Comparisons of Element Yield Rate Prediction Using Feed-Forward Neural Networks and Support Vector Machine

For the complexity of ladle furnace refining production process, it's impossible to establish accurate mathematical prediction model for element yield rate that is an important parameter in the process of alloy adding. Model selection is the key factor of better element yield rate prediction. In this paper, feed-forward neural networks (FNN) and support vector machine (SVM) are chosen as candidate modeling methods. We introduce that, under certain condition, FNN and SVM can be transformed into each other. Then an analysis of the essential difference between two algorithms is carried out. The element yield rate prediction models were set up using different FNN and epsilon-SVR. The comparison results show that modeling by e-SVR can meet the production requirements and has better prediction accuracy than by FNN.