一种改进的CSCA+PCA化工过程异常检测技术

为了解决化工过程异常检测时因参数众多且数据庞杂而导致一些异常无法被有效检出的问题,在Brownlee的克隆选择分类算法（CSCA）基础上,通过引入主成分分析（PCA）技术,进行数据降维和数据重整,探讨了人工免疫算法在化工过程异常检测中的适用效果和技术方案,以TE过程数据作为样本进行异常检测和分类实验。结果表明,过程异常数据的规模、属性的数目对CSCA异常检测效果具有明显影响,而通过主成分分析进行数据降维之后,CSCA检测效果有所提高;进一步的数据重整之后,CSCA对过程异常分类辨识的准确率可提升到85%以上;基于CSCA+PCA的数据降维及重构之后的过程异常检测技术方案,可以获得较高的异常检测准确率,从而一定程度上为化工过程安全运行提供技术保障。     

基于多点关联性的尾矿坝位移监测序列异常值诊断*

针对尾矿坝位移监测序列中噪声和真实异常值的区分问题,提出1种基于多点关联性和改进孤立森林（IF）算法的异常数据诊断模型。通过IF算法对监测序列中的各样本点异常程度进行量化计算,引入云模型（CM）算法确定IF量化的异常得分与异常概念的相互映射关系以实现异常点的初步诊断,根据Apriori算法计算多测点序列间的关联性,找出强关联序列组合,结合序列关联性以及异常点诊断结果区分噪声与真实异常值。以某尾矿坝位移监测序列为例进行模型验证。研究结果表明:基于多点关联性的异常诊断模型能够有效区分尾矿坝位移监测序列中的噪声与真实异常值,提高监测系统的准确性。     

基于时序片段的油气管道运行工况识别方法*

为准确识别管道系统运行工况,提高对油气管道突发事故的响应速度,综合提升管网安全管理水平,提出1种基于时序片段的油气管道运行工况识别方法。首先,构建基于概率分布的状态变化识别模型,提取油气管道中不同运行状态点;其次,建立基于时间序列片段的工况识别模型,快速识别不同时间长度内油气管道运行工况;最后,以国内某成品油管道为例进行方法验证。研究结果表明:该方法可有效识别成品油管道阀门开关状态、泵异常停机和阀门内漏3种运行工况。对比传统的识别方法,该方法可降低状态变化点的漏报率,提升管道运行工况识别的准确率。研究结果可为油气管道系统运行工况识别提供新的借鉴方法。     

基于传递熵与KELM的炼油化工过程风险传播路径分析方法

基于炼油化工过程复杂,设备众多,某一设备的监测变量发生扰动可能会传播至其相邻设备引发出一系列故障链。现有方法多是针对某一设备进行监测与诊断,以期降低事故后果,而忽视了对过程风险传播路径的预测以防止事故的发生。因此,提出一种基于传递熵与核极限学习机的炼油化工过程风险传播路径分析方法,该方法针对某一工艺扰动,分析其在风险发展过程中的扰动传播过程,基于传递熵分析法建立炼油化工过程风险传播推绎模型;并提出一种基于KELM的风险传播搜索方法,预测风险传播路径;将该方法应用于分馏塔冲塔过程。研究结果表明:该方法可辨识出未来一段时间内风险的可能传播路径,以便操作人员及时采取预防措施,保证过程安全及产品质量。     

时序数据分析的复杂化工过程异常智能溯源研究

对复杂化工过程异常工况进行智能推理溯源是实现安全关口前移、降低灾难性事故发生的有效途径。提出了一种基于Spearman-Apriori的化工过程异常智能溯源分析方法,旨在研究复杂化工过程异常工况发生的前置原因,并形成一种智能决策模型。针对化工工艺参数之间耦合性强、关联关系分析难度大的特点,引入Spearman相关系数,通过Spearman实时在线分析过程参数间的相关关系,并设置强关联阈值将Spearman相关系数分析与Apriori算法进行关联耦合,利用Apriori算法中的支持度和置信度二维挖掘各参数之间的超强关联规则。将该方法应用于合成氨工艺中合成工段的异常工况智能推溯,并选取氢氮比、管路工艺气流量、给水换热器冷凝剂流量等8个关键监测指标,研究发现氢氮比增大和给水换热器冷凝剂流量升高分别是导致合成塔入口压力超压、合成塔第一床层温度过低两组异常工况的前置原因,该分析结果与实际生产工艺相符,证明该方法可以有效地对化工过程异常原因进行推溯并筛选主要影响因素。研究为使用生产过程大数据实现化工过程异常智能溯源提供了理论基础,为进一步完善过程风险精细化管控提供了新思路。     

快速路交通流异常数据判断算法研究及实证

对快速路交通流数据进行异常数据判断,有利于避免使用异常数据带来的损失,提高信息利用的有效性。笔者分别根据逻辑推理、交通流的重复性和连续性以及交通流变量之间的机理分析提出了3种判断快速路交通流异常数据的算法,并讨论了这3种算法之间的集成。利用北京快速路实测数据对算法进行了验证,验证结果表明该算法基本是有效的。     

充填管道系统失效模式与影响分析(FMEA)及失效影响模糊评估

根据金川矿区充填管道系统的运行情况,结合充填实践经验,建立了充填管道系统的失效模式与影响分析法(FMEA)分析表,得出系统存在10种失效模式。针对FMEA分析表中失效影响存在较大模糊性而难以有效估计的特点,引入模糊评估方法来进行分析。通过该矿区充填管道系统各失效模式产生失效影响的模糊评估,得到不同失效模式对系统可靠性影响的排序结果。结果表明:该方法的评估结果与实际情况比较吻合;可以使分析定量化,有助于工程人员充分利用系统模糊信息;该方法也能用于其他系统失效模式影响的模糊评估。     

化工系统设备运行可靠性的综合评价模型研究

为提高化工过程系统设备运行的可靠度,讨论可靠性评价在化工过程系统管理中的适用性和发展方向,建立可靠性评价模型体系。将设备故障模式与可靠性评价相结合,分析存在的故障模式类型。用安全、环境以及经济分析方法,得到评价指标。运用无因次函数模型,建立化工过程系统设备运行可靠性的综合评价模型。利用某大型磷化工集团硫酸生产系统的实际运行数据,验证该模型的有效性。结果表明,当磷酸装置运行可靠性的综合评价值大于3.17时,需要对相应设备进行维修维护。     

基于模糊层次分析法的木结构古建筑安全评价

为准确地对木结构古建筑进行安全现状评价,考虑到木结构古建筑的结构复杂性,本文提出一种基于模糊层次分析法的木结构古建筑评价方法。利用该方法对故宫博物院景福宫进行检测和评价。结果表明:利用层次分析法对木结构古建筑进行安全性评价简单可行,与传统的评级法相比,该方法可考虑不同构件对结构安全性的贡献程度以及不同属性对同一构件安全性的贡献程度,提高了木结构古建筑安全评价的准确性和可靠性。     

蒸汽管网疏水系统安全监测预警技术研究

为了预警蒸汽管网疏水系统异常运行状态，通过研究蒸汽管网疏水系统温度及压力变化规律，构建了疏水系统阀门故障判断模型及疏水系统疏水异常判断模型，并基于判断模型提出了一种蒸汽管网疏水系统监测预警方法。该方法利用网关及3G/4G通信模块将传感器采集的温度及压力数据上传至判断模型，从而自动预警蒸汽管网疏水系统运行过程中的安全风险。目前该方法已被应用到合肥市161.5 km的蒸汽管网，应用结果表明，蒸汽管网疏水监测预警方法可以用于蒸汽管网疏水系统异常运行状态监测，并在一定程度上降低了蒸汽管网疏水系统泄漏带来的安全隐患。     

Kullback-Leibler distance-based enhanced detection of incipient anomalies

Accurate and effective anomaly detection and diagnosis of modern engineering systems by monitoring processes ensure reliability and safety of a product while maintaining desired quality. In this paper, an innovative method based on Kullback-Leibler divergence for detecting incipient anomalies in highly correlated multivariate data is presented. We use a partial least square (PLS) method as a modeling framework and a symmetrized Kullback-Leibler distance (KLD) as an anomaly indicator, where it is used to quantify the dissimilarity between current PLS-based residual and reference probability distributions obtained using fault-free data. Furthermore, this paper reports the development of two monitoring charts based on the KLD. The first approach is a KLD-Shewhart chart, where the Shewhart monitoring chart with a three sigma rule is used to monitor the KLD of the response variables residuals from the PLS model. The second approach integrates the KLD statistic into the exponentially weighted moving average monitoring chart. The performance of the PLS-based KLD anomaly-detection methods is illustrated and compared to that of conventional PLS-based anomaly detection methods. Using synthetic data and simulated distillation column data, we demonstrate the greater sensitivity and effectiveness of the developed method over the conventional PLS-based methods, especially when data are highly correlated and small anomalies are of interest. Results indicate that the proposed chart is a very promising KLD-based method because KLD-based charts are, in practice, designed to detect small shifts in process parameters.     

Extension of TACOM to the complexity of tasks designed for abnormal situations in nuclear power plants

TACOM (TAsk COMplexity) is a measure for evaluating the complexity of tasks prescribed for emergency situations in nuclear power plants. Five sub-measures constituting TACOM represent five different aspects of the task complexity exhibited in operating procedures for emergency situations. The practicality of TACOM has been verified through a series of empirical studies. However, tasks designed for abnormal situations that can significantly affect the safety of nuclear power plants, also need a proper measure for evaluating their complexity. TACOM provides a process, a systematic cognitive task analysis method and a set of guidelines to support its application. Therefore, although the characteristics of abnormal task situations are not the same as those of emergency situations, TACOM seems to be reasonably applied to abnormal situations or at least to offer meaningful insights for developing a measure for evaluating the complexity of tasks in abnormal situations. Thus this study examined the applicability of TACOM to abnormal situations through case studies. Particular attention was paid to the sufficiency and appropriateness of the three methodological tools, which are the process, the cognitive task analysis method and the set of guidelines. Collective consideration of the case studies and the characteristics of tasks prescribed for abnormal situations led us to draw the conclusion that TACOM could be reliably used for abnormal situations as well. This paper reports the process of how to apply TACOM to the tasks of abnormal situations and discusses some lessons learned through this application.     

一种新的基于图模型的间歇过程故障监测方法

目前多元统计方法被广泛用于间歇过程故障监测并已经取得了比较好的效果。但是,统计模型的可解释性能比较差,很难直接利用操作人员积累的安全经验。为应对这些不足,提出了一种基于图模型的间歇过程故障监测方法。利用提出的定性建模方法,过程机理及操作人员的安全经验能够方便地表达。利用在线的过程变量数据和正向推理算法推断生产应处于的状态,利用安全知识及时地发现生产异常。当推断的结论与在线测得的结果矛盾或过程变量超过设定的安全限时,给出解释性输出。通过一个间歇反应案例,验证了提出的方法在模型的可解释性和利用安全经验方面的优势。     

Process safety data management program based on HAZOP analysis and its application to an ethylene oxide/ethylene glycol plant

HAZOP analysis is a process hazard analysis method that has been widely applied both within and outside the chemical processing industries. This paper presents a design method for a process safety data management program for petrochemical plants based on HAZOP analysis and demonstrates the steps of application involved in building a process safety data management system for an ethylene oxide/ethylene glycol production plant. Firstly, the production data files and relevant documents of the plants should be classified and stored in the program database as reference documents and treatment schemes for coping with abnormal situations should be collected and summarized as guidance documents. Secondly, the HAZOP analysis method is employed to identify all the dangerous deviations possibly existing in the production process of the ethylene oxide/ethylene glycol plant. Then, the relationships among the deviations, the reference documents and the guidance documents should be considered and evaluated. Finally, each dangerous deviation will be given a corresponding reference document and guidance document. The reference documents and guidance documents stored in the expert system can be utilized to help operators solve the corresponding technical problems and cope with abnormal situations. The process safety data management program will contribute to the identification, analysis and resolution of operation problems. When an abnormal situation occurs, according to the deviations exhibited in the system, the necessary reference documents and guidance documents will be quickly consulted by the operators, and an appropriate decision will be made to address the abnormal situation. Therefore, by using the process safety data management program, plant security and human safety in the petrochemical industries will be improved.     

Misoperation monitoring and early warning during startup and shutdown of petrochemical units

Most petrochemical units run under extreme conditions, such as high temperatures, pressures, and speeds. Consequently, the equipment operators may commit errors because the startup and shutdown processes usually involve complicated operation steps; moreover, the operators may lack experience in handling abnormal situations. Misoperation can lead to accidents, including fires and explosions. Thus, risk analysis for process operations and the development of preventive measures have become an effective means of avoiding misoperation-related accidents. However, it is challenging to ensure the comprehensiveness of risk-analysis results. In this paper, we present a method for misoperation monitoring and early warning in the startup and shutdown processes of petrochemical units. The mechanisms of misoperation occurrence are summarized based on investigations of serious accidents in the recent past. Knowledge regarding the mechanisms of misoperation is crucial for the risk analysis of petrochemical units. The potential risk information, such as causes, adverse consequences, key monitoring parameters, and prevention control solutions, should be acquired and be employed to construct an early-warning knowledge database. Furthermore, misoperation judgment rules need to be formulated to identify misoperations. The data obtained from the monitoring module, misoperation judgment rules, and analysis results can aid in developing schemes to avoid possible abnormal situations. This paper reports a misoperation monitoring and early-warning system for a hydrogenation unit. As demonstrated, conducting risk analysis to determine the potential operational risks and formulating misoperation judgment rules to analyze the process data are essential for enabling early warning. The application of this method will contribute to operational guidance, economic loss reduction, and accident avoidance.     

Fault propagation behavior study and root cause reasoning with dynamic Bayesian network based framework

The Bhopal disaster was a gas leak incident in India, considered the world's worst industrial disaster happened around process facilities. Nowadays the process facilities in petrochemical industries have becoming increasingly large and automatic. There are many risk factors with complex relationships among them. Unfortunately, some operators have poor access to abnormal situation management experience due to the lack of knowledge. However these interdependencies are seldom accounted for in current risk and safety analyses, which also belonged to the main factor causing Bhopal tragedy. Fault propagation behavior of process system is studied in this paper, and a dynamic Bayesian network based framework for root cause reasoning is proposed to deal with abnormal situation. It will help operators to fully understand the relationships among all the risk factors, identify the causes that lead to the abnormal situations, and consider all available safety measures to cope with the situation. Examples from a case study for process facilities are included to illustrate the effectiveness of the proposed approach. It also provides a method to help us do things better in the future and to make sure that another such terrible accident never happens again.     

Risk assessment of chemical process considering dynamic probability of near misses based on Bayesian theory and event tree analysis

With the development of modern automatic control systems, chemical accidents are of low frequency in most chemical plants, but once an accident happens, it often causes serious consequences. Near-misses are the precursor of accidents. As the process progresses, near misses caused by abnormal fluctuation of process variables may eventually lead to accidents. However, variables that may lead to serious consequences in the production process cannot update the risk in the life cycle of the process by traditional risk assessment methods, which do not pay enough attention to the near misses. Therefore, this paper proposed a new method based on Bayesian theory to dynamically update the probability of key variables associated with process failure risk and obtain the risk change of the near-misses. This article outlines the proposed approach and uses a chemical process of styrene production to demonstrate the application. In this chemical process, the key variables include flow rate, liquid level, pressure and temperature. In order to study the dynamic risk of the chemical process with consideration of near misses, according to the accumulated data of process variables, firstly the abnormal probability of the variables and the failure rate of safety systems associated with the variables were updated with time based on Bayesian theory. On the basis of the dynamic probability of key process variables, an event tree of possible consequences caused by variable anomalies was established. From the logical relationship of the event tree, the probability of different consequences can be obtained. The results show that the proposed risk assessment method based on Bayesian theory can overcome the shortcomings of traditional analysis methods. It shows the dynamic characteristics of the probability of different near misses, and achieves the dynamic risk analysis of chemical process accidents.     

公共场所人群加速度异常检测系统

针对目前基于速度检测公共场所密集人群异常行为存在的检测准确率低、使用范围局限的问题,从人群的加速度角度对可能导致公共安全事故的人群异常行为进行研究,提出了一种基于加速度检测人群异常行为的算法,并基于该算法实现了针对人群逃散、人群聚集、人群拥挤和人群逆行4种异常行为检测的系统。首先,利用金字塔Lucas-Kanade光流法进行特征点跟踪;然后,在获取到特征点的速度矩阵基础上进一步计算其加速度矩阵,反映速度的整体变化;最后,从加速度大小和方向两方面检测人群异常行为。结果表明,所提算法检测用时较少,相比基于速度检测的对比算法,检测的正确率提高到80%,误报率降低为5%。