高炉炼铁过程多层面大数据潜结构建模与全方位故障诊断

This project aims at establishing data-driven multi-level big data latent structure modeling and comprehensive fault diagnosis methods for blast furnace ironmaking processes, which are difficult to build global operation models. The blast furnace ironmaking processes, which operate in a multi-level manner, collect irregular index-level data (i.e., temperature of molten iron, silicon content, sulfur content, and coke rate) with a slow sampling frequency, and collect unreliable operation-level data (e.g., surface temperature distribution of molten materials) and process-level data (e.g., lining temperature, cool air flow, and hot air pressure) with a regular high sampling frequency. Traditional projection to latent structures based modeling and fault diagnosis methods encounter three problems. They cannot i) diagnose unexpected surface temperature distribution faults; ii) model irregular sampled index-level data; and iii) diagnose root-cause of the faults exactly.. The project tasks include:. 1. Propose and develop a latent structure modeling method between unreliable surface temperature distribution data of raw materials and process data for the blast furnace ironmaking processes. The corresponding comprehensive fault monitoring method is also proposed.. 2. Propose and develop a latent structure modeling method between irregularly sampled index-level data and regularly sampled process data for the blast furnace ironmaking processes. The corresponding comprehensive fault monitoring method is also proposed.. 3. Propose and develop time-series data modeling and faulty root-cause diagnosis methods for blast furnace ironmaking processes.. 4. Propose and develop big data implementation techniques and application verification with big data from practical blast furnace ironmaking processes. The operation faults, e.g., index-level abnormal quality of molten iron, operation-level unexpected surface temperature distribution, and process-level abnormal temperature, are consequently diagnosed effectively.

本项目以难以建立全局模型的钢铁生产高炉炼铁过程为背景，研究数据驱动的多层面大数据潜结构建模与全方位诊断方法。多层面运行的高炉炼铁过程以慢速率不规则采样指标层数据（铁水温度、硅含量、硫含量、焦比），以快速率规则采样运行层数据（以料面温度分布为衡量指标，且不可靠）和过程层数据（炉衬温度、冷风流量、热风压力等）。传统的数据潜结构建模与诊断方法无法诊断料面温度分布故障，无法处理不规则采样指标数据，无法准确诊断故障原因。. 针对上述问题，本项目研究：（1）高炉炼铁过程不可靠的料面温度分布与过程数据间潜结构建模与联合监控；（2）高炉炼铁过程不规则指标与过程数据间潜结构建模与联合监控；（3）高炉炼铁过程时间序列数据建模与故障原因全方位诊断；（4）上述理论方法的大数据实现技术与应用验证。本项目的研究可实现指标层铁水指标不合格故障、运行层料面温度分布故障以及过程层高炉炉衬温度场故障等的诊断。

本项目以难以建立全局模型的钢铁生产高炉炼铁过程等复杂工业过程为背景，针对传统的数据潜结构建模与诊断方法难以处理不规则采样的指标数据（包括铁水温度、硅含量、硫含量、焦比），难以建立不同时间尺度的过程数据（包括炉衬温度、冷风流量、热风压力、料面温度等）与指标数据之间的动态非线性相关关系、无法分离过程异常变化和指标异常变化，研究数据驱动的多层面大数据潜结构建模与全方位诊断方法。取得的主要研究成果包括：（1）建立由数据动态特征提取、并发潜结构建模、核投影组成的高炉炼铁过程多层次动态潜结构建模体系结构，率先发表的大数据建模方面的综述论文获中国科协顶尖学术论文奖。（2）提出一系列大数据建模与诊断的理论方法，包括：并发典型相关分析、动态并发典型相关分析、动态并发核偏最小二乘、动态加权典型相关分析等建模与全方位监控方法，以及多层次动态潜结构框架内基于多方向重构、动态时间规整和图网络的故障原因诊断方法。（3）突破了多层面不规则动态数据潜结构建模、非线性系统多层次潜结构建模与综合故障诊断的科学难题，解决了高炉炼铁过程铁水指标相关故障和过程特有故障分离与可视化监控问题。（4）利用仿真实验和高炉炼铁过程实际数据对所提出的方法进行了一系列实验验证，结果表明了所提方法的有效性和优越性。.本项目发表或录用标注的高水平学术论文19篇（卓越期刊论文10篇），其中，控制领域国际顶级期刊IEEE汇刊和IFAC会刊长文3篇，以及国内控制领域权威期刊《自动化学报》和《中国科学：信息科学》论文7 篇。依托本项目成果，项目负责人受邀在2020年中国过程控制会议做特邀报告，获选兴辽英才青年拔尖人才。