Echo state network based prediction intervals estimation for blast furnace gas pipeline pressure in steel industry

The pipeline pressure of blast furnace gas (BFG) system in steel industry provides effective information for the energy scheduling operations. However, due to the complexity of the byproduct gas pipeline network and the large fluctuations of the gas flow, it is rather difficult to establish an accurate prediction model for the pipeline pressure. Additionally, the quantitative reliability of the prediction accuracy is the key concerns of energy scheduling workers since there are always a variety of uncertainties in industrial process. In this study, an echo state network (ESN) modelling with output feedback is proposed to predict the BFG pipeline pressure. Given the gas flows data and the pressures sampled from the sensors are generally accompanied with noise, a Bayesian framework for the prediction intervals (PIs) is designed, which can quantify the input and output noises. To verify the effectiveness of the proposed method, a number of prediction experiments coming from industrial data are conducted here. And, the experimental results indicate that the proposed approach has a satisfactory performance on PIs for the pipeline pressure.