Spectrally selective structures for glass cover of silicon photovoltaic module to enhance its energy yield based on neural network surrogate-assisted genetic algorithm

Strategically transmitting harvestable photons and reflecting non-useful photons are essential for photovoltaic (PV) efficiency improvement and temperature reduction. However, it remains a considerable challenge to develop spectrally selective thin-film structures that can maximize PV lifetime energy yield (EY) while maintaining structural simplicity and adequate manufacturability. Herein, a neural network surrogate-assisted genetic algorithm (NNSGA) framework is proposed to facilitate the design of spectrally selective structures. By incorporating spectral utilization weight (SUW) of PV modules, the optimization is guided toward a more effective search direction. Optimal thin-film structures under different design weights are identified using NNSGA with a large population by further leveraging the surrogate model for accelerated prediction. The results reveal that, by adopting SUW, high-performance structures with superior above-bandgap transmittance and below-bandgap reflectance can be designed. Additionally, an extra PDMS top layer degrades the spectral selectivity while introducing an additional heat source, resulting in non-beneficial or deteriorated performance, despite the emissivity compensation. Furthermore, owing to the inherently high transmittance of the air/glass interface, weight of transmission waveband (TWB) must exceed 0.8 to maintain the current, and the operating temperature can only be reduced by about 2.30 °C, with approximately 0.52% efficiency loss under specific working conditions. Moreover, weight of TWB equal to 0.9 can maximize the performance of silicon PV modules. The 25-year cumulative EY can be improved by about 0.94%, mainly due to increased current rather than temperature reduction, where the lifetime EY of silicon PV modules can be enhanced by approximately 3.40% when thermal degradation is accounted for.