A novel RBF neural network based on data dispersion level and its application in BOF endpoint prediction

Abstract

The Basic Oxygen Furnace (BOF) process is a primary method of steel-making. The endpoint targets must be strictly control. However, it is difficult to accurately predict endpoint targets in BOF. In this paper, a clustering method is proposed in which data dispersion level and new metric are introduced respectively. And the proposed clustering method is applied to obtain accurate neural network centers in order to improve accuracy of Radial Basis Function (RBF) neural network. Then a novel RBF neural network is built for the endpoint prediction in BOF process. Finally, an example of endpoint prediction is shown, the simulation results indicate that the influence of disperse and noisy data is decreased, clustering accuracy is increased and the accuracy of endpoint prediction based on RBF neural networks is improved.