Design Of Industrial Alarm Based On Evidence Reasoning

Industrial alarm system is an important part of modern industrial systems.Its role is to monitor process variables in the production process so as to determine whether the system is in an abnormal state or not,and then generate an alarm.The false alarm rate(FAR),the missed alarm rate(MAR),and the averaged delay time(AAD)are usually used as indicators for evaluating the performance of the alarm.The traditional alarm design method generally optimizes the threshold and other parameters for the purpose of minimizing false alarm rate and missed alarm rate under the premise that the statistical distribution of process variables is known,so as to achieve optimal design.But in practical applications,since the sample data of process variables will inevitably be interfered by many uncertain factors in the monitoring and production environment,it is difficult to construct an accurate probability distribution model to describe its random uncertainty,at this time,designing an alarm based on probability theory will have great limitations.Compared with the traditional probability theory,Dempster-Shafer(DS)evidence theory has advantages in the description,reasoning and comprehensive processing of uncertainty information.In the case that the exact probability model of the process variable is unknown,some scholars have applied the evidence combination and update rules in the theory to the alarm design,and achieved remarkable results.On this basis,this paper introduces evidence reasoning theory to give a class of univariate and multivariate alarm design methods to improve the performance of such alarms.The main work is as follows:(1)A method for designing a univariate alarm based on evidence reasoning.For the aleatory uncertainty and mixed(aleatory/epistemic)uncertainty of process variables,this paper construct the sample interval point table and the sample reference value point table respectively,and obtain the alarm evidence corresponding to the process variable at each moment;expert knowledge is used to determine the fusion factors such as the reliability factor and importance weight of the alarm evidence.The evidence reasoning rule is used to recursively combine the current alarm evidence and the historical alarm evidence to obtain the global alarm evidence,and then the alarm result is generated according to decision criteria.Two sets of simulation experiments show that the proposed method has better alarm effect than the traditional method.(2)A method for designing a univariate alarm based on evidence reasoning via multi-objective online optimization.In the research content(1),the relevant fusion factors are determined offline by expert knowledge,and their value has certain subjectivity.Therefore,the forgetting strategy and the target optimization model are introduced,such that the relationship between the new evidence and the historical data can be dynamically adapted to improve the performance of the alarm system.It also provides a more flexible alternative to the adjustment of relevant fusion factors in practical industrial applications.(3)A method for designing a multivariate alarm based on multi-source evidence reliability metrics.Based on the research contents(1)and(2),a multivariate alarm design method based on multi-source evidence reliability metric is proposed.Through the comprehensive conversion of each univariate alarm performance evaluation index,the reliability of their respective alarm evidences is determined,and the evidence reasoning rules are used for multi-source evidence fusion to obtain multivariate alarm results.The simulation experiment and the oil leakage example prove that the proposed method has better alarm effect than the univariate alarm.