Effective scheduling for byproduct gas systems of steel industry is becoming increasingly vital for maintaining their safe operating and improving energy utilization. Considering that the existing studies failed to capture the dynamic changes in the production environment, a novel dynamic scheduling framework is proposed that seamingly integrates expert knowledge with a dynamic programming process. Given the phase characteristics of the steelmaking processes, data series are first partitioned into information granules based on the production plan to form the knowledge-based initial policies. To achieve dynamic scheduling process, a two-stage value function approximation method is proposed, where in the first stage one learns an event-driven Q-function by the fuzzy rule-based states, and then an action fitting strategy is developed for evaluating continuous actions. Considering the difficulties of establishing a mechanism-based model, the state transition process is described by a granular prediction model to simulate taking actions. On their basis, a dynamic compensation for the initial policies is finally achieved. A number of comparative experiments are conducted by utilizing the practical data coming from a steel plant. The results show that the proposed method can deliver effective solutions for long-term scheduling scenarios. Note to Practitioners—Given that the steelmaking process is a discontinuous one and the byproduct gas system can hardly be described by a physical or mechanism-based model, its energy scheduling works is usually performed by manual approach or using static optimization methods, which would lead to low accuracy and a waste of energy. Since a large number of real-time data had been accumulated by the SCADA system implemented in most steel plants, a data-driven dynamic scheduling approach is proposed in this study. The proposed method takes advantages of the expert knowledge and production plan data, and produces dynamic scheduling solutions by utilizing an actor-critic learning process. The application system on the basis of the proposed method can adapt to different scenarios and ensure long-term safety operations of the gas tanks. Furthermore, since there may be missing data or outliners in the acquired data collected by the SCADA onsite, it is necessary to perform data imputation and filtering methods to guarantee the data integrity and reliability. This study avoids the redundant introduction of such preliminary preprocessing methods for the sample data.

中文翻译：

---

结合专家知识和生产计划的副产气系统动态调度框架


钢铁工业副产气体系统的有效调度对于维持其安全运行和提高能源利用率变得越来越重要。考虑到现有研究未能捕捉生产环境的动态变化，提出了一种新颖的动态调度框架，将专家知识与动态编程过程无缝集成。考虑到炼钢过程的阶段特征，首先根据生产计划将数据序列划分为信息粒，形成基于知识的初始策略。为了实现动态调度过程，提出了一种两阶段价值函数逼近方法，其中在第一阶段通过基于模糊规则的状态学习事件驱动的Q函数，然后开发动作拟合策略来评估连续的行动。考虑到建立基于机制的模型的困难，通过粒度预测模型来描述状态转换过程来模拟采取动作。在此基础上，最终实现对初始政策的动态补偿。利用钢厂的实际数据进行了多次对比实验。结果表明，所提出的方法可以为长期调度场景提供有效的解决方案。从业者须知——由于炼钢过程是一个不连续的过程，且副产煤气系统很难用物理或基于机理的模型来描述，其能源调度工作通常采用手工方法或静态优化方法进行，这会导致精度低且浪费能源。 鉴于大多数钢厂实施的SCADA系统积累了大量实时数据，本研究提出了一种数据驱动的动态调度方法。所提出的方法利用专家知识和生产计划数据，并利用演员批评学习过程产生动态调度解决方案。基于该方法的应用系统能够适应不同的场景，保证储气罐的长期安全运行。此外，由于SCADA现场采集的数据可能存在缺失数据或轮廓线，因此需要进行数据插补和过滤方法以保证数据的完整性和可靠性。本研究避免了对样本数据进行此类初步预处理方法的冗余引入。