A Scheduling Method for Blast Furnace Gas Systems Based on Conditional Generative Adversarial Networks and Process Rules

Blast furnace gas (BFG), a vital byproduct energy resource in the ironmaking process, can be efficiently managed through proper scheduling to minimize emissions and augment productivity. A novel scheduling method for BFG systems based on an improved Conditional Generative Adversarial Network (CGAN) and process rules is proposed. Based on the Wasserstein generative adversarial network with gradient penalty (WGAN-GP) and the process production mechanism under ideal conditions, a scenario generative model is constructed to constrain the training process with the rules of the changes among scenario factors. Labels are used as condition information for CGAN to generate the corresponding labeled scenarios. Then, by matching real-time operational data with the generated scenarios, the scheduling solution is derived for the given condition. Using the actual data collected from a large steel enterprise, the validation experiments confirm that the scheduling scenarios generated by the proposed method are consistent with actual operational patterns. Moreover, it delivers viable and rational scheduling solutions for new operating conditions, safeguarding the safe operation of the BFG system.

I. Introduction

To achieve more effective energy conservation and emission reduction, as well as cost-effectiveness, steel enterprises have set higher standards for the utilization of primary energy and the recovery of secondary energy and also increased the requirements for the operational skills and proficiency of on-site operators. The main task of the blast furnace gas (BFG) system scheduling is to ensure the supply of gas consumption while maintaining the gas tank level within a safe fluctuation range. Currently, operators rely primarily on the Supervisory Control and Data Acquisition (SCADA) system to monitor the production rhythm of major users such as ironmaking, steelmaking, and rolling and to allocate energy to various units based on the numerical fluctuations of energy flow, using experience under the premise of ensuring safe production. However, this scheduling method requires operators to continuously pay close attention to the SCADA interface, which is not only labor-intensive but also demands quick response with scheduling solutions in the event of a system safety incident, placing a high demand on computational capabilities. Moreover, to ensure system safety, a tentative pre-scheduling is usually conducted, followed by the formulation of the next plan based on system trends, which results in low precision and a lot of repetitive operations.