A support vector machine based method for parameter estimation of an electric arc furnace model

In the iron and steel industry, electric arc furnaces (EAFs) are used in the melting and refining process of metals. They are known to demand large amounts of reactive power and cause significant power quality (PQ) problems due to their highly non-linear time varying voltage-current characteristic. Several EAF models have been proposed with the purpose to predict the voltage and current waveforms, to assess the performance of different compensating devices such as static var compensator, synchronous static compensator, active power filters, and –still under study– energy storage systems, and also for planning the installation of iron and steel facilities considering existing real data from similar facilities. An important aspect of these models is related to the estimation of their parameters. This paper presents a new method to estimate the parameters of an EAF model. The method utilizes a multiple-input multiple-output regressor based on support vector machine, that maps from voltage characteristics of the electric arc to the values of the model parameters. The multidimensional support vector regressor (M-SVR) is designed in the training phase, using data from several simulations of the EAF model. These simulations are carried out adjusting the parameters of the model within the search space, and considering the real arc current as input to the model. Then, in the validation phase, for the real voltage waveform, the estimated parameters are obtained using each regressor of the M-SVR. The proposed method is validated by the comparison between the waveforms obtained using the EAF model with actual data from a steel plant. Results show that the relative error between the fundamental component of the current and voltage, for real and simulated waveforms, are 2.1% and 6.3% respectively.