Long Short-Term Memory Network Based Tapping Temperature Prediction Model for Electric Arc Furnace

For accurately predicting the tapping temperature of molten steel in electric arc furnace (EAF), a novel prediction model based on the long short-term memory (LSTM) network is proposed in this paper. The smelting process of EAF is firstly analyzed. Then, the prediction model is established based on the energy balance of molten steel in the furnace. Part of the input features of the model are denoted in the form of time series in order to reflect the time-variation characteristic of smelting variables. This characteristic is a critical factor affecting the tapping temperature and is usually ignored in traditional intelligent model. Subsequently, the LSTM network is applied to describe the long time-dependent relationship between the tapping temperature and time-series features. Finally, the model is examined by practical data. The influence of the length of time-series features on the prediction result is also discussed. Comparative experiments with some other intelligent models demonstrate that the proposed model can improve the prediction accuracy, and is usually more reasonable.