Quantitative identification of dominant controlling factors and productivity prediction for deep–ultra-deep tight sandstone reservoirs

Deep to ultra-deep tight sandstone reservoirs are characterized by low porosity, low permeability, and strong heterogeneity, and their productivity is jointly controlled by geological, engineering, and development factors. However, the high-dimensional nonlinearity of these variables and their complex interactions make it difficult for conventional methods to accurately identify dominant controls and achieve reliable productivity prediction. To address this issue, this study developed an integrated workflow combining data preprocessing, multidimensional correlation-based factor screening, single-factor trend analysis based on simple linear regression, and machine-learning-based prediction. Pearson, Spearman, and Kendall correlation coefficients were jointly employed to identify the dominant controlling factors, and multiple predictive algorithms were systematically compared to select the optimal model for productivity prediction at the single-well level and the block-level sample-set level. The results indicate that stable water cut, shut-in time, fracturing-fluid volume, and oil saturation are the core factors controlling productivity. Among the tested models, XGBoost exhibited the best predictive performance, with overall predictive accuracy exceeding 95%. At the block-level sample-set level, errors in initial productivity prediction were mainly associated with excessively high stable water cut, whereas deviations in first-year cumulative production prediction were closely related to high water cut and insufficient shut-in time. These findings demonstrate that the proposed framework provides an effective tool for productivity evaluation and prediction within the present dataset and development conditions.