Abstract: X gas reservoir is a massive fractured edge-water gas reservoir controlled by anticline structures, where rapid water breakthrough induced by edge water invasion significantly compromises recovery efficiency, thereby necessitating scientific theories and methodologies to guide development practices. This study systematically investigates the influence of key parameters on recovery efficiency and establishes a prediction model for recovery efficiency in fractured water-drive gas reservoirs, aiming to provide a scientific basis for dynamic optimization of development strategies. Leveraging basic geological and production datasets of the reservoir, a single-well mechanistic model was constructed using the Embedded Discrete Fracture Model (EDFM). Through single-factor sensitivity analysis, the study elucidates the governing mechanisms of water body multiple, gas production rate, matrix permeability, and permeability anisotropy: recovery efficiency exhibits a negative correlation with water body multiple, characterized by accelerated water-gas ratio growth and shortened stable production duration as the multiple increases; gas production rate exerts a pronounced impact on stable production time, with an optimal operational threshold identified to maximize recovery efficiency; matrix permeability demonstrates a positive correlation with recovery efficiency, where lower permeability correlates with faster water-gas ratio escalation and shorter stable production time; excessively low permeability anisotropy ratios degrade recovery efficiency due to inadequate seepage capacity, while higher ratios exacerbate water invasion and further diminish recovery efficiency. 125 sets of cross-experimental schemes were designed. Basic data were obtained through numerical simulation to establish a recovery rate prediction model for fractured water-driven gas reservoirs. Raw data were discretized to enhance predictive accuracy, and a decision tree algorithm was employed to construct the model, which achieved a prediction accuracy of 96% following parameter optimization. Validation against Blasingame production decline analysis using field dynamic data from two production wells in the X gas reservoir demonstrates a high degree of consistency between model predictions and actual production data, underscoring the model's reliability and practical utility. This research provides an efficient and precise technical methodology for predicting recovery efficiency in fractured water-drive gas reservoirs, offering valuable insights for similar reservoir systems.

Key words: machine learning, embedded discrete fracture model, recovery rate prediction, fractured water-drive gas reservoirs, numerical simulation