Abstract: As different kinds of sensitive loads connected to the power system, the impact of voltage sag has become more serious. The damage or outage of equipment caused by voltage sag is often accompanied by huge economic losses. In addition to a series of post-analysis, people also hope to reduce losses through pre-prevention. Due to the continuous expansion of the power monitoring system, a large number of historical monitoring data have been accumulated. Mining the hidden information makes it possible to find the laws between different characteristics of the voltage sag. This paper proposes an analytical method of data mining on voltage sag based on gray target theory and cloud theory, to obtain the knowledge which reflect voltage sag severity of site in different situations. For each voltage sag event, a large amount of real-time monitoring data can be collected through the monitoring station, and the information of voltage sag contains various kinds of contents, such as the cause of the fault, can be obtained through the post-report. Each voltage sag event record contains multiple types of content, which together form a description of a fault scenario. By mining the association rules of the historical database, each site in the power grid is used as the description object to find the relationship between the different factors in the fault scenario and the voltage sag severity of site. Because the power system is characterized by complexity, uncertainty and fuzziness, it contains a lot of qualitative and quantitative knowledge that can be used to describe the system. In order to solve the problem that the traditional association rule mining algorithm can’t deal with the continuous data, this paper firstly combines the cloud transformation algorithm to divide the definition domain of the quantitative data into the corresponding cloud qualitative concept, then uses the AprioriTid algorithm to mine the strong association rules. In the actual scene that users are concerned with, the characteristics associated with the voltage sag severity of site may be partially or completely the same as the fault scenarios described by multiple strong association rules. In order to find the association rules that can reflect the voltage sag severity of site in the actual scene, this paper combines the grey target theory to establish the matching model between the fault scenarios described by the strong association rules and the actual scenario, to filter different strong association rules in the actual scenario. Finally it can obtain the correlation knowledge that reflects voltage sag severity of site in the current scenario. The history power quality monitoring records of an electric power company are used as the data source, and combine with the fault location and the type of user and so on obtained from the event report, to build a database for mining association rules. The results show that the proposed method can effectively reflect the voltage sag severity in different scenarios, and the knowledge obtained will help the power departments to make appropriate governance decisions.