Predicting the Thermodynamic Stability of Perovskite Structures through Machine Learning
Abstract
Perovskites solar cells have been improved tremendously over the years, making them a material of interest in the solar photovoltaic (PV) world. However, the perovskite material is limited by the thermodynamic instability of its structure. Creating and testing of structural and molecular combinations in a laboratory is expensive and time consuming as there is a plethora of these combinations. The thermodynamic instability of perovskite materials can be improved through various chemical processes; however, this research focuses on component choice alteration to tackle this. This component choice alteration is facilitated by machine learning, where energy above convex hull of the analyzed perovskite structures is used as basis for the prediction of stable materials. On the other hand, machine learning, using presently available data, can be used to predict the molecular system in which the structural frame is more likely thermodynamically stable. In this research project, seven supervised classification machine learning models have been developed to predict the stability of perovskite structures with different molecular systems. Two datasets (sample size 1,929 for Dataset 1 and 5,329 for Dataset 2) which consist of perovskite structural compounds are evaluated for this project. The models, which include logistic regression, K nearest neighbors (KNN), decision tree, random forest, support vector machine (SVM), multi-layer perceptron, and gradient boosting are used to predict the thermodynamic stability of the perovskite structures, and the results of these different models are compared. Random Forest model performed best for the prediction of thermodynamic stability of within these datasets with accuracy of 96.6% for Dataset 1 and 94.9% for Dataset 2.
Subject Area
Mechanical engineering|Artificial intelligence
Recommended Citation
Moyinoluwa Ibironke Eniola,
"Predicting the Thermodynamic Stability of Perovskite Structures through Machine Learning"
(2021).
ETD Collection for Tennessee State University.
Paper AAI28767898.
https://digitalscholarship.tnstate.edu/dissertations/AAI28767898