复杂化工过程故障诊断及其根源分析关键技术研究

化工行业是国民经济的支柱产业之一，其规模和产量日益增加，但生产过程故障频发导致事故，造成了严重的人员伤亡和财产损失，因此，有效处理过程故障，保障化工企业安全、稳定生产刻不容缓。故障诊断技术是处理过程故障的有效途径。本项目针对流行学习小样本问题，研究提出基于“量”改进和“质”改进的流行学习新方法，提升其判别能力，更有效地提取过程故障特征信息；针对单一分类器诊断精度低的问题，提出基于Adaboost同质集成和异质集成的多故障诊断新方法；针对传统贝叶斯网络结构学习效率低和易于过拟合的问题，研究提出改进模块化贝叶斯网络新模型，借鉴“分布式”策略，并引入新型评分函数和惩罚项，避免过拟合，提高结构学习效率，形成更高效的模块化贝叶斯网络模型用于故障根源分析。最后，采用田纳西-伊斯曼仿真平台和工业应用案例，验证本项目所提方法的有效性与可行性，为保障复杂化工生产过程安全、稳定提供可靠的技术支持。

The chemical industry is one of the pillar industries of the national economy. The scale and the yields of chemical processes have been increasing. However, frequent process faults may lead to accidents, causing serious casualties and property losses. Therefore, it is urgent to effectively handle process faults for ensuring the safe and stable production of chemical companies. The fault diagnosis technology is an effective way to deal with process faults. In this project, to solve the small sample problem of manifold learning, improved manifold learning methods based on "quantity" improvement and "quality" improvement are studied and proposed; as a result, the discriminating ability is enhanced and the process fault feature information can be extracted more effectively. To improve the diagnosis accuracy of a single classifier, homogeneous and heterogeneous ensemble models based on Adaboost are proposed for multi-fault diagnosis. In traditional Bayesian networks, the efficiency of the structure learning is low and the learning process is easy to be over-fitting. To deal with these problems, a novel and improved modular Bayesian network is studied and proposed. By drawing on the distributed strategy and introducing new scoring functions and penalties, the learning efficiency is accelerated and the overfitting problem is avoided. Thus, a more efficient modular Bayesian network model can be established for root cause analysis. Finally, case studies based on the Tennessee-Eastman simulation platform and real-plant implementation are carried out to verify the effectiveness and feasibility of the proposed methods in this project. This project aims at providing reliable technical support for ensuring the safety and stability of complex chemical production processes.