In the context of high traffic volume and passenger traffic, the track line is the basic facility for subway operation, and its irregularity directly affects the safety and comfort of driving. Take Beijing Subway Line nine as an example, in this paper, the fiber grating sensor was used to monitor the track irregularity of a typical section. The statistics and the over-limit analysis of the monitoring data were performed by mathematical statistics. A combined prediction method combining gray model and particle swarm optimization for Elman neural network was proposed by Combining with the data characteristics. The effects of the model were verified by the seven parameters of track irregularity monitored. The results show that the prediction effect of the model is better than that of the single gray model. It has high prediction accuracy in terms of average relative error, root mean square error, determination coefficient and correlation coefficient. It can achieve over-limit warning for each individual indicator, which can better monitor the development of track quality.