Study On Floods Numerical Simulation And Intelligent Prediction For Songhua River

Numerical simulation of river flow can provide reliable and accurate data base for flood analysis and prediction. For example, it can be used to calculate the corresponding surface curves of design flows in the planning and design of flood control project; and in real-time flood control, it can be used to calculate the corresponding surface curve according to forecasted river flow, which can provide important support for flood control decision-making. However, numerical simulation model is only to analyze the current hydrological situation, but is difficult to predict future floods. Therefore, it is meaningful to establish a water information system which combining the hydrodynamic model and intelligence forecasting model. Therein, the hydrodynamic model is used for simulating historical flood; the intelligence forecasting model is used for making flood forecasting.In the past few decades, various numerical simulation methods have presented. However, the applications of these methods are constrained for the low simulation efficiency and heavy data preparation. In the hydrological forecasting, many prediction methods also have been developed. Among them, the artificial neural network method has been widely used in river flood forecasting because of its strong nonlinear mapping ability. Based on the above, with Songhua River net as the engineering background, this article firstly built an one-dimension hydrodynamic model with a semi-implicit finite element scheme; then, a flood intelligence forecasting model were built with two kinds of modified BP algorithms. The main research results are summarized as follows:(1) Songhua River Basin was briefly introduced, including meteorology, hydrology and historical floods. Then, existing data were analyzed, and the study area, time and typical years were identified.(2) Numerical simulation of river flow generally exsits these problems:low efficiency, cumbersome modification, and poor coupling. In order to improve these defects, a semi-implicit elementary scheme was introduced. Combined with historical data of Songhua river, an one-dimensional hydrodynamic model was established. The simulation results showed that the model not only had strong computational stability and fast convergence speed, and it was easy to prepare data.(3) BP algorithm has the disadvantage of slow convergence and falling into local minimum. In order to overcome these, two kinds of modified BP algorithms, Momentum-Adaptive-Learning-Rate algorithm and Levenberg-Marquart (L-M) algorithm, were applied to forecast the flow of Songhua River net. The results showed that both of modified BP algorithms could improve prediction accuracy and convergence speed, and the convergence speed of L-M was faster than Momentum-Adaptive-Learning-Rate algorithm's, but the forecasting performance of Momentum-Adaptive-Learning-Rate algorithm was better in the historical flood of 1998.