基于操作参数深度优化的高炉铁水质量窄窗口智能化稳定控制

With the deterioration of the iron-making fuel quality of the blast furnace, to alleviate the situation that the mid-high level steel in China is heavily dependent on import, it is necessary to strengthen the quality control of the blast furnace molten iron and strictly control the fluctuation range of its quality parameters to show a "narrow window" feature. However, the production process in blast furnace has characteristics of harsh environment, serious coupling effect between the parameters and large delay, which brings great challenges for the stable control of molten iron quality. Therefore, a data reduction and customization algorithm is designed to construct a complete set of input parameters which best matches the demand of multivariable parameter prediction of molten iron quality. A multivariable parameter prediction model of molten iron quality based on local linear deep learning is established, and a multivariable prediction method for the changing trend of molten iron quality based on the probability density function is proposed. Moreover, a predictive control method based on improved cases, which is used for the stable control of the narrow window of molten iron quality, is also discussed. On the basis of the above methods, an intelligent deep optimization strategy of the operating parameters of the blast furnace iron-making process based on deep learning is put forward to deeply optimize the operating parameter range, and the closed-loop stability control of molten iron quality index will be realized. The feasibility of the proposed method is verified by the combination of simulation researches and practical engineering applications. And an intelligent stability control method, which is safety, robustness and agility, is finally developed for the stable control of the narrow window of molten iron quality. The industrial applications of the research results will enhance the quality of China's molten iron.

在高炉炼铁原燃料质量日益劣化的条件下，为缓解我国中高端钢种严重依赖进口的局面，必须加强高炉铁水质量控制，严格控制铁水质量参数的波动范围，呈现“窄窗口”特征，同时高炉生产过程具有环境恶劣、各参数间耦合严重、大滞后等特点，给铁水质量稳定控制带来巨大的挑战。为此，本项目设计数据约简定制算法，构建与多元铁水质量参数预测需求最匹配的完备输入参数集，建立基于局部线性深度学习的多元铁水质量参数预报模型，提出基于概率密度函数的多元铁水质量参数变化趋势预报方法，探讨基于趋势分布特征的铁水质量窄窗口稳定预测控制方法，在此基础上，提出基于深度学习的高炉炼铁过程操作参数智能深度优化，深度优化操作参数范围，实现铁水质量指标闭环稳定控制；结合仿真研究和实际工程应用研究，验证所提方法的可行性和有效性；形成具有安全性、鲁棒性与敏捷性的高炉铁水质量窄窗口智能化稳定控制方法，成果的工业应用将提升我国铁水质量。

铁水质量信息是直观表征冶炼过程中铁水质量、炉温状态、高炉能耗水平以及炉缸活跃程度的重要参数。高炉冶炼过程的密闭、高温和强粉尘等特点使得高炉铁水质量信息难以直接在线检测，造成高炉铁水质量精细化调控盲目。为此，本项目系统地开展了相关研究，取得丰硕的研究成果：①提出了高温工业特谱测温仪光学设计及防护技术，研制了高温工业特谱测温仪，获取了高炉出铁口铁水流的特谱热图像面源温度信息和感兴趣的铁水流区域，基于铁水与其他物体间的温度差异确定温度阈值，区分铁水区域与其他非铁水区域，建立了铁水温度与感兴趣区域面源信息间的映射模型，解决了高温、强辐射、易飞溅等特点下的出铁口铁水温度在线检测难题。②提出了基于粗糙集预处理的优质数据集构建方法，针对冶炼过程工况的渐变特性，为充分挖掘数据的分布特点和模糊聚类特征，提出了一种自适应局部加权聚类算法，实现对滑动窗口内高炉数据样本模糊属性的挖掘，获得样本的类别划分；针对冶炼过程的大时滞特性, 定义相邻时间节点间的硅含量工况迁移代价函数, 并提出多源路径寻优算法, 实现冶炼过程中硅含量最优工况迁移路径及当前时刻硅含量最优预测值的求解。③提出了高炉冶炼过程多层次特征描述和提取策略，并将铁水硅含量的值在时间轴上进行划窗分割使用多项式拟合，实现了在动态性、无明显周期性和聚类性下的硅含量趋势分布特征的提取，克服了硅含量趋势特征难以定性描述的难题，设计了一种基于目标相关的深度网络实现了不平衡样本集下的铁水硅含量变化趋势在线精确预报。④研制了大型高炉铁水质量监控平台，包括基于特谱热图像高炉出铁口高温铁水温度在线精确检测系统、基于数据驱动的铁水硅含量及其变化趋势实时预报系统。为现场提供了更全面的铁水实时温度、实时硅含量及其变化趋势等关键铁水冶炼数据，为钢铁企业实现安全生产、优化控制、节能降耗、提质增效提供了有力的保证。