A Grey Correlative Analysis and Pattern Recognition Applied to Mid-long Term Runoff Forecasting

This paper addresses the problem of mid-long term runoff forecasting. An extended dimension approach from one dimension to multiple dimensions in hydrological time series was developed in terms of the idea of pooling hydrological information from different sources whereby historical data were used to form a family of runoff pattern and a set of corresponding forecasting vector. A workable and practical method called as the grey correlative pattern recognition was developed to determine a most similar hydrological pattern and forecast runoff in the leading time. Eight catchments under different climate conditions in the basins of the Yellow River and the Yangtze River were selected for verifying annual and monthly runoff forecasting. It was found that the proposed method yields a better runoff forecasting than the AR (p) model.