Prediction of bending force in the hot strip rolling process using artificial neural network and genetic algorithm (ANN-GA)

Abstract

An artificial neural network (ANN) optimized by genetic algorithm (GA) is an established prediction model of bending force in hot strip rolling. The data are collected from factory of steel manufacture. Entrance temperature and thickness, exit thickness, strip width, rolling force, rolling speed, roll shifting, target profile, and yield strength of strip are selected to be independent variables as network inputs. MATLAB software is utilized for establishing GA-ANN model and achieving the purpose of obtaining the bending force as results of setup model, as well as the GA method is used to optimize the initial weights and biases of the backpropagation neural network. Mean absolute error (MAE), mean absolute percentage error (MAPE), root mean squared error (RMSE), and correlation coefficient are adapted to evaluate the performance of the model. The predictive results are compared with the measured results to verify the accuracy of the GA-ANN prediction model. It is found that the optimization effect is the best with the population size 40 crossover probability of 0.7 and the mutation probability of 0.05 at the same time, the fitness function value can reach 80.7. In addition, the ANN architecture 9-11-1 trained with Bayesian regulation “trainbr” function has the best performance with mean absolute error of 0.01 and correlation coefficient of 0.983. With a deeper understanding of neural networks through the analysis of the GA-ANN model, the proposed model can be flexibly used for on-line controlling and rolling schedule optimizing.