The prediction of the silicon content in hot metal based on GRU using DTFT denoising

Abstract

The silicon content in hot metal reveals the blast furnace(BF) thermal state, which is closely related to the energy consumed and the quality of the product. So it is vital to predict the silicon content in hot metal for the monitor and operation of BF. Aiming at the challenge brought by industrial data noise and singularity during BF ironmaking process, this paper proposes the method of the prediction of silicon content using the time-series data. First there is the introduction of designing low-pass filter(LPF) based ondiscrete-time fourier transform(DTFT), and then the completion of missing value based on multivariate regression will be discussed. The algorithmof LSTM, GRU are explained as the theoretical basis of the models. Finally, their performance will be provided to prove the validation of the theoretical analysis. The result shows that GRU with DTFT denoising and completion of missing value has faster convergent rate than other models and can achieve the lowest MSE and highest hit rate.