Abstract: The Zimi Gas Field is a typical low-permeability tight gas field with great differences in reservoir physical properties and production characteristics among different gas well reservoirs, so the development strategy is in urgent need of improvement. To address this problem, a classification method for tight gas wells based on XGBoost algorithm is proposed. The input features of the model are determined by calculating Pearson correlation coefficients of the feature parameters and judging the correlation degree of 4 dynamic and 5 static feature parameters with gas well productivity. Then, based on the XGBoost algorithm, the model training is completed and gas wells are classified in Zimi Gas Field through parameter optimization. The study shows that the main factors affecting gas well classification are production allocation, original formation pressure, effective thickness, porosity and open flow capacity; the gas well productivity of Category 1 and 2 in Zimi Gas Field is mainly affected by reservoir thickness and matrix permeability, while the main factors affecting the gas well productivity of Category 3 are effective thickness and gas saturation; the accuracy of the model classification results is 92.3% compared with the expert classification results. This study improves the effectiveness of gas well classification, reduces human subjectivity, and the classification results are in line with the actual mine field, which has a certain guidance for gas well classification management and development strategy formulation.

Key words: machine learning, gas well classification, XGBoost algorithm, Zimi Gas Field