全氧化物窄带热光伏发射器的代理辅助逆向设计与温度相关电热分析
Surrogate-Assisted Inverse Design and Temperature-Dependent Electrothermal Analysis of an All-Oxide Narrowband Thermophotovoltaic Emitter
- Bangladesh University of Engineering and Technology（孟加拉国工程与技术大学）
- Chittagong University of Engineering and Technology（吉大港工程与技术大学）
机构由 AI 辅助整理，请以论文原文为准。
AI总结：
本文提出一种基于代理模型（ExtraTrees）和贝叶斯优化的逆向设计框架，用于优化全氧化物窄带热光伏发射器，实现高且窄带的峰值发射，并通过FDTD验证。
AI中文摘要：
与底层太阳能电池带隙对齐的窄带发射器对于提高热光伏（TPV）系统的光谱效率和热稳定性至关重要。基于氧化物材料的发射器为解决传统发射器的光学和机械性能退化提供了一种有前景的方案，传统发射器采用金属-介电结构，在高温下会经历金属氧化和结构劣化。在此，我们提出了一种代理辅助逆向设计框架，用于设计一种由蓝宝石衬底上的ITO和Al2O3层组成的窄带一维光栅发射器。我们在包含峰值发射约束（E_peak > 0.90）的惩罚增强目标上，对训练好的ExtraTrees代理模型进行了贝叶斯优化，同时最小化半高全宽（FWHM）并最大化选定峰值中心光谱带内发射的占比，以获得高且窄带的峰值发射。代理模型使用通过时域有限差分（FDTD）方法获得的发射光谱数据集进行训练，该数据集通过系统改变层厚度和结构周期作为输入特征而生成，并以窄带发射的品质因数（FOM）（包括E_peak、峰值发射波长λ_peak、FWHM、带内占比f_in（用于确定峰值带外发射量）以及峰值附近发射集中度）作为预测目标。优化结构的预测结果进一步通过FDTD方法进行了验证。
英文摘要：
A narrowband emitter aligning with the bandgap of the underlying solar cell is essential for improving the spectral efficiency and thermal stability of thermophotovoltaic (TPV) systems. Emitters based on oxide materials present a promising solution to the optical and mechanical performance degradation of traditional emitters, which employ metal-dielectric structures that experience metal oxidation and structural deterioration at high temperatures. Here, we presented a surrogate-assisted inverse-design framework for a narrowband 1D grating emitter comprising ITO and Al2O3 layers on a sapphire substrate. We performed Bayesian optimization over the trained ExtraTrees surrogates on a penalty-augmented objective containing a peak-emission constraint (E_peak > 0.90) while minimizing the full width at half maximum (FWHM) and maximizing the fraction of emission concentrated within the selected peak-centered spectral band to acquire a high, narrowband peak emission. The surrogate model was trained using a dataset of emission spectra obtained from the finite-difference time-domain (FDTD) by systematically varying the layers' thicknesses and the structure's period as the input features, and the narrowband emission's figure of merit (FOM) ((E_peak), wavelength of peak emission (_peak), FWHM, in-band fraction (f_in) for determining the amount of emission outside the peak band, and concentration of peak emission near the peak) was used as the prediction target. The resulting set of predictions for the optimized structure was further validated using the FDTD method.
