A systematic and critical review of artificial intelligence applications for enhancing well stimulation operations and sustainable reservoir management

Abstract

Well stimulation remains essential for improving reservoir productivity, but conventional workflows are constrained by geological heterogeneity, limited data, and dependence on empirical correlations. This systematic and comparative review examines artificial-intelligence applications in hydraulic fracturing, acidizing, sand control, and ultrasonic stimulation, focusing on 2020–2025 while selectively considering earlier foundational studies. The quantitative synthesis of 23 key application studies identified 11 studies on acidizing and acid fracturing (47.8%), 8 on hydraulic fracturing (34.8%), 3 on sand control (13.0%), and 1 on ultrasonic stimulation (4.3%). Neural-network architectures were reported in 16 studies (69.6%), while tree-based learners or ensembles appeared in 13 studies (56.5%); these categories overlap because several studies compared multiple algorithms. Among studies reporting numerical performance, R2 values for acidizing-related regression tasks ranged from 0.82 to 0.96, while reported classification accuracies for selected acidizing and sand-control applications ranged from 87.5% to above 90%. These values were not statistically pooled because the studies used different prediction targets, data scales, validation procedures, and performance definitions. Tree-based ensembles and neural networks currently dominate fracture-design and acidizing applications, whereas reinforcement learning, hybrid optimization, and physics-informed approaches remain less frequently investigated. Field-scale and cross-field validation are limited, particularly for sand control and ultrasonic stimulation, for which the available evidence is predominantly laboratory- or simulation-based. The principal barriers to operational deployment are proprietary and small datasets, inconsistent validation, limited interpretability, and uncertain model transferability. Future research should prioritize benchmark datasets, standardized reporting, external field validation, explainable and physics-constrained models, and carefully evaluated closed-loop stimulation systems. Provides an integrated cross-method review of AI applications in hydraulic fracturing, acidizing, sand control, and ultrasonic stimulation, with emphasis on model selection, field readiness, and closed-loop decision support. Proposes a taxonomy of AI models (supervised, unsupervised, deep learning, hybrid/physics-informed) with comparative analysis of strengths, limitations, and use cases. Introduces updated literature tables and figures that map AI adoption trends, stimulation-specific applications, and research gaps. Identifies key challenges including data scarcity, model interpretability, limited field deployment, and underrepresentation of sand control and ultrasonic stimulation. Outlines future directions and strategic recommendations for industry adoption, emphasizing explainable AI, benchmark datasets, real-time control, and ESG-aligned workflows. Provides an integrated cross-method review of AI applications in hydraulic fracturing, acidizing, sand control, and ultrasonic stimulation, with emphasis on model selection, field readiness, and closed-loop decision support. Proposes a taxonomy of AI models (supervised, unsupervised, deep learning, hybrid/physics-informed) with comparative analysis of strengths, limitations, and use cases. Introduces updated literature tables and figures that map AI adoption trends, stimulation-specific applications, and research gaps. Identifies key challenges including data scarcity, model interpretability, limited field deployment, and underrepresentation of sand control and ultrasonic stimulation. Outlines future directions and strategic recommendations for industry adoption, emphasizing explainable AI, benchmark datasets, real-time control, and ESG-aligned workflows.