The invention discloses a power communication network alarm association mining method based on an improved GSP, and the method comprises the following steps: (1) collecting historical alarm data, building a historical alarm database, and carrying out the preprocessing of the historical alarm data; (2) generating a network topology model by analyzing the topological relation between the network element devices in the synchronous network topology database in the step (1); (3) performing data mining on the preprocessed historical alarm data in the step (1) by accessing the network topology modelassociation frequent sequence in the step (2) to form an alarm association rule and a sequence experience library; (4) collecting real-time alarm data, establishing a real-time alarm database, and preprocessing the real-time alarm data; and (5) performing data analysis on the preprocessed real-time alarm data in combination with an alarm association rule and a sequence experience library, and constructing an alarm real-time dependence model in combination with the network topology model in the step (2), thereby realizing positioning analysis of the root alarm.