Nuisance alarms are recognized as the major problem in the process industry. However

current alarm optimization methods usually suffer difficulties such as low sensitivity and poor reliability. In this paper

based on alarm interval and alarm duration of alarm data

the types of nuisance alarms are determined before an adaptive method to effectively determine the delay timers and alarm deadbands. An ARMA model is established to forecast adaptive alarm deadbands which are employed to deal with high frequency alarms

while the alarm delays updated by adaptively adjusting alarm time intervals are used to deal with low frequency alarms. Experiments with industrial process data are carried out

illustrating that the proposed method can reduce the numbers of false and missed alarms as well as improve the performance of alarm systems.