Updating stochastic models of arc furnace reactive power by genetic algorithm

The time varying nature of electric arc furnace (EAF) gives rise to voltage fluctuations. The ability of static VAr compensator (SVC) in flicker reduction is limited by delays in reactive power measurements and thyristor ignition. To improve the SVC performance in flicker compensation, EAF reactive power can be predicted for a half cycle ahead by using appropriate ARMA models. This paper uses huge field data, collected from eight arc furnaces and demonstrates that the EAF reactive power models coefficients and their variations are different from one data record to another. Therefore it is necessary to update the model coefficients for prediction purposes. For this purpose, genetic algorithm (GA) is used to determine the prediction relationship coefficients online. By applying the method to the data records and using some indices such as newly defined indices based on concepts of flicker frequencies and power spectral density, the transient and steady state performances of the method are studied in EAF reactive power prediction and compared with those of normalized least mean square (NLMS) and recursive least square (RLS) algorithms. It is demonstrated that the overall performance of online GA is better than of other two algorithms.