Forecasting Reservoir Discharge Using ANN Model & Its Software System

The results of mid-long term flow forecasting are required for effective hydropower reservoir management and scheduling. The results can be widely used in many fields such as environment protecting, flood control, drought protecting, operation of reservoir, running of hydropower station, ship management and water resource distribution and show huge economic value in operation of reservoir and hydropower station. But the precision of these results is not satisfactory to us because of the complexity of intrinsic mechanism of raining and flow, the influence of human activity to the river basin and the long period of anticipation. Artificial Neural Network is the most useful algorithm currently, which has already been applied in forecast problems of water resource system such as raining forecast, flow forecast, water quality forecast, water demand forecast and load forecast and provides a effective approach to improve accuracy of mid-long term flow forecasting. The focus of the paper combined with the advanced water resource dispatch project of Yunnan Power Group is as follows:1 , An artificial neural network (ANN) model was developed to forecast river flow in the Manwan Reservoir. Results from neural network forecasting are compared with those obtained from traditional time-series forecasting and show that the ANN model is a better approach in hydrology forecast.2, Base on numeric experimentation, the structure of ANN which adapts to flow of Manwan Reservoir is found. This ANN model improves the accuracy of forecasting.3, There are many training algorithms when we are try to train an ANN model. Every one has its advantage and disadvantage. In this paper, the best algorithm is found through comparing with various training algorithms.4, Combining with the advanced water resource dispatch project of Yunnan Power Group, the ANN model is successfully applied to mid-long term forecast system of water resource.