复杂工业过程报警识别分析与依赖溯源关键技术研究

Accurate and efficient alarm system is crucial for improving the safety and operability of production processes. In view of alarm system for complex industry process, alarm flooding and identification difficulties are the main problems. In this project, firstly, a nonlinear interval estimation method for alarm threshold is proposed. An upper lower bound estimation model is established for alarm threshold. Then, a comprehensive evaluation index is constructed to measure the quality of interval estimation. And the optimization is made to acquire the alarm threshold. Secondly, a feature reconstruction method for alarm features is proposed to enlarge the difference for alarm features of different types. And information feedback mechanism is introduced to construct dynamic information memory feedback-based deep-learning network, which is used to realize on-line alarm classification identification and improve identification accuracy and anti-interference ability. Thirdly, an alarm dependence root-cause method is proposed. By introducing extended finite state machine (EFSM), multi-layer EFSM model is established for alarm system. Then, the intra-transition and inter-transition data dependence and transitivity are explored, EFSM data dependence graph and dependence root-cause model for alarm system are established, so as to analyze and locate the alarm roots. Taking the Tennessee Eastman (TE) process and typical petrochemical ethylene production process as the application objects, the new theory and methods of alarm identification analysis and dependence root-cause are formed for complex industry process, and the software prototype of online identification and root-cause for alarm system is explored, which provides effective theoretical and technical supports for the safety of process production.

准确、高效的报警系统对提高生产过程安全性和可操作性至关重要。针对复杂工业过程报警泛滥、难以辨别等问题，本课题提出一种报警阈值非线性区间估计方法，通过建立阈值上下限估计模型，构建阈值区间估计质量综合评价指标，实现报警阈值优化设置；提出报警特征重构方法，放大报警特征差异，并引入信息反馈机制，构建基于动态信息记忆反馈的深度学习网络，实现报警在线分类识别，提高报警识别准确度及抗干扰能力；提出报警依赖溯源方法，引入扩展有限状态机(EFSM)，建立报警系统多层EFSM模型，探究迁移内和迁移间数据依赖及传递性，构建报警系统EFSM数据依赖图和依赖溯源模型，实现报警根源有效分析与定位；以TE过程和典型石化乙烯生产过程为应用研究对象，形成面向复杂工业过程报警识别分析与依赖溯源的新理论与新方法，研发报警系统在线识别与溯源分析软件原型，为保障过程安全生产提供有效的理论和技术支持。

准确、高效的报警系统对提高生产过程安全性和可操作性至关重要。针对复杂工业过程报 警泛滥、难以辨别等问题，本课题提出一种报警阈值非线性区间估计方法，通过建立阈值上下 限估计模型，构建阈值区间估计质量综合评价指标，实现报警阈值优化设置；提出报警特征重构方法，放大报警特征差异，并引入信息反馈机制，构建基于动态信息记忆反馈的深度学习网络，实现报警在线分类识别，提高报警识别准确度及抗干扰能力；提出报警依赖溯源方法，引入扩展有限状态机(EFSM)，建立报警系统多层EFSM模型，探究迁移内和迁移间数据依赖及传递 性，构建报警系统EFSM数据依赖图和依赖溯源模型，实现报警根源有效分析与定位；以TE过程 和典型石化乙烯生产过程为应用研究对象，形成面向复杂工业过程报警识别分析与依赖溯源的新理论与新方法，研发报警系统在线识别与溯源分析软件原型，为保障过程安全生产提供有效的理论和技术支持。