A prediction model for the activated roasting leaching composition of titanium-containing blast furnace slag based on PSO-GRNN

Activation roasting is a new method to recover and utilize titanium resources in titanium-containing blast furnace slag. In order to quickly obtain the content of components in the recycled slag through the reaction conditions, a Generalized Regression Neural Network prediction model based on Particle Swarm Optimization （PSO-GRNN） is established. With the help of Spearman’s correlation analysis to screen the characteristic variables as model inputs, PSO is used to optimize the weights and node thresholds of the GRNN neural network, and the advantages of the proposed model are identified by comparing it with the Partial Least Squares Regression （PLS）, Random Forest （RF） and Support Vector Regression （SVR） algorithms. The results show that the PSO-GRNN has the smallest RMSE and the largest R², indicating that the designed PSO-GRNN has the best model performance on this dataset, which can provide some guidance for subsequent experimental or industrial applications.