Alarm systems play important roles for the safe and efficient operation of modern industrial plants. Critical alarms are configured with a higher priority and are safety related among many other alarms. If critical alarms can be predicted in advance, the operator will have more time to prevent them from happening. In this paper, we present a dynamic alarm prediction algorithm, which is a probabilistic model that utilizes alarm data from distributed control system, to calculate the occurrence probability of critical alarms. It accounts for the local inter-dependences among the alarms using the n-gram model, which occur because of the nonlinear relationships between variables. Finally, the dynamic alarm prediction algorithm is applied to an industrial case study. (C) 2016 The Chemical Industry and Engineering Society of China, and Chemical Industry Press. All rights reserved.