| Real-time correction's kernel is that real-time correction system should sense reason and site that error takes place in time, and in that site modifies error in time. By this way conflict between forecasting precision and foreseeable period can be given attention to. So, many stations (water level station for example) should be needed in the predicting system, and can collect more state variables directly or indirectly. Those methods that make use of exit on-the-spot survey data can not be given attention to between forecasting precision and foreseeable period due to lake of information. Karman filtering's condition is very rigor, and needs special disposal so as to meet these conditions sometimes. Preferable effect can be achieved by proper simplification too in the predicting system; the rest real-time correction methods should be applied to the small basins mostly, because they modify exit quality only. If they are applied to the large basin, foreseeable period may be shortened, and conflict between forecasting precision and foreseeable period can be given attention to.In this article the 1996-1999 total 20 floods' hydrologic datum of four years of Qingjiang region Enshi basin is used to proceed the network's training, forecasting and correction, and it is proved elementarily ANN(Artificial Neural Networks) has certain feasibility in the hydrologic real-time correction .In the study it is found that analytical approach (flood classifying) gets more effect than synthesis approach.In this article it is also discussed how the flood as training samples is selected. In the study author finds that whether training samples' selection is felicity has the most effect on whether the hydrologic real-time correction succeeds and networks is stable .And for this reason the principle of training samples' selection is discussed and some practical conclusions are deduced further. All these can be regarded as one new thought talked over with readers.
|
PDF Full Text Request