The use of a high-precision a priori zenith wet delay can effectively improve the convergence time and increase the accuracy of precise point positioning (PPP). Considering the limitations of empirical and blind ZWD models, a new lightweight ZWD model was developed based on GNSS observations and ERA5 precipitable water vapor (PWV) and temperature information that can provide reliable ZWD with 0.25° spatial resolution and hourly temporal resolution without relying on massive meteorological parameters. The correlation between the ERA5 PWV/temperature and the GNSS-derived ZWD was first analyzed in 2020 in this study. Subsequently, a functional mapping model based on machine learning was established, considering the effects of spatial and temporal factors, such as time, topography, geographical location. The validation results showed the proposed model's RMSE/STD/Bias were 1.70/1.70/0cm compared with GNSS ZWD, respectively. Furthermore, an independent GNSS and radiosonde ZWD in 2021, was introduced to validate the generalization ability of the proposed model, and the RMSE/STD/Bias values were 1.63/1.49/0.2cm and 2.77/2.63/0.23cm, respectively. The ZWD model developed in this study effectively reduces the dependence on meteorological data from numerical weather models, which is beneficial for broadcasting ZWD data and enhances precise regional positioning and navigation.