AI-Based Deflector Control for Vertical Axis Wind Turbines: A Systematic Literature Review

Vertical axis wind turbines (VAWTs) have become an attractive option for use in cities and other built environments because they can capture wind from different directions. This gives them an advantage over horizontal axis wind turbines in locations where wind direction changes frequently. However, VAWTs still face several challenges, including unstable airflow, dynamic stall, flow separation, and negative torque during certain parts of their rotation cycle. These issues can reduce overall performance and efficiency. This systematic literature review focuses on three closely related areas: VAWT aerodynamic performance; the use of flow deflectors to improve airflow; and the application of artificial intelligence for prediction, optimization, and control. The review followed a Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA)-based method and collected studies published between 2008 and 2026 from Scopus, Web of Science, ScienceDirect, and Google Scholar. Following PRISMA guidelines, 1127 records were identified, 792 remained after duplicate removal, 136 full texts were assessed, and 21 studies were included in the final review. The selected studies were organized according to turbine type, deflector design, operating conditions, research methods, and the role of artificial intelligence. An assessment was also carried out to compare evidence from experiments, validated simulations, optimization studies, and emerging AI applications. The findings show that well designed deflectors can improve the aerodynamic performance of VAWTs when compared with their original configurations. However, the level of improvement depends on factors such as turbine design, wind speed, Reynolds number, tip speed ratio, and deflector shape. Because of these differences, reported performance gains should be considered specific to each study rather than a general result. Artificial intelligence has mainly been used for performance prediction, optimization, surrogate modelling, and turbine control. Reinforcement learning appears promising for adaptive deflector control. However, very few studies have tested a complete system that combines sensors, an adjustable deflector, artificial intelligence, and real-time closed-loop control. Based on the reviewed studies, a five-layer research framework is proposed. The framework includes aerodynamic and mechanical design, data collection and processing, AI model development, real-time control and actuation, and experimental validation. This framework is presented as a research direction for future investigation. Future studies should compare reinforcement learning with traditional control methods, evaluate energy consumption, address the gap between simulation and real-world operation, and conduct more experimental testing.