Abstract: In order to provide a basis for determining the anti-floating fortification water level of buildings within the design reference period T, this study takes the confined water level as the research object, based on long-term groundwater level observation records of a residential project near the Beijiang Embankment and the river hydrological data, the empirical distribution method and the maximum value distribution method were employed to conduct probabilistic modeling of the observation records. The density function of the maximum confined water level within the reference period T was extrapolated to predict the long-term maximum confined water level. The results indicate that when the probability model is related to the confined water level results of 100-year return period obtained from the inversion method/numerical method, the standard values of the confined water level of 50-year return period are 10.32 m (using the empirical distribution method) and 10.26 m (using the maximum value distribution method). When the probability model is established solely based on observation records, the standard value of the confined water level of 50-year return period is 9.96 m (using the maximum value distribution method). Based on the standard values and the analysis results of the observation records, a method for determining the anti-floating fortification water level of the residential project during the construction period and the service period was proposed. Additionally, load statistical parameters were provided for the calibration of the load partial coefficient and the combined value coefficient. By taking probabilistic load modeling as a breakthrough, we solve key problems in anti-floating design and providing a basis for the application and optimization of reliability analysis. The technical method demonstrates strong compatibility and operability.