Abstract
The coalbed methane reservoir in the southern margin of the Junggar Basin exhibits significant heterogeneity, posing considerable challenges for the prediction of gas well productivity. Machine learning, which has robust data analysis capabilities, can effectively determine the relationships among reservoir parameters, fracturing parameters, and gas well productivity. In this study, based on fracturing operation data and production dynamics from coalbed methane wells in the Fukang mining area in the southern margin of the Junggar Basin, gray correlation analysis was employed to quantitatively evaluate multiple influencing factors and to identify the key controlling factors governing gas well productivity. An improved model that integrates a back-propagation (BP) neural network with a genetic algorithm was developed to predict coalbed methane well production, and this model incorporates the main controlling factors. The production predictions generated by the new GA-BP neural network model exhibit a high degree of agreement with actual production data. Compared to the traditional BP neural network, the improved model was found to have significant advantages in terms of predictive accuracy and stability, and it can more precisely capture the dynamic changes in gas well productivity. This method provides reliable technical support for optimizing fracturing technology and conducting comprehensive productivity evaluations of the coalbed methane reservoir in the southern margin of the Junggar Basin as well as facilitating the efficient development and utilization of regional coalbed methane resources.