Machine Learning-Assisted Accelerated Research of Energy Storage Properties of BaTiO3–BiMeO3 Ceramics

The exploration of dielectric materials with excellent energy storage properties has always been a research focus in the field of materials science. The development of a technical method that can accurately predict the energy storage characteristics of ceramics will significantly accelerate the pace of research into energy storage materials. In this research, a machine learning method was utilized with the aim of accurately predicting the energy storage density (Wrec) and energy storage efficiency (η) of BaTiO3–BiMeO3 (BT-BMO) ferroelectric ceramics. Initially, a data set was established on the energy storage properties of BT-BMO bulk ceramics by consulting relevant published literature. Three distinct feature vector spaces were constructed based on the physicochemical characteristics of constituent elements, related property information, and sintering process parameters. After that, multiple machine learning algorithm models were built to train and predict Wrec and η. For the prediction of energy storage density, the 10-fold cross-validation coefficient of determination (r2) and root-mean-squared error (RMSE) of the GBR model are 0.974 and 0.142, respectively. For the prediction of energy storage efficiency, the 10-fold cross-validation r2 and RMSE of the LGBM model are 0.894 and 0.068, respectively. To enhance the interpretability of the models, the Shapley additive explanation method was introduced, revealing and briefly analyzing the important features that influence the target performance. Finally, the material system of xBaTiO3-(1 – x)Bi(Zn2/3Ta1/3)O3, which was not included in the data set, was synthesized experimentally and tested. The experimental results were found to be close to the model’s predictions, thereby validating the effectiveness of the method. This study provides a new approach that could accelerate the development of dielectric materials with excellent energy storage properties.