Reactive power compensation in electric arc furnaces using prediction intervals

This paper proposes a probabilistic method to model the uncertainty of reactive power compensation by static VAr compensators (SVCs) in electric arc furnaces (EAFs). The time-varying characteristics of EAF accentuate the voltage fluctuations and produce flicker in power lines as well as neighboring loads. In order to solve this issue, quick and accurate response of SVC within a half-cycle ahead is required. This paper proposes a nonparametric approach based on lower upper bound estimation method to construct prediction intervals (PIs) for the reactive power in EAFs. Due to the nonlinear nature of reactive power signals in EAFs, a set of PIs are produced and combined to find an optimal aggregated PI. The proposed prediction method provides a faster-than-real-time monitoring of SVC, which aims at high speed and efficient reactive power compensation. In order to find the most satisfying PIs with high coverage probability and low average width, an optimization algorithm is developed to improve the training process of neural networks. The appropriate performance of the proposed method is examined on the practical data gathered from the Mobarakeh Steel Company, Iran.