To clarify the occurrence law of different types of typical obstructive flow patterns,and predict the typical obstructive flow patterns that may occur in the future based on the known objective conditions,this study focuses on summarizing and sorting out the characteristics of different flow patterns and carries out the research on the prediction method for the surface flow pattern characteristics of sharp bends channels to address the safety risks of navigation caused by obstructive flow patterns.The prediction method based on CNN-GRU-BP combined neural network is used,and compared with the measured values and traditional numerical methods.The results show that the combined network prediction method has good prediction performance for the surface flow characteristics of the channel,and it can obtain the characteristic values of different flow patterns in real time,accurately,and purposely under the input nonlinear influencing factors,which can provide data support for the navigation safety control of the actual channel,and technical support for the prevention and resolution of related navigation safety risks.