Alarm Correlation Analysis Methods Based On Alarm Timesequences Mining

The alarm system is an important part to ensure the safety of modern process industry.However,the widespread occurrence of alarms in the industrial process has a great impact on the performance of the alarm system.The alarm correlation analysis is one of effective ways to find out the root cause of alarms as well as to suppress the alarm flooding.The alarm correlation analysis needs to calculate and process a large amount of alarm data.However,the current methods always suffer limitations to deal with such time sequence data.In addition,the estimation and representation of the time lag between the alarm sequences is far from accurate.In response these limitations,this thesis presents an alarm similarity analysis method based on alarm time sequences mining.The main contents are organized as follows.Firstly,the calculation of the similarity between alarm time series is presented in this thesis.In order to improve the computational efficiency,a novel block-based data mining approach is suggested to analyze industrial alarm data.By converting alarm data sequences into the time node sequences,both the time complexity and the space complexity of the correlation analysis are reduced.Secondly,to solve the problem that the time lag estimation is not accurate enough in the current method,a variable time window based on blocked sequence is proposed,which includes a complete alarm cycle time lag for each time window.Through the statistical analysis,the time lag field between the alarm sequences is generated,and the time lag relationship between the alarms is more accurately represented.Finally,the TE process is used to verify this method compared with the traditional one,showing that the partitioning method can improve the computing efficiency.Then statistical analysis of the time lag domain is performed to represent the time lag between alarm sequences.