Hydrology prediction is a complexnon-linear dynamic process and the station water-level often shows dynamic changing character owing to all kinds of factors.In the Huaihe Basin Wuhe station water-level will be influenced by the backwater influence of Hongze lake and shows the non-statinoary changing.In the paper based on the neural-network model of time series and the data characteristics of hydrology a non-stationary multi-station variable dynamic sequence prediction model is made by using artificial neural-network and practised in Wuhe station water-level prediction of Huaihe River.The calculation results indicates that the model is not only reasonable but also its predicting period is longer.It is valuable when being used in practices.