ABSTRACT： Large amounts of data for voltage sag events have been accumulated in on-line power quality monitoring. Massive data contains the relationship among the items, which can be used to predict the law of events according to association rules. In this paper, a method is designed to convert feature dimension data in the database of voltage sag events into one-dimensional array. Through a single scan executed on the database, the mode mining of multi-dimensional frequent patterns based on that array greatly improves the computation efficiency. According to the generated rule base, integrated with knowledge reasoning technique, calculating the similarity between the predicted data and the regular data, voltage sag prediction is realized. The proposed method is suitable for event data mining and prediction.