A Machine Learning Approach for Metal Oxide Based Polymer Composites as Charge Selective Layers in Perovskite Solar Cells. Issue 5 (18th May 2021)

• Record Type: Journal Article
• Authors: Yildirim, Murat Onur; Gok, Elif Ceren; Hemasiri, Naveen Harindu; Eren, Esin; Kazim, Samrana; Oksuz, Aysegul Uygun; Ahmad, Shahzada
• Is Part Of: ChemPlusChem. Volume 86:Issue 5(2021)
• Page Start: 785
• Page End: 793
• Publication Date: 2021-05-18
• Subjects: conjugated polymers -- machine learning -- perovskite solar cells -- tungsten trioxide; Chemistry -- Periodicals; 540.5
• DOI: 10.1002/cplu.202100132
• Languages: English
• ISSNs: 2192-6506

Abstract

Abstract: A library of metal oxide‐conjugated polymer composites was prepared, encompassing WO3 ‐polyaniline (PANI), WO3 ‐poly(N‐methylaniline) (PMANI), WO3 ‐poly(2‐fluoroaniline) (PFANI), WO3 ‐polythiophene (PTh), WO3 ‐polyfuran (PFu) and WO3 ‐poly(3, 4‐ethylenedioxythiophene) (PEDOT) which were used as hole selective layers for perovskite solar cells (PSCs) fabrication. We adopted machine learning approaches to predict and compare PSCs performances with the developed WO3 and its composites. For the evaluation of PSCs performance, a decision tree model that returns 0.9656 R 2 score is ideal for the WO3 ‐PEDOT composite, while a random forest model was found to be suitable for WO3 ‐PMANI, WO3 ‐PFANI, and WO3 ‐PFu with R 2 scores of 0.9976, 0.9968, and 0.9772 respectively. In the case of WO3, WO3 ‐PANI, and WO3 ‐PTh, a K‐Nearest Neighbors model was found suitable with R 2 scores of 0.9975, 0.9916, and 0.9969 respectively. Machine learning can be a pioneering prediction model for the PSCs performance and its validation. Abstract : Predicting the performance of perovskite solar cells with input datasets using a machine learning approach and its experimental validation are reported in this work. Machine learning approaches used to predict and compare solar cell performance with the developed WO3 and its composites. For performance evaluation, a decision tree model is ideal for the WO3 ‐PEDOT composite, while a random forest model was also found to be suitable for some.