Optimization of monocrystalline silicon solar cell using Box–Behnken design and machine learning models

Abstract

This work integrates PC1D simulation, Box–Behnken design (BBD), and machine learning models (artificial neural network—ANN and particle swarm optimization-artificial neural network—PSO-ANN) to optimize monocrystalline silicon solar cells. Using the global desirability function, the optimal efficiency of 23.29% is obtained under certain conditions: p-type doping concentration (3.32 × 10¹⁷ cm⁻³), n-type doping concentration (6 × 10¹⁷ cm⁻³), textured wafer pyramid height (1 µm), textured wafer pyramid angle (80.67°), and temperature (20 °C). Notably, the PSO-ANN model outperforms the ANN model with an RMSE of 0.0149 and a correlation coefficient of 0.9997. This study demonstrates the effectiveness of advanced modeling and machine learning in increasing solar cell efficiency and highlights the superior performance of the PSO-ANN model.