ML-Based Bayesian Optimization of Indoor Perovskite Solar Cells for IoT Integration

Mitashi Jain, Lakshya

Abstract: The rapid growth of Internet of Things (IoT) devices has created a need for maintenance-free indoor power sources that can replace primary batteries. Indoor photovoltaics (IPVs) based on metal-halide perovskites are well suited to this role because their bandgap can be tuned to the narrow emission spectra of LED and fluorescent lamps. This paper reports an integrated machine-learning (ML) and experimental workflow for indoor perovskite solar cells (IPSCs). A curated dataset drawn from the Perovskite Database Project, published literature and in-house laboratory records was used to train a gradient-boosted forward model that predicts optical bandgap from A-, B- and X-site composition, achieving R² = 0.90-0.92 with a mean absolute error below 0.05 eV. The trained model was then embedded in an Optuna-driven Bayesian inverse search of 10,000 trials to locate compositions with a bandgap near the indoor-optimal value of 1.8 eV. The recommended composition MA_0.37FA_0.31Cs_0.31PbI_1.33Br_0.44Cl_1.23 (Eg ? 1.812 eV) was cross-validated against independently retrained models. Triple-cation n-i-p devices fabricated on FTO/TiO₂ substrates delivered power conversion efficiencies of 14.08-24.7 % under AM 1.5 G illumination and successfully drove an LED load, demonstrating feasibility for self-powered IoT nodes.

Keywords: Indoor photovoltaics, perovskite solar cells, Bayesian optimization, machine learning, IoT energy harvesting