Machine Learning Prediction and Efficient Screening Method for Thermally Induced Fluorescence Enhancement of Er³+ Doped Materials

Rare earth luminescent materials exhibit a phenomenon of fluorescence intensity reduction at high temperature, that is, thermal quenching, which seriously affects the luminescent efficiency. Thermally induced fluorescence enhancement is the most effective method to counteract fluorescence quenching, however, the intrinsic mechanism of thermal enhancement remains unclear and can only be probed through extensive trial-and-error experiments. In order to solve these problems, an innovative method based on machine learning is proposed to predict the thermally induced enhancement effect of Er³+ upconversion luminescence. The aim is to establish the relationship between the enhancement intensity and the excitation wavelength, temperature, doping concentration, absolute sensitivity and other characteristic conditions. Eight machine learning models were trained by constructing feature data sets. The research results show that integrated tree models (such as XGBoost and gradient boosting) perform optimally in predicting the intensity of green and red light enhancement, with their R² coefficients all exceeding 0.84 and their mean absolute errors (MAE) below 0.35. Furthermore, we applied the trained models to predict the thermal enhancement properties of two new material systems, fluorides (NaYF₄, NaGdF₄, etc.) and tungsten molybdates (NaY(WO₄)₂, NaLa(MoO₄)₂, etc.), achieving prediction errors as low as 3.58%, which verifies the model's excellent generalization ability and high prediction accuracy. This study not only provides a data-driven new perspective for understanding the complex mechanism of fluorescence thermal enhancement but also offers an efficient and reliable new approach for the rational design and performance optimization of upconversion luminescent materials for high-temperature applications.