Water level is closely related to the navigability of a waterway，and is particularly important for large ships(convoys)passing through certain typical sections.Water level is affected by many factors，and developing an accurate and reliable water level prediction model is a challenging problem.In this paper，a water level prediction model based on Transformer's multi-feature spatio-temporal fusion network is proposed，which is able to capture the complex spatio-temporal patterns and interactions of the water level data，study the correlation relationship between the water level and the different influencing factors，and generate the future water level prediction results based on the fused features.The research results are helpful to ensure the navigation safety of ships，give full play to the navigation capacity of waterways，and provide references for shipping management and planning.