Machine learning-assisted highly efficient thermal management in function-oriented thermochromic smart windows

Abstract

Windows contribute significantly to energy inefficiency in electric vehicles and buildings, and the development of smart windows with customized optical responses catering to various purposes presents a significant challenge. Vanadium dioxide (VO2) Fabry-Perot resonator has emerged as an effective device to modulate heat transfer passively to enhance energy-saving performance, with its efficacy highly limited by the complex interrelationship between the structure and multispectral selectivity. This study introduces a physics-guided neural network to design energy-efficient thermochromic smart windows for directional privacy protection (DPP). Guided by machine learning predictions, the VO2 nanoparticle size- and spacer-controlled DPP smart window exhibits a targeted luminous transmission (<0.15) with both high modulation of near-infrared transmittance of 0.12 and longwave infrared radiation emissivity of 0.56, surpassing both commercial and the best-reported windows. When applied in electric vehicles and building envelopes, the DPP smart window demonstrated superior thermal management compared to commercial counterparts. This work provides a framework for the inverse design of function-oriented smart windows, advancing energy-efficient solutions for real-world applications.