Millions of optical layer alarms may appear in optical transmission networks every month, which brings huge challenges to network operation, management and maintenance. Aiming at this problem, this paper proposes an optical network alarm compression method based on multi-dimensional attributes. First, the alarm transaction set is extracted according to the time when the alarm occurs and the network location. Then, the weight of the alarm attribute is obtained through the neural network to calculate the alarm similarity. Finally, the alarm information is compressed by alarm similarity. The actual network data verification results show that this method can effectively improve the alarm compression ratio by about 17%. By using association mining algorithm to mine association rules of compressed alarm set, the strong association alarm transaction increased by 55%, which is helpful to further analyze the correlation between alarm information and locate the fault source.