基于HSMM及SPCA的复杂石化过程多模态报警设计与溯源研究

Petrochemical industry is the pillar industry of national economy and is also a high-risk manufacturing process. It is crucial to implement effective monitoring and alarm management, which is the safeguard to move towards the new-generation intelligent manufacturing. At present, the prominent problem of "alarm overload" is common in its alarm system. How to realize accurate and timely alarm monitoring is an urgent problem to be solved. From the viewpoint of complex characteristics and operational datasets of the petrochemical industry and targeting at the key stages of design, operation and monitoring in the lifecycle of alarm management, we shall study new methods of multimodal alarm monitoring in this project. Aimed at the characteristics of multimodality, strong correlation and wide distribution of petrochemical process, multimodal, multi-module, multidimensional and multilayer causal modeling are studied to provide an accurate and reliable model basis. Aimed at static and independent alarm thresholds, a multimodal association threshold design method based on alarm probability plot and hidden semi-Markov model is proposed to generate accurate alarms. Aimed at the wide range of alarm propagation and the difficulty of centralized traceability, a distributed fusion diagnosis method based on online reduced kernel sparse principal component analysis is studied and reconstructed contribution based alarm propagation path mining is studied to achieve fast and accurate traceability. The effectiveness and applicability of proposed methods is verified via case study of complex ethylene production process. Thus, theoretical basis and technical support can be provided for alarm management and safe and efficient operation of petrochemical industry.

石化工业是国民经济支柱产业，也是高风险制造过程，对其实施有效的监控和报警管理至关重要，是走向新一代智能制造的安全保障。目前其报警系统普遍存在“报警过载”这一突出问题，如何实现准确及时的监控报警是迫切需要解决的问题。本项目从石化工业复杂特征和运行数据出发，面向报警管理生命周期中设计、运行、监控关键阶段，研究多模态报警监控新方法。针对石化过程多模态、强关联、分布广的特征，研究多模态、多模块、多维度、多层次因果建模，提供准确可靠的模型依据；针对报警阈值静态、独立问题，提出一种基于报警概率图和隐半马尔科夫模型的多模态关联阈值设计方法，实现准确报警；针对报警传播广、集中溯源难问题，研究基于在线约减核稀疏主元分析的分布式融合诊断方法，并基于重构贡献值挖掘报警传播路径，实现快速准确溯源。以复杂乙烯生产过程为应用对象，验证所提方法的有效性和普适性，为石化工业报警管理与安全高效运行提供理论依据和技术支持。

石化工业是国民经济支柱产业，是我国“智能制造”的先行领域和绿色发展的主战场，也是高风险工业。近年来，石化工业朝着大规模、复杂化、智能化方向快速发展，但其报警系统性能水平和安全运行效率与国际先进水平还有差距，仍存在较为严重的报警过载的问题，一个重要原因就是缺乏保障石化过程安全稳定高效运行的报警监测和溯源诊断技术。本项目从石化工业复杂特征和运行数据出发，面向报警管理生命周期中设计、运行、监控关键阶段，针对包括石化过程在内的工业过程报警监测和溯源诊断的关键难题，系统分析过程本身以及过程数据的复杂特征，研究面向不同特性的复杂过程报警监测新方法，取得了若干创新性研究成果，主要包括：（1）针对大规模全流程复杂过程，克服了过程大规模、多模态、多阶段相关特性难点，提出了基于相关特征学习和改进聚类算法的过程多模态、多阶段、多模块、多层次建模方法；（2）针对非线性、动态性过程，克服了强关联、强非线性、动态性的难点，提出了基于多元统计分析和深度特征提取的智能报警和预警策略；（3）针对大规模、因果关联过程，克服了复杂特性带来的报警溯源诊断难的难点，提出了基于因果分析和深度特征提取的智能报警溯源诊断方法。项目执行期间，项目负责人和成员展开密切合作研究，取得了良好的研究成果，并培养了多名研究生：发表学术论文18篇，其中SCI/EI收录17篇，申请发明专利7项，授权发明专利1项，获得软件著作权1项。项目负责人独立培养硕士研究生4名，协助培养博士研究生4名、硕士研究生8名。本项目围绕报警过载这一难题展开研究，通过方法集成和融合，形成了面向智能报警监测和溯源诊断的基础理论、方法和应用体系，并逐步研发系统原型，为提升石化过程安全稳定运行水平提供有效的理论依据和技术支持。