The magnetic properties prediction and composition design of La-Co substitution Sr-hexaferrite based on high-through experiments and machine learning

La-Co co-substitution is an effective method to improve the magnetic properties of Sr-hexaferrite but knowing the most suitable substitution amount requires many trial and error experiments. Combining high-throughput ex-periments with machine learning techniques is a promising way to quickly realize the composition design. With that in mind, we adopted three frequently-used machine learning models, namely Gaussian process regression (GPR), support vector regression (SVR), and radial basis function network (RBFN) as candidate prediction models to learn the 145 samples accumulated from the high-throughput experiments. Furthermore, three nature-inspired algorithms called particle swarm optimization (PSO), genetic algorithm (GA), and grey wolf algorithm (GWA) were applied to search for the optimal combination of the hyper-parameters, improving the performance of the machine learning models. To compare the accuracy of these models and simultaneously validate the experi-mental data's reliability, the 20 samples under the same experimental conditions obtained from literature were selected as the testing data. The comparison results showed that the SVR model with the GWA algorithm (SVR-GWA) performed better than the other methods. After that, the predicted figures of saturation magnetization (MS) and coercivity (HcJ) were obtained by the SVR-GWA model. Moreover, five compositions not involved in training and testing data were randomly selected from the prediction figures to further verify the reliability of the SVR-GWA model, which achieves the goal of fast and accurate composition design of La-Co substitution Sr-hexaferrite.