基于概率图模型的钢铁企业副产煤气系统动态调度方法研究

Currently, the overall level of recovery and utilization of byproduct gas in steel enterprises is not high enough to met the criterion of green development and manufacturing proposed by our government. Besides some process mechanism restrictions of the occurrence and consumption aspects, the lower ability of scheduling and decision-making for the system is also a very significant reason. In this project, the probabilistic graphical models (PGMs)-based theories and methods are studied for the dynamic scheduling problem of the byproduct gas system. The detailed research contents are described as follows. First, a mixed PGM with latent variables that can be learned based on the incomplete training dataset will be studied for classification of the operational conditions. Second, a new fast online-learning method for the PGM will be researched for constructing and updating the scheduling model. Third, as for the extraction of the scheduling solution, a reasoning method that can guarantee the global convergence will be developed for the mixed PGM. Finally, an equivalent PGM of the byproduct gas system will be constructed for verifying the extracted scheduling solution. Based on this project, it is expected to form some new theories and methods based on PGMs for the dynamic scheduling problems of the byproduct gas system. The application of the research achievements will effectively improve the level of dynamic scheduling of the byproduct gas system. And then, the current situation of high energy consumption and emission in the domestic iron and steel industry can be changed, the cost of the steel production can be effectively reduced, and the international competitiveness of China's iron and steel industry can also be enhanced.

当前钢铁企业副产煤气整体的回收和利用水平不高，达不到国家提出的绿色发展和绿色制造的标准，除了发生和消耗环节的一些工艺机理限制外，系统的调度决策水平不高也是重要原因。本项目以钢铁企业副产煤气系统为研究对象，拟基于概率图模型的理论和方法研究系统的动态调度问题，具体包括面向副产煤气系统工况划分的非完备数据集下的含有隐变量的混合概率图模型的构建方法；面向副产煤气系统调度问题建模的概率图在线快速学习理论与方法；面向系统实时调度方案的提取研究一种全局收敛的混合概率图推理方法；最后研究系统的等价概率图构建方法，及基于等价概率图模型的调度方案在线验证方法。通过本项目的研究，有望形成基于概率图模型的副产煤气系统动态调度新理论和新方法，成果的应用将有效提高副产煤气系统的调度决策水平，改变国内钢铁行业高能耗、高排放的现状，降低钢铁生产的成本，增强我国钢铁行业的国际竞争力。

本课题基于上海宝钢能源中心已建立的能源管理系统这一信息平台，将理论研究与应用研究相结合，分析系统中的大量历史数据、实时数据和调度决策数据，调度专家经验数据，采用基于概率图的相关方法，开展面向钢铁企业副产能源系统的预测及优化调度方法研究。理论研究方面，针对副产煤气系统的产消流量动态特性复杂，波动趋势存在不确定性的问题，对副产煤气系统建立了基于高斯粒度网络集成的流量区间预测模型，估计流量趋势波动的不确定性。针对工业数据存在缺失和不完全性，提出了基于变分推理的动态贝叶斯网络模型用于构建输入不确定的副产煤气流量预测模型。针对副产煤气系统的优化调度问题，分别提出了数据驱动的预测调度方法，将知识驱动与数学规划相结合的副产煤气系统优化调度方法，知识驱动的高炉煤气系统控制与优化调度方法，基于因果概率图模型的副产煤气系统调度方法，基于隐树模型的副产能源系统的实时调整方法。应用研究方面，将提出的若干种副产煤气系统的预测及优化调度方法以软件模块的形式实现，并将其集成于实验室所开发的钢铁企业能源系统优化调度软件，为工业现场的调度专家提供指导，以提高我国钢铁企业副产煤气的利用效率，降低能耗水平。