River Ice Forecasting Based on Genetic Neural Network

Based on the analysis of the factors caused ice flood, the paper selected appropriate forecasting factors, established neural network model of ice forecasting combining genetic algorithm (GA) with Levenberg-Marquardt BP(LMBP) neural network. The GA-LMBP algorithm is to train globally using genetic learning algorithm firstly, then train accurately using LMBP algorithm, overcoming the defects of traditional BP algorithm such as slow convergent rate and local minimum. It has achieved good results through applying the model to forecast break-up date in Yilan and Jiamusi sections of Songhua River.