Recurrent neuro-fuzzy predictors for multi-step prediction of v-i characteristics of electric arc furnaces

Presents an application of recurrent neuro-fuzzy systems to predict electric arc furnaces voltage and current. The primary objective is to investigate capability of adaptive fuzzy systems to predict the v-i characteristics of nonlinear, multivariable, complex systems such as electric furnaces. The novelties of this work are proposing a combination of recurrent neuro-fuzzy networks deemed suitable for prediction and using a wider window of observation whereby multi-step predictions can be made. In particular, the paper investigates the likelihood of long-term prediction for both furnace current and voltage. Successful implementations of recurrent neuro-fuzzy predictors are described and their performances are illustrated using the results obtained from adaptive neuro-fuzzy networks and recorded data