Optimization Design Method Of Industrial Alarm System Based On Conditional Evidence Linear Combination Updating Rule

In industrial processes,the changes of main process variables can reflect the operation condition of the monitored equipment.The Alarm system acts on processing the sampling signals of the process variables by comparing them with alarm threshold so as to monitor the abnormal state of the equipment.In the design of alarm,many scholars have generally taken the false alarm rate(FAR),missed alarm rate(MAR)and averaged delay time(AAD)as the measure of performance indices.The traditional design methods of alarm system are usually based on the assumption that the statistical distribution of process variable is known,and then use the former two indices to optimize the alarm threshold and other parameters.Because of the impacts of all sorts of uncertainties in the actual operation and condition monitoring of equipment,it is difficult to obtain the statistical distribution of process variable.However,Dempster-Shafer evidence theory has itself advantage over the probability theory in representation,reasoning and comprehensive processing of uncertainty information.The information fusion theory has been introduced into the alarm system.At the same time,the optimization as well as design of the alarm system via the update and fusion of alarm evidence also is given.As a result,some preliminary research results are obtained.This thesis mainly aims at an in-depth study on alarm evidence generation,performance indices about evidential alarm system and the optimization of alarm evidence parameters so as to improve application of evidence theory in industrial alarm design,the main works are as follows:(1)Alarm evidence generation based on Sigmoid function.In the traditional transformation from the process variable to the corresponding alarm evidence,the usage of the “subsection” trapezoidal fuzzy membership function leads to the information loss contained in the process variable.Aiming at this problem,we propose an alarm evidence generation method based on the “continuous” Sigmoid function,and the theoretical proofs and the simulation experiments show that the new transformation technique never causes the information loss.(2)Optimization method of alarm evidence based on static convergence index.We define the static convergence index of the probability assignments in alarm evidence based on the Jousselme distance of evidence,and further analyze the relationship between the static convergence index and the Sigmoid function's parameters.Furthermore,the corresponding relationships among the alarm threshold,FAR/MAR and the static convergence index are researched.On this basis,the contextual discounting vector on alarm evidence is introduced and the objective function based on the static convergence index is designed to improve reliability of alarm evidence by optimizing the discount vector of the current alarm evidence and adjusting the Sigmoid function's parameters.(3)Conditional alarm evidence linear combination updating method based on the dynamic convergence index.Firstly,the definition of dynamic convergence index is given.Secondly,on the basis of the above optimization of static convergence index,the alarm evidence updating and parameter optimization method based on dynamic convergence index is designed.Finally,the superiority of the proposed method is illustrated by comparing with the traditional alarm design method and the traditional evidence update alarm optimization design methods.