Surrogate-based optimization of a folded solar cell structure with enhanced optical efficiency

One of the most effective approaches to improving the efficiency of a solar cell is to alter the geometric structure of the panel. Convex and folded structures can enhance absorption of solar radiation in crystalline cells by reducing average reflectance from the surface. In order to properly design these modified surface geometries, precise optical radiation simulations such as ray tracing optics must be executed. Monte Carlo ray tracing simulations can be extremely time-consuming, and therefore thorough optimization that relies on them is inefficient. One way to overcome the aforementioned challenge is to utilize surrogate models. Approximate models learned from data can be used for optimization instead of original MC simulations. This work applies surrogate-based optimization to ray tracing optical simulations in order to improve the efficiency of silicon solar panels using a folded surface. We study optimal design of the geometry of a stationary solar cell structure with a three-folded surface. The efficiency of the cell is measured with respect to average radiation angle with standard AM1.5 solar irradiance, and the design goal is to maximize the average optical absorptivity of the radiated sunlight in the panel. We use neural networks as surrogate models and demonstrate that the model can accurately estimate spectral absorptivity of a random geometry, and furthermore, it can be reliability used in optimization. Results of optimization using the proposed surrogate model suggest that the optimal panel with folded surface has an improved efficiency over that of the flat panel with the same length by as large as 13%.