Abstract
Xiasiwan Oilfield, located in the southwest Yan'an block of the Ordos Basin, has large geological reserves, but the actual productivity of some oil wells is low. To analyze the reasons for poor oil well productivity, this paper examines the QZ block of Xiasiwan Oilfield as a case study. Based on investigation and actual production characteristics, it adopts a multivariate nonlinear regression model, stochastic forest model, and BP neural network model to comprehensively evaluate the impact of various geological development factors on oil production and clarify the main controlling factors of oil well productivity in tight reservoirs. By comparing the evaluation criteria among the three models, the study analyzes the explanatory power and generalization ability of the models and verifies their accuracy. The results show that oil saturation, porosity, permeability, closing stress, and reservoir thickness are the main factors influencing oil production, in order of significance, with oil saturation being the primary controlling factor affecting oil well productivity. The stochastic forest model is the most balanced and robust among the three models, demonstrating strong generalization ability and resistance to overfitting. This study provides a scientific basis for the efficient development of Xiasiwan Oilfield and helps optimize the oilfield development strategy.