A Bayesian Networks Structure Learning and Reasoning-Based Byproduct Gas Scheduling in Steel Industry

Abstract

It is very crucial for the byproduct gas system in steel industry to perform an accurate and timely scheduling, which enables to reasonably utilize the energy resources and effectively reduce the production cost of enterprises. In this study, a novel data-driven-based dynamic scheduling thought is proposed for the real-time gas scheduling, in which a probability relationship described by a Bayesian network is modeled to determine the adjustable gas users that impact on the gas tanks level, and to give their scheduling amounts online by maximizing the posterior probability of the users' operational statuses. For the practical applicability, the obtained scheduling solution can be further verified by a recurrent neural network reported in literature. To indicate the effectiveness of the proposed data-driven scheduling method, the real gas flow data coming from a steel plant in China are employed, and the experimental results indicate that the proposed method can provide real-time and scientific gas scheduling solution for the energy system of steel industry.