Generation of chemical library of near-IR dyes for photovoltaics applications

Abstract

The dyes with light absorption ability in extended spectrum have huge potential in various photovoltaics applications. Absorption maxima (Lambda max) is predicted using machine learning (ML). Multiple machine learning models are used. Random Forest is best model among the tried ML models. A library of new dyes is created using python-based tool. Absorption maxima of newly generated dyes is predicted using best ML model (Random Forest). The generated library of dyes is visualized using Uniform Manifold Approximation and Projection (UMAP) plot. 30 dyes with absorption in extended spectrum are selected. Synthetic accessibility assessment is done to check ease of synthesis of selected dyes and to further narrow down the number of potential candidates. Structural behavior of selected dyes is studied using hierarchical cluster analysis. This study offers a theoretical framework for developing potential dyes that could exhibit light absorption in near-IR region of solar spectrum.