Research Of Flood Forecasting Model For The Dahuofang Reservoir Based On BP Neural Network

Dahuofang Reservoir, the controlling key project on the upper reaches of Hunhe River, has great significance for flood control. The current flood forecasting model of Dahuofang Reservoir is Dahuofang Model (DHF).This model is a kind of Lumped hydrological model which does not consider hydrological processes, therefore it leads to the problem that the flow rate of tributaries and water levels cannot be effectively described. And the optimum seeking method or the manually try the errors method are used into the selection of DHF’s parameters, which also require real-time correction. So it is a demanding work for the workers and the work of parameter calibration is more complicated. In view of the above shortcomings, this paper takes a full consideration of objective factors such as data limitation, data input, boundary conditions and watershed characteristics and established a semi-distributed BP neural network model for flood forecasting. After observed in practice, it is easy to fall into local minimum point and the duration of forecasting is short.so then some improvements have been made on this basis and established the DHF-GA-BP neural network coupled flood forecasting model. This model is applied to upstream basin of Dahuofang dam site for flood forecasting. The improved model combines the advantages of Dahuofang model and BP neural networks, also uses the Genetic Algorithms, and offsets the deficiency of the semi-distributed BP neural network model. The main content and the corresponding results of this paper are as follows:(1) Dividing the basin of above dam before building model. When dividing the basin, each hydrological station’s nearest natural river basin watershed is used to be boundary line, using DEM data and the ArcGIS divide Ⅰ,Ⅱ,Ⅲ,three areas eventually, where the Beikouqian hydrological stations exits is area Ⅰ, where the Zhanbei hydrological station exists is area Ⅱ, where the NanZhangdang hydrological station exists is area Ⅲ.(2)Pre-process existing data which will be input into the model. Using the stepwise analysis method to select the initial data to ensure that the input data has a significant impact on output. The result of this selection is as below:both area Ⅰ, Ⅱ are eight significant factors, area Ⅲ is six significant factors, the reservoir section, namely, the whole basin is 11 significant factors.(3)Establishment of semi-distributed BP neural network model for Dahuofang Reservoir real-time flood forecasting, the prediction model is divided into two parts, the first part is of sub-basin flow forecasting, the second part is the storage section flow forecasting, the output of the first part is the input of the second part. It emerged that the results of sub-basin forecasting and the storage section flow forecasting are preferably and the model also plays a supporting role in the flood regulation of sub-basin, but the prediction accuracy, efficiency, and the duration of prediction need to be improved. So the model still needs to be perfected.(4)The improved model is established after the application of genetic algorithm calibration of Dahuofang model parameters. That calibrate the parameters of Dahuofang model by using genetic algorithms can make it apply separately to each sub-basin flood forecasting. The forecast results of inspection period show that the prediction is better, although some unqualified phenomenon appeared in some sessions, the overall standard can meet the request of formal forecasting.(5)The defects of the semi distributed BP neural network model are improved. Application of genetic algorithms to optimize the initial weights and threshold value of the original BP neural network, combined with the results of using genetic algorithms to calibrate parameters of Dahuofang model, to establish an improved semi-distributed BP neural network model --DHF-GA-BP neural network coupled model. Forecast results show that the flood forecasting model has been improved in the efficiency and accuracy than before, extended the forecast duration. And to a certain extent, avoid the forecasting errors accumulation of Dahuofang Model which was used the Genetic Algorithms to calibrate the parameters.