Sticker breakout recognition based on temperature characteristics and stacking multi-classifier model

Abstract

All 61 sticker breakouts were obtained based on the on-line mould monitoring system during the conventional slab continuous casting. The temperature characteristics and temperature velocities were extracted and combined to characterise the process of sticker breakout. A new recognition model of stacking multi-classifier was developed and compared with four single classifiers, such as k-nearest neighbour, support vector classification, decision tree and logistic regression. The results show that all 61 sticker breakouts can be predicted correctly. The accuracy rate of stacking multi-classifier, decision tree, support vector classification, k-nearest neighbour and logistic regression are 98.3%, 96.0%, 91.6%, 91.6% and 85.4%, respectively. The stacking multi-classifier has the maximum accuracy rate, because it integrates the advantages of different single classifiers. The ROC curve demonstrates that the stacking multi-classifier has a better applicability and generalisation ability. This method not only improves the prediction accuracy, but also reduces the disturbance of slab continuous casting, which is helpful to produce the conventional slabs with high quality.