Hydrological Forecasting Method Research Based On Neural Network

The flood disaster in China happens very frequently, which not only have an impacton the country’s social development, but also threat the people’s life. In order to reduce theloss brought by the flood, lots of measures have been used, such as the flood forecastingmodel. For the flood control, the accuracy of flood forecasting is one of the mostimportant factors for flood prevention and disaster reduction.Neural network is the most widely used method in hydrological forecast, and the typeof forecast factor to establish the neural network input influences the accuracy of thehydrological forecast.In this paper, the copula entropy theory was putted into partial mutual information(PMI) method. The PMI base on the entropy of the Copula was used to the selection of thepredictor, which calculates the mutual information value through the copula entropy. Inthis paper, the BP neural network, the RBF neural network and the GRNN neural networkwere used to establish the hydrological forecasting model, the copula entropy method andcorrelation coefficient method were used to select the predictor of the hydrological model,to make the hydrological forecasting for the Yichang hydrological station in Yangtze Riverbasin and the Pingshan hydrological station in the Jinsha River Basin. Test resultsindicated that in the same neural network of hydrological forecast model, the result basedon the entropy of copula are better than the result based on the method of correlationfactor, in the same forecast factor conditions, the result based on BP network are betterthan the forecast result based on RBF network and GRNN network. This paper improvedthe accuracy of the hydrological forecast through the comparative analysis based on thedifferent forecast factor and the hydrological forecast model for a particular basin, and hadvery important practical significance on the flood resources and flood prevention anddisaster reduction.