Mid-Long Term Runoff Forecast Model Of Zhaopingtai Reservoir

The forecasting of monthly runoff into reservoir is an important reference for the management and regulation of reservoir. Researches on these aspects would be of great benefit to the security of the reservoir and downstream area, the rational use of water resource and the utilization of reservoir.Zhaopingtai Reservoir lies in the Shahe River in the upstream of Huaihe River Basin. Inflow is usually less in the low flow periods. In the reservoir regulation, more outflow used for irrigation and hydropower plant in the prophase of the low flow period often results in less water storage and low water level, which directly reduces the benefit of the reservoir. To solve this problem, forecasting models of monthly inflow to the Zhaopingtai Reservoir in the low flow period from September to March are developed using artificial neural network (ANN).Correlation analysis method is used for the monthly inflow series of Zhaopingtai Reservoir to find main influencing factors of the inflow in months of low flow period. Forecasting models based on multivariate regression analysis were established to predict monthly inflow in low water period. In the ANN forecasting models of monthly inflow, the standardized inflow of the influencing months is taken as the ANN input, and the inflow to be predicted as the ANN output. The activation function of the hidden layer and output layer use tansig and pureline functions, respectively, and the train algorithm is Levenberg-Marquartdt. Other ANN parameters are determined with trial-and-error method. The qualified rate of forecasting is used for model evaluation based on the Standard for hydrological information and hydrological forecasying (SL250-2000). The results show that the precision of forecasting models for Jan., Feb., Sep., Oct. and Dec. are all in the first class, that for Nov. and Mar. are in the second class and third class, respectively. To increase the forecasting precision, monthly precipitation is added as another ANN input. The results indicate that the forecasting precision are all improved, with the...