Optimization of CO2 flooding in tight oil based on improved machine learning

This study develops an innovative methodology for CO 2 flooding optimization in tight oil reservoirs through a novel hybrid analytical framework that effectively combines Grey Relational Analysis (GRA), Entropy Weight Method (EWM), and Analytic Hierarchy Process (AHP) in a logically structured decision-making process. The research achieves significant methodological advancement by intelligently integrating type-2 fuzzy logic with artificial neural networks to establish robust high-precision predictive models, while developing a comprehensive optimization system that innovatively correlates geological-fluid characteristics with engineering parameters through systematic parameter coupling. The proposed framework introduces a new classification-based approach to formulate specific optimization criteria for different reservoir grades. Detailed comparative studies confirm that the staggered well-fracture network configuration provides optimal performance, offering clear technical advantages over alternative patterns in terms of both productivity enhancement and recovery efficiency. The framework's distinctive analytical capability enables a thorough evaluation of key controlling factors, with rigorous sensitivity analysis identifying porosity, oil saturation, effective reservoir thickness, crude oil viscosity, and crude oil density as dominant geological controls in Block W, while precisely quantifying the impact of operational parameters including fracturing clusters, horizontal interval length, and total gas injection volume. The optimization process employs an enhanced Particle Swarm-Genetic Hybrid System, where 151 iterations of intelligent algorithm refinement yield field-validated performance improvements. The resulting scheme demonstrates consistent effectiveness across reservoir grades, delivering 10%–25% cumulative production increase and over 8% recovery factor enhancement. This multidisciplinary approach provides a technically sound and practically viable solution for tight oil reservoir management, establishing a new methodological standard for CO 2 flooding optimization that successfully overcomes several key limitations of traditional analytical approaches through its integrated, data-driven methodology. • Machine learning-based multi-parameter coupling analysis boosts CO 2 flooding optimization in tight reservoirs. • Production and efficiency prediction by type-2 fuzzy neural network system. • Formation-specific CO 2 flooding strategies based on reservoir classification-driven optimization.