To enhance the expressway informational management, a model for the expressway travel time prediction problem based on exit toll data was proposed. The model combines support vector regression (SVR) and bayesian optimization algorithm (BOA), and incorporates mean filtering (MF) to denoise the dataset, while fully considering the influence of historical travel time, vehicle type, and other characteristic variables on travel time. Selected G22 Qing-Lan Expressway exit toll data for prediction experiments, the results show that the prediction model based on MF-BOA-SVR demonstrates better performance in terms of expressway travel time prediction compared to benchmark models such as back propagation algorithm (BP) and random forest (RF), and exhibits strong generalizability, on the dataset of the section from Yuchong toll station to Lanzhou East toll station, the mean absolute error is reduced by 54.80%, root mean square error by 56.11% and mean absolute percentage error by 53.94% on average.