Abstract: Wheel-rail adhesion model serves as a key parameter for vehicle dynamics analysis and traction/braking control, directly determining the calculation precision of creep force in the wheel-rail interface. Although some adhesion models have been developed for adhesion behavior at large creepages, their validation is limited to speeds below 200 km/h, leaving high-speed applicability uncertain. This work first conducted adhesion tests using a full-scale wheel-rail rolling test rig under water and antifreeze lubrication conditions with speeds up to 450 km/h and creepages up to 30%. Dynamic creep curves during the increase and decrease of the creepage were measured and analyzed, followed by a comprehensive evaluation of several existing modeling methods for wheel-rail adhesion behavior at large creepages based on Polach adhesion theory. Some improvements to the modeling method were made, and wheel-rail adhesion models suitable for high speeds and large creepages were finally constructed. This model was expected to improve the calculation accuracy of wheel-rail creep forces in vehicle dynamics analysis and traction/braking control. Test results revealed significant dynamic phenomena, including the phenomena of adhesion recovery, hysteresis, and fluctuation, where the creepage corresponding to the limit of adhesion recovery (the second peak) was within 10%~15% and 15%~20%, respectively. This finding suggested that the maximum creepage should be controlled within this range during traction/braking control to achieve maximum utilization of wheel-rail adhesion. Under water lubrication conditions, increasing the creep rate was beneficial to adhesion recovery, and the degree of adhesion recovery increased significantly with increasing speed. Under antifreeze lubrication conditions, the degree of adhesion recovery was better than that under water lubrication conditions at speeds less than 300 km/h, while the adhesion recovery phenomenon could be neglected at speeds equal to 400 and 450 km/h. Regarding the modeling of adhesion behavior, the traditional Polach model failed to capture the above dynamic phenomena. After introducing instantaneous parameters such as specific dissipated energy, it was possible to effectively simulate the phenomenon of adhesion recovery, and partially simulate the phenomena of adhesion hysteresis and fluctuation. When the short-term cumulative dissipated energy was introduced, the existing model lost most of its fluctuation characteristics and leading to discontinuous mutations in the creep curve. Consequently, an improved modeling method was proposed by optimizing the relationship function between the static friction coefficient and the dissipated energy and simplifying the model structure. The model constructed in this way achieved better simulation for the phenomena of adhesion recovery and hysteresis. Moreover, the root-mean-square error between the simulated curves and the measured dynamic creep curves was reduced by nearly 25% compared with that before improvement. Notably, given the inherent discrepancies between laboratory settings and real-world operational environments, as well as the model’s remaining limitations in fully capturing dynamic fluctuation, caution is advised during the practical application of the model.