Track geometry data is obtained by checking the measured track, however, the performance of historical data from different time is often accompanied by the existence of cumulative mileage errors due to changes in inspection environment and conditions, it will lead to a phenomenon of data that cannot be aligned, then it is impossible to predict the development of track irregularities accurately; It is proposed that the multiple sets of raw data should be verified in subsection at a certain step, cross correlation function is used to evaluate each other, the effective observations are obtained after each group's raw data is aligned; then, the historical data in Guangzhou Railway Group Huizhou Railway Section HangzhouShenzhen Line Chaoshan Railway Station No.4 Road K1317+150—K1317+350 between 2013—2015 as the test sample is used to predict the track irregularities by building the ARIMA model. The result shows: research on the prediction of track irregularity after the raw data of track geometry size has been aligned that can achieve higher test accuracy, the maximum relative error is less than 5%, the average relative error is 1.75% in the sample.