Alarm System Design Via Evidence Updating Filter And Belief Rule Inference

The complex modem industrial production process mainly uses alarm system to monitor the process variables.The alarm system can determine whether the production process is abnormal or normal by analyzing process variables,so as to ensure the safe,stable and efficient operation of the monitored plants.The false alarm rate(FAR)and missed alarm rate(MAR)are often used as indicator to measure accuracy performance of the alarm system,and the averaged delay time(AAD)is used as an indicator of sensitivity performance.In traditional alarm system,the design methods are usually assumed that the statistical distribution of process variables is known and optimize the threshold value of alarm system to realize the optimal design in the sense of probability by minimizing false alarm rate and missed alarm rate.However,in the actual production conditions,the uncertainty in the monitoring environment is changeable or even unknown,hence it is very hard to give the accurate distributions of process variables,so the traditional probability-based alarm design methods are no longer applicable.Dempster-shafer(DS)Evidence theory has great advantages in describing and reasoning information with the stochastic uncertainty and cognitive uncertainty.Some scholars have designed an industrial alarm based on evidence updating filter method under the condition that the probabilistic characteristics of process variables are not completely known or even unknown.On this basis,this paper introduces the belief rule inference and the multi-order form of the evidence updating filter to improve the performance of this kind of design methods,the main work is as follows:(1)The design of alarm system based on multi-order evidence updating filter.There is a problem of missing information in the process of transforming process variable into alarm evidence.The continuous fuzzy membership function is used to replace the segmented fuzzy membership function to realize such transformation,which effectively reduces the missing of process variable information.In order to integrate more alarm information and obtain more accurate alarm results,the second-order evidence update method is extended to the multi-level form,so that the importance of historical information is emphasized in the fusion,and the function of the current evidence information is refined.Finally,the more reliable alarm evidence and decision result are obtained.The superiority of the method is illustrated by comparative experiments.(2)The design of multi-order evidence updating filter alarm based on belief rule inference.In the second-order evidence updating alarm design method,the fusion weights are designed by the linear relation between the incoming and historical alarm evidence,but this linear model can hardly completely describe the complex relationship between alarm evidence.Therefore,the belief rule inference method is introduced to construct a nonlinear model between the alarm evidence support(input)and the fusion weight(output).The belief rule based expert system describes the nonlinear relationship between input and output more accurately by optimizing the relevant parameters.The superiority of the proposed method is demonstrated by comparing the experimental design with the multi-order evidence updating alarm design method given in(1)and other traditional methods.