稀土发光材料在高温下表现出荧光强度降低的现象，即热猝灭，这严重影响了发光效率。热诱导荧光增强是抵消荧光淬灭的最有效方法，然而，热增强的内在机制尚不清楚，只能通过广泛的试错实验来探索，慢而且缺乏规律性。为了解决这些问题，我们提出了一种基于机器学习的创新方法来预测Er³⁺掺杂发光材料的热诱导增强效应，目的是建立荧光强度增强与激发波长、温度、掺杂浓度、绝对灵敏度和其它特征条件之间的关系。通过构建特征数据集训练了八个机器学习模型。研究结果表明，集成树模型（如XGBoost和梯度增强）在预测绿光和红光增强强度方面表现最佳，其R²系数均超过0.84，平均绝对误差（MAE）低于0.35。此外，我们应用训练好的模型预测了Er³⁺掺杂氟化物（NaYF₄、NaGdF₄等）和钼酸钨（NaY(WO₄)₂、NaLa(MoO₄)₂等）两种新材料体系的热增强性能，预测误差低至3.58%，验证了该模型具有出色的泛化能力和较高的预测精度。本研究不仅为理解荧光热增强的复杂机制提供了数据驱动的新视角，而且为高温环境应用中发光材料的合理设计和性能优化提供了一种高效可靠的新方法。该研究成果被美国光学学会Optics Express, 2025（IF=3.3）接受出版！

全文链接：https://doi.org/10.1364/OE.574940

Fig. 1 The mechanism of heat-induced fluorescence enhancement effect, (a) The distance between atoms decreases with the increase of temperature, d represents this distance; (b) The surface molecules of nanoparticles evaporate as temperature increases, producing specific surface phonons. These phonons can not only assist in the energy transfer between sensitized ions and activated ions, but also promote the realization of the multi-photon upconversion process by assisting the radiation-free relaxation between the energy levels of specific excited states.

Fig. 2 Flow chart and step-by-step analysis of machine learning to predict the thermally induced fluorescence enhancement effect of rare earth-doped upconversion luminescent materials.

Fig. 3 Prediction and Actual Comparisons of Green, red Gain and thermal sensitivity.

Fig. 4 Algorithm error of fluoride and tungsten-molybdate systems.