Dimension Compression And Clustering Analysis Based Alarms Thresholds Optimization In Chemical Processes

In modern chemical industry,it is necessary to conduct alarm threshold settings for some process variables via alarm system to enhance the security and stability of production process.Improper threshold settings will generate too many invalid alarms,increase operation load,and even trigger serious accidents and invalid alarm system finally.Therefore,it is requisite to do optimization research for variable alarm threshold.Study on the method of multivariate alarm threshold optimization was conducted based on Tennessee Eastman(TE)process in this paper,and its validity was verified by the case of an industrial atmospheric-vacuum crude distillation operation.Two aspects about multivariate alarm thresholds optimization method based on the TE process are conducted.On the one hand,the method of principal component analysis(PCA)is used to calculate variable weight.Then Johnson normal transformation is used to transform variable data.False alarm rate(FAR)and missed alarm rate(MAR)are determined via probability density estimation,and the conditions of reduced FAR and appropriate alarm frequency less than the limited frequency specified in the international standards(normally one alarm per minute)are used to optimize alarm thresholds.The PCA weight and Johnson transformation based alarm threshold optimization method in chemical processes is finally proposed.On the other hand,the standardized Euclidean distance is used to realize alarm clustering,and the entropy method is utilized to calculate variable weight.Then probability density functions of variables are fitted in normal and abnormal conditions respectively.Objective function about FAR,MAR and average alarm delay(AAD)is established and optimized by the method of ant colony optimization(ACO).The alarm clustering analysis and ACO based multivariable alarms thresholds optimization method in chemical processes is finally proposed.Both of them realize the purpose of optimizing thresholds to a certain extent.Case studies of TE process and an industrial atmospheric-vacuum crude distillation operation show that the proposed methods can effectively decrease the number of alarms and alarm rates compared with traditional methods.So our method has significant advantages in terms of alarm threshold optimization.