With the increasing energy consumption of buildings, transparent heat mirror films have been widely used in building windows to enhance energy efficiency owing to their excellent spectrally selective properties. Previous studies have typically focused on spectral selectivity in the visible and near-infrared bands, as well as single-parameter optimization of film materials or thickness, without fully exploring the performance potential of the films. To address the limitations of traditional design methods, this paper proposes a deep reinforcement learning-based approach that employs an adaptive strategy network to optimize the thin-film material system and layer thickness parameters simultaneously. Through inverse design, a Ta2O5/Ag/Ta2O5/Ag/Ta2O5 (42 nm/22 nm/79 nm/22 nm/40 nm) thin-film structure with broadband spectral selectivity was obtained. The film exhibited an average reflectance of 75.5% in the ultraviolet band and 93.2% in the near-infrared band while maintaining an average visible transmittance of 87.0% and a mid- to far-infrared emissivity as low as 1.7%. Additionally, the film maintained excellent optical performance over a wide range of incident angles, making it suitable for use in complex lighting environments. Building energy simulations indicate that the film achieves a maximum energy-saving rate of 17.93% under the hot climatic conditions of Changsha and 16.81% in Guangzhou, demonstrating that the designed transparent heat mirror film provides a viable approach to reducing building energy consumption and holds significant potential for practical applications.
Keywords: building energy efficiency; deep reinforcement learning; inverse design; spectrally selective properties; transparent heat mirror.