Abstract: For the difficulty in accurately and swiftly obtaining bottomhole pressure in gas storage reservoirs,in this study,Xiangguosi Gas Storage was taken as the research object to analyze 12 characteristic parameters from 14 datasets across 10 wells based on the data-driven principles and supervised learning methods.Three bottomhole pressure prediction models:Gaussian Process Regression (GPR),Support Vector Regression (SVR),and Artificial Neural Network (ANN),were developed and evaluated for prediction accuracy.The study shows that the main factors influencing bottomhole pressure are daily injection-production volume,wellhead pressure,formation pressure,and wellhead temperature,and the prediction accuracy of SVR,GPR,and ANN models are 99.2%,97.4%,and 95.1%,respectively.This indicates that data-driven methods can effectively predict bottomhole pressure,with the SVR model offering a more reliable prediction for gas storage injection-production control.This research is of great practical significance for enhancing the safety and economy of gas storage operations.

Key words: underground gas storage, bottomhole pressure, machine learning, data-driven, production monitoring data, prediction model