Application of shallow neural networks in electric arc furnace modelling

Klimas, Maciej; Grabowski, Dariusz

doi:10.1109/eeeic/icpseurope51590.2021.9584512

Cited by 5 publications

(1 citation statement)

References 23 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Here, a genetic algorithm is utilized to calculate the parameters using the arc current and voltage waveforms. In [24], three ANN-based models for modelling EAF are used in addition to a stochastic moving average (MA)-based extension. ANFIS is used in [25] to represent the modelling EAF.…”

Section: D) Black-box Modelsmentioning

confidence: 99%

Optimized time varying parameters of the power balance equation model for electric arc furnaces

Babaei,

Samet

2024

IET Generation Trans & Dist

View full text Add to dashboard Cite

An accurate model of electric arc furnaces (EAFs) is required to analyse their impact on the system's power quality. One of the well‐known approaches to model EAF is using the power balance equation. This model has two terms, each one has a coefficient named the model parameter. Also, both terms are powered to values, which are called model orders. So, the model includes two model parameters and two model orders. Model orders highly affect the modelling precision. The optimum model orders have not been obtained yet in previous studies. Here, based on many actual records from EAFs, the best model orders for refining, melting, and scrapping stages are attained. Unlike the original power balance equation which has constant model parameters, the first proposed model uses the attained optimal model orders with the time‐varying model parameters. In the second proposed model, the model orders are also considered time‐varying. Model parameters, which are updated every 10 ms, are obtained using autoregressive moving average (ARMA) models. The correctness of the models is assessed by calculating some important electric quantities using the actual data and the proposed models. Finally, the time‐varying parameters are generated using two MATLAB codes.

show abstract

Section: D) Black-box Modelsmentioning

confidence: 99%