Research On The Multidimensional Hierarchical Model For Complex Process Industry Alarm And Its Application

An alarm system is the collection of hardware and software that detects an alarm state, communicates the indication of that state to oper-ators, and records changes in the alarm state. Alarm systems play criti-cally fundamental roles for the safe and efficient operation of modern process industries. Its performance is directly related to process safety,product quality, production costs and even casualties. However, there is a widespread problem namely "alarm overloading" in the current alarm systems around the world, which seriously affects the proper protection function of the alarm systems, and thus acts as a great threat to the eco-nomic benefits, process safety and even personal safety. Concequently, it is urgent to pursuit scientific and reliable theory and technology to solve this problem. In our study, four main causes are indentified as the perpe-trators of alarm overloading. That is, abnormality spread due to compli-cated relationships, unreasonable design of alarm limits, unclear divi-sion of alarm priority, and nuisance alarms. Then, the research status covering domestic academia, international academia and industry are summarized by focusing on the current studies related to these four main causes. The lacks and improvements are analyzed, and a new systematic framework is proposed to provide theory and technology for solving alarm overloading.The modern process industry is characterized by nonlinearity,strong coupling, strong correlation and wide distribution. In order to solve the problem of alarm overloading in process industry, this paper studies the establishment of multi-dimensional and hierarchical model for complex process industry alarm based on the complex nature of the process. Taking the hierarchical model as the basis, a series fundamental study, namely, alarm threshold optimization, alarm root cause diagnosis,and alarm priority division are performed, constituting a systematic framework to solve the alarm overloading problem. The key benefits of our proposed framework have been demonstrated through industrial examples. The achivements of our study are listed as follows:(1) Aiming at the inherent complexity of process industry, multi-dimensional hierarchical models of process industry alarm based on da-ta-driven and knowledge fusion are established. Firstly, the multidimen-sional hierarchical model of process industy alarm based on partial cor-relation analysis and shallow knowledge fusion is studied. The visuali-zation of process topology is realized in four dimensions: macroscopic time dimension, microscopic time dimension, macroscopic spatial di-mension and microcosmic spatial dimension. Then, to realize the multi-variate analysis, a forward sparse pincipal component analysis algorithm is proposed to study multivariate hierarchical sparse model. Some re-dundant paths can be removed. The application results and comparison results of Tennessee Eastman (TE) process support the effectiveness of these two model.(2) Aiming at the unreasonable design of alarm threshold, an alarm threshold design strategy based on multidimensional hierarchical rele-vancy analysis is proposed. In order to make the alarm threshold design fully reflect the inherent relevant characteristics of the process, firstly,based on the model, the multi-dimensional hierarchical correlation analysis algorithm for both process variable and alarm variables are proposed. Then, the correlation consistency between process variables and alarm variables is regared as the objective function. The particle swarm optimization algorithm is used to optimize the threshold.Through the TE application we can see that using this strategy, the op-timal multivariable alarm threshold under different abnormal conditions are configured, leading to a great reduce of the false negative rate. Be-sides, the order of abnormal propagation is fully reflected.(3) Aiming at the problem that a single centralized diagnosis and control strategy can not be applied to large, non-linear and uncertain complex processes, a distributed alarm monitoring and alarm root-cause analysis method based on extreme learning machine (ELM) and data fu-sion is proposed. Still based on the model, to realized the distributed monitoring, multiple diagnostic modules are divided, and multiple di-agnostic models are built using ELM. Then, a BPA-IAHP data fusion algorithm is proposed to realize the effective fusion of multiple diagnos-tic module and find the real alarm root-cause. Finally, the method is ap-plied to TE process. The results show that the proposed method has strong robustness, high diagnostic accuracy and short diagnostic delay.(4) Aiming at the problem of unreasonable alarm priority division,an alarm priority partitioning method based on improved Likert scale is proposed. The method is objective and universal. Based on the multidi-mensional alarm hierarchy model, the improved Likert scale method is used to score the key performance index (KPI) of the alarm variables on the alarm propagation path and sort the alarm variables based on the fi-nal score, the higher the ranking, the higher the priority of the variable.The method is applied to TE process, and the result shows that the method can realize the priority division on the basis of guaranteeing the process safety and reduce the blindness in dealing with the alarms.