Application Of BP Neural Network Based On Swarm Intelligence Methods For Prediction Of The Stage And Thickness For Ice Jam

Ice is a common phenomenon in the rivers of cold regions. Rivers in these regions often form the ice cover, the ice jam or the ice dam in winter, and they will bring various ice disasters. The variety of the water level during the period the ice jam is one of the phenomena that forming mostly in the ice jam segment. When analysis the field data from Hequ region to the Yellow River, we used four methods to building the model of stage and thickness for ice jam in the Hequ region to the Yellow River. They are the traditional BP artificial neural network, particle swarm optimization, ant colony algorithm optimize neural network and multiple regression analysis. While in the control of experimental conditions, we also analysis the measured data from Ice jam test and Backwater test by the same four methods were used to set up stage and thickness for ice jam prediction model, predictive value will be obtained and measured values within the scope of the information contrast. The results can be seen by comparison, both in natural rivers and in the laboratory under controlled conditions, traditional BP neural network has been enhanced in the prediction accuracy contrasting to multiple regression analysis. On this basis, particle swarm optimization and ant colony algorithm optimize neural network have more advantage both in the prediction accuracy and complex environmental adaptability. This study provide useful and important help reference for the frozen rivers of ice forecasting model established.