Optimizing Spectral Properties of Cesium Tungsten Bronze Films Doped with Silver Nanowires Based on the Machine Learning Method

Abstract

Glass exhibits high transmittance in the solar radiation band but high absorbance in the mid-far infrared (MIR) band, which causes a poor energy-saving effect. Thermal insulation coatings offer the most effective solution to address this. Among these, nano-cesium tungsten bronze (Cs_xWO₃) had a strong blocking effect in the solar radiation band due to its intrinsic absorption, local surface plasmon resonance, and small polaron absorption, but its reflectivity in the MIR band was very low. Conversely, silver nanowires (AgNWs) formed a dense network structure with low transmittance and high reflectance in the MIR band; however, when mixed into Cs_0.32WO₃ slurries, the solar radiation-blocking ability was weakened. In order to assess the impact of AgNWs on the properties of Cs_0.32WO₃ films, this study collected experimental data from different Cs_0.32WO₃ films doped with AgNWs for multilayer perceptron (MLP) neural network machine learning. The trained models exhibited efficient and accurate prediction abilities. A large number of extrapolated independent variables were input to the trained MLP models using the grid search method, and then the predicted results of dependent variables were displayed in three-dimensional (3D) models to more intuitively show the influence of doping AgNWs on the optical performance of different Cs_0.32WO₃ films. Through an optimization analysis of two 3D models of T_550 nm (transmittance at 550 nm) and SC (shading coefficient), two transition values of T_550 nm were found: 69.5 and 64.2%. When T_550 nm surpassed 69.5%, the SC value of nondoped Cs_0.32WO₃ films was the lowest. Conversely, when T_550 nm was below 69.5%, doping with AgNWs decreased the SC value of Cs_0.32WO₃ films. Using the optimal mixture of Cs_0.32WO₃ slurries and AgNW slurry at a ratio of 1:3, the SC value of nondoped Cs_0.32WO₃ films was lower when T_550 nm exceeded 64.2%. Conversely, when T_550 nmwas below 64.2%, the same mixture resulted in a lower SC value. These predicted results and their accuracy were verified by experiments and provided important technical guidance and method support for future research and the application of high-performance Cs_0.32WO₃ films.