Generative adversarial network–based real-time temperature prediction model for heating stage of electric arc furnace

For accurately predicting the molten steel temperature of heating stage in electric arc furnace (EAF) in real time, a novel prediction model based on the generative adversarial network (GAN) is proposed in this paper. First, the generator is specially designed based on the simplified energy balance of molten steel combined with long short-term memory (LSTM) network. The sequential smelting variables are used as the input of generator, which is an effective representation of the time-variant EAF operations. Meanwhile, the discriminator is established to indicate the deviation of the changing trend between the generator predicted temperature and the simulated temperature. Here, the simulated temperature is produced according to smelting experience which is a good supplement to the sparse temperature measurements. Subsequently, the loss function of the generator is improved to consider both the accuracy of predicted temperature and the correctness of temperature changing trend. Through alternate training the discriminator and generator, the generator is finally able to predict the temperature of molten steel in real time with a better precision. Experiments with practical data verify the effectiveness of the proposed model.