A Novel Model Based on PSO-GA-RBF Approach for Predicting Roll Gap During Unsteady Processes

To enhance the control accuracy of strip thickness in the non-stationary process of cold continuous rolling, a PSO-GA-RBF based prediction model for the non-stationary roll gap is proposed. This model combines big data algorithms with actual production data, including roll gap, rolling force, and rolling speed. A 1450mm series cold rolling production line is selected as the research object. Pearson analysis is utilized to perform a correlation analysis on the original data indicators, while principal component analysis is employed to reduce the dimensionality of the sample data. The optimization process of particle swarm optimization is enhanced by introducing crossover and mutation operators from genetic algorithms. This optimization aims to improve the parameter selection mechanism of the radial basis function neural network, thereby enhancing the model's prediction accuracy and generalization performance. The performance of different prediction models is analyzed, comparing them based on evaluation indicators such as mean square error, average absolute error, and average absolute error percentage. The results indicate that the PSO-GA-RBF based roll gap prediction model demonstrates good predictive ability and generalization performance. To further validate the effectiveness of the prediction model, a comparison is drawn between the prediction model and the roll gap data obtained from engineering practice. The experimental results show a close correspondence in trend and period between the two, verifying the strong prediction performance of the roll gap prediction model. The validated model offers benefits such as reducing the workload of the automatic control system and improving the control accuracy of strip steel thickness.