Application of Shallow Neural Networks in Electric Arc Furnace Modeling

Abstract

Electric arc furnaces (EAFs) are important appliances in the steelmaking industry, but they are characterized by a nonlinear, dynamic, and stochastic nature. Due to this fact, EAFs can have a negative influence on power systems. Measures to mitigate such problems can be designed properly only with knowledge of the influence of the load on the system. Therefore, it is necessary to have accurate EAF models that reflect the complicated character of such loads. Researchers use different approaches for EAF modeling, such as stochastic process analysis, differential equation models, or neural networks. This article presents the application of three models based on artificial neural networks (ANNs) in EAF modeling along with an extension based on the stochastic moving average (MA) process. The goal was to provide ANN models that are simple in structure in comparison to, e.g., deep learning methods used by other researchers. The first two models are built on multilayer perceptron networks, and the third applies a nonlinear autoregressive exogenous model with the help of differential equation transformed into the Hammerstein–Wiener model. ANN models are improved by adding an MA ingredient. The article describes the measurement data, the design of each approach, and the results of the EAF modeling.