A machine learning-based method for recovery rate prediction in fractured water-driven gas reservoirs
Year: 2025
Type: article
Abstract: Gas reservoir X is a block-fractured edge-water gas reservoir controlled by anticline structures. Due to edge-water invasion, rapid water breakthrough severely reduced recovery efficiency, highlighting the urgent need for theoretical and methodological guidance for reservoir development. To address this, based on a systematic analysis of the influence patterns of key parameters on recovery rate, a recovery prediction model for fractured water-driven gas reservoirs was established. This model provides a scientific basis for the dynamic optimization of development schemes. Using basic geological and production data, a single-well mechanistic model was constructed with the embedded discrete fracture model (EDFM). Through single-factor sensitivity analysis, the influence patterns of water body multiple, gas production rate, matrix permeability, and permeability anisotropy were revealed. Recovery rate exhibited a negative correlation with water body multiple. As the water body multiple increased, the water-gas ratio of gas wells rose significantly faster, and the stable production time was greatly shortened. Gas production rate had a pronounced impact on the stable production time of the reservoir. An optimal gas production rate threshold existed that maximized the recovery rate. Matrix permeability was positively correlated with recovery rate. Lower matrix permeability led to lower recovery rate. The faster the increase in water-gas ratio, the shorter the stable production time. When permeability anisotropy was too low, poor seepage capacity resulted in a reduced recovery rate. An increased ratio accelerated water invasion, further decreasing the recovery rate. Based on these findings, 125 sets of cross-experimental schemes were designed, and basic data were obtained through numerical simulations. A recovery rate prediction model for fractured water-driven gas reservoirs was established. To improve the prediction accuracy of the model, the original data were discretized, and a decision tree algorithm was used. After parameter optimization, the prediction accuracy of the model reached 96%. The model was validated using actual dynamic data from two production wells in the gas reservoir X. The prediction results were compared with those from the Blasingame production decline analysis method. The results showed high consistency between model predictions and actual production data, indicating high reliability and practicality of the model. This provides an efficient and precise technical method for recovery rate prediction in fractured water-driven gas reservoirs.
Author: SUN Qiufen, QIN Jiazheng, FENG Qiao, QIAO Yu, LIU Yaxin, ZHAO Qiyang, XU Liang, YAN Chun
Language: English
Subfield: Ocean Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7090773566
FWCI:
Cites:
Cited by:
Related to: