Research Of Bayesian Probabilistic Forecasting About River Flood Based On BP ANN

Research on the river channel flood forecasting is an important part of hydrologic forecasting .The calculating results are also very important for the reduction of disasters and the synthesis use of water resource. So based on the traditional Muskingum method ,the back-propagation artificial neural network (BP ANN) is used to forecast the river channel flood ,because the BP ANN has good nonlinear characteristic.Considering the uncertainty of the hydrologic forecasting,the Bayesian probabilistic forecasting and the BP ANN were combined for forecasting the river channel flood more scientificly.By this way,the BP model is proposed with the prior distribution and the likelihood function of the discharge to be forecasted .The research is made on the flood forecasting between the Shuibuya and Yuxiakou on the Qingjiang River ,and the research results are analysed .Meanwhile the results show that probabilistic forecasting is feasible and the data information got is very helpful for the decision-making aiming at defending flood.The major contents of this paper are as follows:(1) The development of the flood propagation theory,the real-time correction methods of flood forecasting and hydrologic probabilistic forecasting are reviewed.And it is pointed out that research on the river channel flood forecasting has a pretty strong academic meaning.(2) The theory of Muskingum method used broadly for the flood forecasting is expatiated.The concents contain the nonlinear results of the Muskingum method and the Muskingum method especially for the river ,where there are anabranches.(3) The paper introduces the BP ANN ,which is used comprehensively in the world,so the BP ANN can correct the discharge results of Muskingum method forecasting and the corrected results can also be used as the basis for the Bayesian forecasting system (BFS).(4) The Bayesian probabilistic forecasting is introduced as one of the probabilistic hydrologic forecasting frameworks.The Bayesian forecasting system (BFS) was first introduced,and how to get the system model assumed as linear-normal model was discussed.This paper shows the posterior distribution of the discharge to be forecasted by the linear-normal model. (5) A new prior density and likelihood function model is developed with BP ANN to study the river channel between the Shuibuya and Yuxiakou on the Qingjiang River .First,the discharge results is calculated by Muskingum method considering the anabranch Zhaolaihe.Then BP ANN is used to update the discharge results by correcting error.And the discharge results calculated are compared and analyzed.In the end, the probability density of discharge to be forecasted is showed by the Bayesian probabilistic forecasting model.So the destination of this paper comes true by realizing the hydrologic probability forecasts.Different from the deterministic hydrologic forecasting ,Bayesian probabilistic forecasting can provide the density distribution for the users,which contain more useful information,very helpful for the decision-making.