Research On Data-driven Alarm Analysis Method For Process Industry

Industrial process alarm system has great means to help operators deal with process anomalies by the combination of software and hardware in industry,it is important for ensuring safe and efficient production.With the deep advancement of Industry 4.0 and "Made in China 2025",large-scale industrial processes have developed rapidly.As a result,higher requirements have been placed on the alarm systems.However,most of the traditional industrial process alarm systems are difficult to solve the "alarm floods"problem,in which the false alarm event is difficult to handle,the alarm information is too simple to analyze,and the alarm threshold does not work efficiently.All these factors bring security risks for industrial process.Based on the problems existing in the current alarm system,this paper proposes a data-driven industrial alarm analysis framework by the following three parts to ensure the safety of industrial processes,pre-alarm prevention,alarm response and post-alarm optimization.Due to the complex and wide distribution of variables in industrial process,it is difficult to design the alarm system efficiently.At the same time,the clear distribution of various units in the industrial process,rich process knowledge and convenient process data collection also lay a foundation for alarm analysis framwork.In this paper,the pre-alarm prevention,alarm response and post-alarm optimization are implemented in five parts:real-time key alarm variable analysis,rapid alarm root area localization after alarm,accurate alarm root analysis after alarm,threshold optimization after alarm and pre-optimization before alarm,respectively.Through these five researches,this paper builds an alarm analysis framework to solve the problem of "alarm floods".This paper also verifies the effectiveness of the proposed methods through industrial simulation examples.In these studies,the main results are shown as follows:(1)Multi-variables simultaneous alarm is difficult to analyze,this paper proposes a key alarm information analysis technology based on multi-correlation blocks partial least square method,and this paper proposes a corresponding real-time key alarm information analysis strategy.When an alarm occurs,this method can first give the main unit that affects the normal operation of the process,and then specify several main process variables that affect the normal operation of the process,which helps the operator to deal with a large number of alarms in time.(2)Aiming at the difficulty of determining the source of the alarm,this paper proposes an alarm propagation network based on the multi-blocks transfer entropy and a corresponding alarm root cause analysis method.The multi-blocks transfer entropy algorithm reduces the computational burden of building an alarm propagation network,improves the accuracy of alarm root cause analysis,and achieves the effect of quickly locating the alarm root area.This method can assist the operator to deal with the root cause of the alarm in time.(3)The accuracy of the alarm propagation network is easily affected by the similarity of the alarm data,this paper proposes an active dynamic alarm propagation network based on the active transfer entropy algorithm and a corresponding alarm root cause analysis method.The alarm propagation network has the ability to automatically eliminate false causal relationships,accurately provide causal relationships between alarm variables,and achieve the effect of accurately locating the root cause variables of alarms.It is an important breakthrough in dealing with the phenomenon of "alarm flooding".(4)Aiming at the problem of unreasonable threshold setting in the traditional alarm system.In this paper,two alarm threshold optimization methods are proposed,which are the threshold optimization method based on multi-correlation blocks based false alarm probability and missing alarm probability and the threshold pre-optimization method based on active transfer entropy based simplified multi-layers bayesian network.The threshold optimization method has achieved the goal of optimizing the multivariable thresholds of each unit at the same time,and has achieved a certain degree of improvement.The threshold pre-optimization method adaptively adjusts the alarm threshold before the alarm occurs,which improves the accuracy of the alarm.In addition,the threshold pre-optimization method has the ability to continuously optimize the performance of the alarm system,which can fundamentally guarantee the production safety of industrial processes.