In the process of alarm association mining, the average length of alarm ID is more than 60 characters, and the basic operation will occupy a lot of resources, which makes the mining task difficult to complete in the effective time. In this paper, the mining process is optimized for the follow problems. Firstly, the alarm data set is divided into subsets in the computer room. Secondly, the statistics of alarm-related matters in each subset are realized by merging. Finally, standardize the alarm ID according to the statistical characteristics of the alarm data set. Furthermore, the operation speed of the model of alarm association mining is greatly improved.