Extended Kalman filter approach to electric arc furnace forecasting

In this thesis, we propose and implement a new approach for building an online self-adjusting model for prediction of v-i characteristic of a multivariate time series obtained from an operational electrical arc furnace. The proposed methodology is based on the Kalman filtering method, and is used for prediction of the arc furnace voltage using the past history of the current and voltage. The main advantage of the proposed approach over similar earlier related work is the ability to adapt during the operation of the furnace. In this study, three different hybrid models have been developed based on the extended Kalman filtering technique and one of the following methodologies: (i) a linar auto regressive model; (ii) fuzzy logic, (iii) wavelet analysis. The results compare well with those of earlier work and clearly indicate that the augmentation of the above mentioned approaches with the extended Kalman filter improves the prediction accuracy.