Research And Application Of Alarm Threshold Optimization Strategy In Petrochemical Industry

The alarm system is an important part of the petrochemical industry process and an important barrier to ensure the safety of production.Through the alarm system prompts,the operator can grasp the running status of the equipment in real time and handle the process faults accurately and efficiently.With the development and extensive use of the advanced control systems,alarm points can be set easily.However,the situation of setting alarm points randomly causes a lot of false and redundant information in the generated alarm data.There are often correlations between variables in the petrochemical industry.If the alarm settings are unreasonable,when an abnormal situation occurs,it may cause an "alarm flooding" problem.A large number of invalid alarms cause the great trouble to the system operator.In severe cases,the operator cannot even handle the key alarms in time,which leads an accident.As an important part of alarm management design,whether the threshold is reasonable or not is related to whether the correct alarms can be generated at the best time,whether the false and missed alarms can be reduced effectively,and whether the process production safety can be guaranteed.Therefore,aiming at the problem of alarm flooding and unreasonable threshold design in petrochemical industry,this paper carries out the optimization design of alarm threshold,and presents an improved multivariable alarm threshold optimization design method.The method firstly divides the system variables based on the chemical process knowledge,secondly selects the target optimization sub-module by using the principal component analysis method,chooses target optimization variables by partial least squares and then establishes the association relationship between the target sub-modules by applying the Partial Least Squares method.Finally,the objective optimization function based on the false alarm probability,the missed alarm probability and the similarity factor of the projection matrix of the correlation module is constructed to obtain the optimal alarm threshold group.The main research work of this paper is as follows:(1)Selecting the target optimization sub-module and the target optimization variables of the sub-module and constructing the association correlation of the target optimization sub-modules.In this paper,the target optimization sub-modules are selected for the variable module with chemical continuous correlation characteristics through the principal component analysis method.The sub-module target optimization variables are chosen by partial least squares.The association correlation between the target optimization sub-modules is established by the partial least squares projection matrix.(2)The alarm threshold design method based on the false positive rate,the false negative rate and the similarity factor of the correlation module projection matrix.In this paper,two variables alarm optimization based on variable correlation analysis in the previous threshold optimization design process is discussed.The similarity factor of the associated module is introduced.The false alarm probability and Missed alarm probability of two important safety indicators in the alarm management are integrated.The objective optimization function based on the false alarm probability,the missed alarm probability and the similarity factor of the projection matrix of the correlation module is constructed.The optimal threshold group are obtained by combining the Gaussian non-parametric kernel function estimation method and the genetic algorithm.The goal of multivariable alarm threshold optimization is achieved.(3)Taking the TE simulation process model as an example,based on the three different fault types of the TE process,the alarm threshold optimization design method proposed in tihis paper is used to verify the example.The effectiveness of the proposed method is verified by comparison with the commonly used threshold design method.