Predicting photovoltaic parameters of perovskite solar cells using machine learning.

Abstract

Perovskite solar cells (PSCs) have garnered significant attention owing to their highly power conversion efficiency (PCE) and cost-effectiveness. Traditionally, screening for PSCs with superior photovoltaic parameters relies on resource-intensive trial-and-error experiments. Nowadays, time-saving machine learning (ML) techniques serve as an artificial intelligence approach to expedite the prediction of photovoltaic parameters using accumulated research datasets. In this study, we employ seven supervised ML methods to forecast key photovoltaic parameters for PSCs such as PCE, Short-Circuit Current Density (Jsc), Open-Circuit Voltage (Voc), and Fill Factor (FF). Particularly, we design an artificial neural network (ANN) architecture that incorporates residual connectivity and layer normalization after the linear layers to enhance the scope and adaptability of the network. For PCE and Jsc, ANN demonstrates superior prediction accuracy, yielding root mean square errors of 2.632% and 2.244 mA/cm2, respectively. The Random Forest (RF) model exhibits exceptional prediction performance for Voc and FF. Additionally, an interpretability analysis of the model is conducted to elucidate the impact of features on PCE prediction, offering a novel approach for accurate and interpretable ML methods in the context of PSCs.