Hydrological Long Term Prediction And Analysis In Qingjiang Basin

Exact and timely prediction of water resource system, which is a large complex system, not only can make significant academic theory, but also can make important practical value. Among prediction of water resource system, streamflow prediction is a substantial part. Since streamflow in one basin is influenced by climate characteristic, physical geography characteristic and human beings activities etc., its characteristic is very complicated, showing randomicity, grey, nonlinear, and so on. So, prediction of streamflow, especially long term prediction, is very difficult, and always be a hotspot or a nodus in the prediction field.Based on the streamflow data and precipitation data from Qingjiang basin, prediction of annual streamflow, lunar streamflow and daily streamflow in a week is developed:In the prediction and analysis of annual streamflow part, first, two traditional models, auto-regression model and least-squares linear regression model, are analyzed and applied. Then, four new prediction models, partial least-squares linear regression model, dynamic grey model, nearest neighbor bootstrapping regressive model, and artificial neural network model are also tried. Through analysis, it is found that among these prediction models, nearest neighbor bootstrapping regressive model can give a best prediction result of Qingjiang basin.In the prediction and analysis of lunar streamflow part, seasonal auto-regression model, seasonal least-squares linear regression model, seasonal threshold regression model, recession model, seasonal nearest neighbor bootstrapping regressive model and seasonal artificial neural network model are suggested. As a whole, the prediction result from seasonal threshold regression model and seasonal artificial neural network model is better than other models.Prediction of daily streamflow in a week is the last part, also is the most difficult part of this paper. Threshold regression model and multivariate nearest neighbor bootstrapping regressive model are used. At the same while, three new hybrid models are put forward, one is combined by nearest bootstrapping regressive model and nearest bootstrapping disaggregation model, one is combined by threshold regression model and nearest bootstrapping disaggregation model, the other one is called threshold artificial neural networks model.Through prediction and analysis, both traditional mature prediction methods and recent new prediction technology are taken into account in this paper. Thus, not only makes a deep analysis to streamflow prediction in Qingjiang basin, but also helps to afford new use of prediction models and presents some new prediction thoughts.