers demonstrate the effectiveness of their method by comparing it to traditional techniques across five precision metrics: Mean Absolute Error (MAE), Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and Coefficient of Determination (R²). The results indicate significant improvements, showcasing the method's ability to maintain high accuracy with limited sample data.
Previous studies have indicated limitations of existing models when addressing the fluctuative nature of bottom hole pressure during gas injection and production phases, particularly within the dynamic environment of UGS. This novel approach stands out because it integrates both theoretical principles and empirical samples during training, ensuring outputs closely align with actual measurements taken during gas storage operations.
Eventually, this integration allows for greater flexibility and adaptability. A substantial portion of the underlying model relies on historical production data gathered from real-world scenarios across 12 wells within the Zhongyuan Oilfield, amassing over 12,130 data records collected across various operational phases. These existing records serve as invaluable benchmarks against which the TDDNN method's predictive capabilities can be validated.
Notably, the research emphasizes the wellbore structure, focusing on the mid-depth perforation as the primary point of interest for bottom hole pressure estimations. This specificity aids developments not only within vertical wells but also opens avenues for adapting methodologies to the more complex structures of horizontal wells.
Moving forward, there is potential to refine this method and expand its applications beyond traditional paradigms, enhancing the future of gas reservoir management and smart energy solutions. With the necessity of enhanced computational models becoming increasingly apparent, particularly in energy sectors facing stringent environmental regulations, the TDDNN may represent a new frontier for integrating artificial intelligence with foundational engineering practices.