The Method Of Multivariate Alarm Thresholds Design In Industry Process

Alarm system plays an important role in industrial process, and it can monitor the whole process. While the process is deviating from the normal state, the alarm system can give the corresponding alarm message. And the operator can take processing measures according to the alarm information, which can ensure the enterprise's safety. Alarm flooding is a common problem of alarm system, and an important influencing factor is the unreasonable of alarm threshold. Once the alarm flooding occurs, a large of the nuisance alarms may be produced. Because there are so many alarms, it may result in that the critical alarms are overwhelmed in nuisance alarms. Thus the operator cannot get the effective alarms in time, so they would not take the right response, which has become a serious threat to the safety of factory. Therefore, aiming at the alarm flooding, this paper has presented an improved design method for multivariable alarm thresholds design.Firstly, time delay estimation method was used to construct the topology relation between the process variables, thus we could determine the propagation path of alarm variables and then selected the optimized alarm variables. Secondly, to ensure both the enterprise's safety and correlation consistency, three factors-false alarm probability (FAP), missed alarm probability (MAP), and the correlation between the alarm information and process information-were combined to establish the objective function for the first time. Finally, the optimal thresholds were obtained by the genetic algorithm.The followings are the main study contents in the paper:(1) The selection of correlated alarm variables. For industrial process data, topology relation between variables was preliminary constructed by the correlation matrix, the directionality matrix and the causal matrix using time delay estimation method. And then the process knowledge was combined to modify this relationship, thus we could select the correlated alarm variables.(2) The alarm thresholds design method that combing the FAP, MAP and the consistency analysis was studied. Because that the security problem was ignored by the correlation analysis method in the process of threshold optimization. And the FAP and MAP are two important security performance index of alarm system. Hence, in our paper, we considered both the two index and the consistency analysis to establish the objective function of thresholds optimization, and the genetic algorithm was introduced to obtain the optimal thresholds.(3) In TE process, the 3cr method, the correlation analysis method, the FAP-MAP method and the improved method were used to set alarm thresholds, respectively. According to the results of false alarms and missed alarms, it is shown that the improved way has an advantage on reducing nuisance alarms, which provides a new way for alarm thresholds design.(4) Based on GUI in MATLAB, the user interface was designed for threshold optimization, and the following functions such as processing the process data, displaying the real-time data and optimizing the thresholds and so on, were implemented.