Artificial intelligence in heliostat control and optimization for CSP plants: A critical review

This review evaluates how artificial intelligence (AI) enhances heliostat control and optimization in concentrated solar power (CSP) systems, a critical need as global emissions reached 41.6 billion metric tons in 2024. The paper systematically evaluates AI and optimization techniques across four critical domains including heliostat field layout optimization, tracking accuracy enhancement, aiming and alignment optimization, and heat flux prediction and control. Results demonstrate significant performance improvements, with genetic algorithms increasing optical efficiency by 14.62 % in heliostat layouts. Notably, layout optimization techniques have successfully reduced the required number of heliostats from 864 to 541. Deep learning models reduce computation time from 600 s to 1 s (≈99.8 % reduction) and achieve 99.7 % recognition accuracy for heliostat detection. Neural network-based adaptive control systems reduce tracking errors to within 0.1 mrad, while StyleGAN architectures achieve 90 % flux prediction accuracy. The integration of machine learning with differentiable ray tracing delivers higher irradiance predictions using only 6000 rays versus traditional methods requiring 1,500,000 rays, achieving a lower L1 loss of 0.4 compared to 0.59. These technical advancements translate into substantial economic benefits, including a 3–4 % reduction in the levelized cost of electricity and annual operational savings of up to $30,000 per MW. Furthermore, the AI-optimized system can avoid approximately 57,624 tons of CO2 emissions annually, strengthening their environmental value. The review highlights challenges such as data quality limitations, computational complexity, and practical deployment barriers. Future work should focus on developing robust, scalable control approaches that enhance the economic viability and long-term sustainability of CSP technologies.