A hybrid machine learning approach for predicting the performance of perovskite solar cells under varying temperatures

Renewable energy sources, particularly solar cells, play a crucial role in energy production, with silicon-based cells being the most common. However, perovskite solar cells have emerged as a promising alternative due to their diverse structural configurations and lower cost compared to traditional silicon cells. This study develops a unified model that integrates both classification and regression approaches to predict the optimal absorber material and assess the impact of temperature on solar cell performance. In the classification task, gradient boosting and random forest models demonstrated a higher area under the curve compared to other models. Before perovskite solar cells can be commercialized, experimental research must be conducted to better understand the factors influencing their performance. However, experiments are time-consuming and costly, and testing under varied conditions has its limitations. To overcome these challenges, machine learning is applied to improve and expand experimental data. With its high accuracy and speed, machine learning is widely used across various fields, including the development of perovskite solar cells. In this research, the impact of temperature on perovskite solar cells is examined. The goal was to gather experimental data and predict missing information. Among the three regression techniques applied, random forest regression yielded the highest accuracy at 98 %, while linear regression had the lowest at 75 %. Using the random forest approach, the power conversion efficiency at predicted temperatures of 55 °C, 75 °C, and 85 °C was found to be 99 %, 89 %, and 88 % of the initial value, respectively.