Research Of The Ice Condition Forecasting Based On The Artificial Neural Network

River ice condition forecasting is an important problem that is always concerned by people. Because there is no distinct function relation between factors and ice condition, lots of problems have not been solved completely, especially in the forecasting of break up, freeze up, ice jam and ice dam. Good nonlinear mapping power gets artificial neural network the advantage to solve the ice condition forecasting problem.With the analysis of the appear, disappear and evolvement of the river ice condition, sum up six aspects of the factors that affect the river ice condition, consist of heat factor, drive factor, river channel configuration, people action, effect of up and down stations and direct runoff. Except the river channel configuration is settled in a short period, any big change of one of the factors could be the dominate factor of the ice condition's development.Two break up date forecasting mathematical models of Ha'er bin Station on Song Hua River were founded, based on the mechanism of the break up and the ice condition data of the station. The first model is a within prospect of and forward looking production, because its data of factors is close to the break up date. But its precision is restricted by the precision of air and hydrology forecasting. The second model is easily to handle and isn't restricted by the precision of air and hydrology forecasting, but its precision is influence by considering the data is far with the break up date. The two models can be used together in actual use. Considering the actual practice of the first model, the applied mode of the first mode was founded with the iterative algorithm. It is proved that the method is doable. The artificial neural network of this paper is BP network with one hidden layer, and the training method is Levenberg-Marquardt algorithm.The program of ice condition forecasting system which is produce by Visual Basic and Matlab can be used as the platform for training the model of the artificial neural network and the forecasting of the ice condition. The program is designed to be the form of menu and command button, and you can use it without the knowledge of artificial neural network.