Optimizing the efficiency of perovskite solar cells for improved performance and energy conversion using temporal dynamic graph neural network

Abstract

Perovskite solar cells represent a promising solution for next-generation solar energy due to their high power conversion efficiency (PCE) and cost-effective fabrication. However, enhancing their performance remains a major challenge, largely because existing material selection and optimization methods rely heavily on time-consuming, trial-and-error experimentation. To overcome these limitations, a hybrid framework combining temporal dynamic graph neural networks (TDGNN) and human evolutionary optimization (HEO) referred to as the TDGNN-HEO method is proposed for improving prediction accuracy and optimizing the efficiency of perovskite tandem solar cells. The main goal of this strategy is to precisely forecast cell performance and maximize PCE. TDGNN is used to capture temporal and structural dependencies among solar cell parameters, enabling precise prediction of short-circuit current. HEO is applied to optimize the neural network’s weight parameters, enhancing learning effectiveness and overall model performance. The methodology is implemented in MATLAB and evaluated against established techniques, including convolutional neural networks, random forest algorithm, and K-nearest neighbors. Results demonstrate that the TDGNN-HEO method achieves a PCE of 20.5%, significantly outperforming the benchmark models, which yield 18.7%, 15.5%, and 10.13%, respectively. In terms of prediction accuracy, TDGNN-HEO attains 97%, compared to 85%, 75%, and 65% for the other techniques. These outcomes highlight the effectiveness of the TDGNN-HEO framework in improving both the efficiency and predictive reliability of perovskite solar cells, offering a robust data-driven solution for advancing solar cell design and performance optimization.