Abstract: Accurate prediction of the number of passengers sent at large train stations is one of the main basis for the compilation of train operation plan and train timetables to complete the task of passenger trasportation based on passenger travel demand. Firstly, this paper gives a brief introduction to the theory and principle of support vector regression. Taking the actual daily passenger flow of Hankou Station from January to December in 2017 as the sample data set, the characteristics of passenger flow of the large railway station are analyzed. And the analysis shows that the annual passenger flow fluctuates in obvious cycles and the passenger flow spikes suddenly due to several holidays in a long period. The sample data set is then divided into one training set and one test set and the numbers of passengers daily sent at the station before and after the elimination of holidays are respectively predicted by using the support vector regression model and the comparison of the errors of the predication results indicates that the accuracy of the number of passengers sent at a station derived by using this model can be enhanced subtantially after eliminating the impact of sudden spikes of holidays' passenger flow.