Study On ANN Model Application In Middle And Lower Reaches Of Yangtze River

Research on the stream flood forecasting is an important part of hydrological forecasting. This paper is based on the Sub-Project of the National Science Fund Project that is to research on flood characteristics and reduction of disasters, study on the new way to improve forecasting period and precision. Research on the method of flood forecasting in mid-down stream is the study direction of this paper. Based on the hydrological characteristics of mid-down stream, using a Artificial Nerve Network way, a new way of flood forecasting is established to make the stream flood. The stream from Luoshan to Hankou is selected as sample, this model is to predict flood course in Hankou. The most research contents is as below:(1)This paper expatiated the flood disasters and the ways of reducing it, and analyzed some methods of the stream flood forecasting simply. The ANN model and its application status are introduced; some kinds of ANN model, its advantage and disadvantage are compared and analyzed.(2) The stream flood transmition is incertitude, the Artifical Nerve Network can describe the complexity course of input-output patter for its enough non-linearity. This paper emphasis intruduced back-propagation(BP) ANN model and its advanced. Based on the flood of Yangtze River at Luoshan and Hankou station, by using a self-adapt back-propagation(BP) algorithm, this paper established a BP nerve network model about riverway flood forecasting of Hankou considering the interzone inflow. The predicting results indicate that this model can preferably reflect the influences of interzone inflow to the discharge process of Hankou and the precision is acceptable under predicting period improved.(3) While ANN model using in stream flood forecasting, In allusion to the problem that the simulation precision is not very good because ofthe lack of high peak samples, this paper began with the method of network planning and set up a BPPR model, based on the self-adapt BP model and importing peak identified theory which is adding error correct coefficient to network error of peak samples. Comparing BP model with BPPR model, through real calculation, we find BPPR model improved precision of peak forecasting, as well as the total simulation precision.