A hybrid model for the prediction of aluminum foil output thickness in cold rolling process

Abstract

This study proposes a hybrid model composed of multiple prediction algorithms and an autoregressive moving average (ARMA) module for the thickness prediction. In order to attain higher accuracy, the prediction algorithms were globally combined by simple voting to reduce the effect of the inductive bias imposed by each algorithm on the dataset. The global multiexpert combination (GMEC) system included the multilayer perceptron neural network (MLPNN), radial basis function network (RBFN), multiple linear regression (MLR), and support vector machines (SVM) algorithms. The proposed hybrid model extends the GMEC system by integrating an ARMA module for the output. On the test dataset, the mean absolute error (MEA) and root mean squared error (RMSE) were better for the hybrid model than the GMEC system. The GMEC system had approximately twice better performance than the MLPNN, which was the best among the learners. The performance was significantly improved via the hybrid model in terms of correlation coefficient (R). The results suggested that the proposed hybrid model can be used for more accurate and precise prediction of aluminum foil output thickness.