With the development of the Internet, there are frequent network attacks and crimes against university websites, and traffic attacks are one of the common network attacks. In order to analyze the distribution characteristics of the extreme value of website visits and set up the early warning line of website visits scientifically and reasonably, this paper builds a Generalized Pareto Distribution model with the visit data of logistics website of UESTC as the research object. Firstly, the optimal threshold is selected using KS test -Hill graph. Secondly, the model parameters are estimated based on maximum likelihood estimation, and the fitting effect of the model is analyzed. Finally, through the analysis of different return periods, the Generalized Pareto model is proved to have a certain extrapolation ability, and the scheme of setting the early warning line of website visits according to the return level corresponding to different return periods is proposed.