基于知识和顺惯决策的挥发窑操作模式识别与迁移优化

Volatile kilns are widely used in metallurgical, building materials, chemical industries and so on, which are a kind of typical high energy consumption and high pollution thermal furnaces. Their operation involves complex heat and mass transfer mechanism in addition to being closely related to the inlet solid characteristics. Moreover, the operational pattern identification and transition optimization are the key to the safe operation of such kilns. A new method based on the knowledge automation and heterogeneous information fusion for operational pattern identification and transition optimization is proposed after in-depth investigating volatile drying processes and making the best of their mass data. The knowledge automation control system is decomposed into independent two-dimensional framework for continuous control/ discrete learning, in which the ensemble learning based methods on mining, extraction, reducing and online correction of good operational pattern base are discussed in the learning dimension, and the operational pattern identification and transition optimization are processed in the control dimension. Aiming at solving measurement difficulty of thermal conditions in kilns, the flame image modeling methods and multi-model fusion theories are studied. On this basis, the operational pattern identification for given steady states is solved by using ensemble knowledge and the operational pattern transition optimization is coped with the sequential decision making optimization methods based on reinforcement learning. New solutions for complex industrial process modeling and optimization control such as volatile kilns are provided, and the important guiding significance to energy saving and emission reduction is got for producing process of processing industry.

挥发窑广泛应用于冶金、建材、化工等多个行业，是一类典型的高能耗高污染热工窑炉，其运行不仅涉及复杂的传热、传质机理，还与入料成分特性密切相关，而操作模式识别与迁移优化是这类工业窑炉安全运行的关键。本课题通过深入分析挥发窑生产过程机理，充分利用生产过程的海量数据，提出基于知识自动化与异类信息融合的操作模式识别与迁移优化新方法，将知识自动化系统分解成独立的连续控制/离散学习二维框架，在学习维上探讨基于集成学习的优良操作模式库挖掘、规则抽取、精简及在线修正方法；在控制维上进行操作模式识别与迁移优化。针对烧成带燃烧测量问题，研究火焰图像建模方法及多模型融合理论，在此基础上运用集成知识解决挥发窑稳态操作模式识别和给定问题，提出基于强化学习的顺惯决策优化方法，实现挥发窑操作模式动态迁移优化。本项目的研究为挥发窑这类复杂工业过程建模和优化控制提供新的解决方法，对流程工业生产过程节能减排有重要的指导意义。

挥发窑广泛应用于冶金、建材、化工等多个行业，是一类典型的高能耗高污染热工窑炉，其运行不仅涉及复杂的传热、传质机理，还与物料的特性密切相关，同时窑本身是一个大惯性、大滞后性、存在诸多不确定性参数的控制对象，其主要部分在回转，导致关键工艺参数不能准确及时测量，窑内热工状况难以定量描述，生产过程技术经济指标波动较大，严重制约企业节能减排水平的提高。为此，项目组通过深入分析挥发窑生产过程机理，并充分利用生产过程海量数据，研究基于知识自动化的复杂工业过程操作模式识别与优化控制方法，提出了将知识自动化控制系统分解成独立的连续控制/离散学习二维框架，在学习维上研究以机理模型为基础，融合支持向量机、极限学习机等智能混合建模方法，建立了挥发窑生产过程智能集成关键状态与操作模式知识库及挥发窑生产过程模拟仿真模型，探讨了基于多核支持向量机的包含不确定信息的优良操作模式库挖掘、规则抽取、精简及在线修正方法；在控制维上进行操作模式识别与迁移优化研究，提出了基于强化学习的顺惯决策优化方法，探讨了在严格反馈形式下的多智能体分散控制策略，实现了基于性能评价和顺惯决策的操作模式动态迁移优化。针对烧成带燃烧测量问题，获取了烧成带高温摄像头图像记录数据，研究了火焰图像建模方法及多模型融合理论，提出了基于多信息融合的分层极限学习机方法预测产品质量，建立了挥发窑应用验证平台，实现了挥发窑生产过程高效运行。本项目的研究为挥发窑这类复杂工业过程建模和优化控制提供了新的解决方法，对流程工业生产过程节能减排有重要的指导意义。同时通过本项目的研究，目前共发表SCI、EI等论文16篇，授权发明专利7项，软件著作权1项，获湖南省科技进步奖三等奖2项，湖南省仪器仪表协会优秀论文一、二等奖各1项，培养硕士生6名，通过自动控制原理线上精品课程及自动化工程实训虚拟仿真课程湖南省一流课程认定，专著出版中，实现了预期目标。