Further Studies On Methods Of Hydrograph Forecasting Of Sediment Concentration For Sediment-Laden River

In order to alleviate silt of sediment in the downstream and heavy flood with small discharge, there are a series of methods to be used, including warping of the Xiaobei Mainstream of the Yellow River and water-sand joint operate of Xiaolangdi Reservoir. Those methods to be used will be work well in conducting with sediment concentration forecasting. This thesis improves the existed methods of sediment forecasting and explores new method based on previous studies. The main content and research results are as follows:1. Multivariate regression equation is often used to forecasting. Selecting impact-factor correctly is one of the keys to make it effective. The existed model of sediment forecasting only considers flow discharge and sediment concentration. Sand composition is also one of the important factors to affect sediment concentration of flood. The thesis built multivariate regression model with considering sand composition effect to sediment concentration. The result indicates that the forecasting performance of this model is bad to the model without considering sand composition. These could be that current measure about sediment composition unmeet the forecast, therefore causes most estimates of the sediment composition were used with the model.2. System response function model is one of the conventional forecasting models, which most be used to forecast runoff based on rainfall. The thesis built multi-input and single output system response function model based on sediment transport character of "the more will be transported if more sediment will come" and the consideration of that the character of flood wave similar to convergence of flow on a basin. For the model disregard details and grasp the sediment transport with flood from macroscopic view, so the model has a good forecast performance. In the condition of current hydrometrics about sediment and knowledge about sediment transport in river, system response function model has a certain practicality and reliability.3. Taking into account the continuity development of a thing, self-memorization principle is a universal principle. The thesis first used this principle to forecast sediment concentration. The dynamic function in the self-memorization method is built according to sediment discharge character. Forecasts indicated that the model was not good as the response model, possible reason is that dynamic function cannot effectively reflect the law of sediment discharge transport. Self-memorization model can overcome hysteretic error the system response function model has, so it is worthy to do further study.4. Artificial Neural Networks is often used to forecasting in different field for its nonlinear mapping ability. In order to explore the ability of forecasting sediment concentration in single channel and the effect of different input to output, the thesis built Artificial Neural Model to forecast sediment concentration, which inputs were made in two case, one was the inputs from four factors, the other was from two. Forecasts indicated that if input factors are on dependence each other, the contribution of each input might be counter balanced by harmful actions among them, therefore influences the performance of the model.5. Comparing the four above mentioned models in terms of performance, we can make a order for them: the best one is the system response function model, the second is Artificial Neural Networks of two input, the third is self-memorization model and the forth is multivariate regression equation model.