Research on the Forecast of Runoff by Chaotic Neural Networks
Tang Tian-guo · China Rural Water and Hydropower · 2009
Runoff is often affected by various factors and its character is complex and multivariate.Evolution regularity of runoff cannot be found only by using hydrographical data from observing stations.Based on chaotic theories,phase space reconstruction had been completed by mean runoff from the Cuntan Station in the Three Gorges.Saturated association dimension is obtained.Learning and teaching values of Neural Networks can be achieved from multi-dimensional phase space of chaotic theories.Buildup of networks is built according to saturated association dimensions.So,the chaotic network model has been finished for runoff forecast.The results show that reasonable evolution regularity has been found from hydrographical data from the Cuntan Station by using the chaotic network method.These conclusions may serve as a reference to exploitation and utilization of hydrographical resources in the Three Gorges.