Bayesian neural networks based pipeline pressure prediction for steam system in steel industry

The steam is a very important energy resource medium in steel industry, and its reasonable utilization is of great significance for saving the energy and reducing the production cost. In practice, the accurate prediction for steam pipeline pressure is the prerequisite for the energy scheduling and ensuring the stability of the pipeline networks. Considering that the industrial data are always mixed with high frequency noises, a Bayesian neural network based prediction method is proposed in this study, where the uncertainties of the input and the output are both taken into account to reduce the impact of the industrial noises on the prediction accuracy. By using the proposed method, not only the predicted pressure values, but also the predicted range can be apparently quantified, which help the energy scheduling workers to well evaluate the prediction performance and implement the scheduling process. To verify the effectiveness of the proposed method, a large number of real energy data coming from a certain steel plant in China are employed, and the experimental results indicate that the proposed method is applicable to the industrial production.