Application of Gray-self-memory-neural Network Model to Prediction of the Annual Runoff

摘要

Runoff time series is a non-linear, weakly dependent and complicated dynamic system. The key of improving the accuracy of runoff prediction is to dig the information in the limited sample sufficiently. Gray system modeling uncovers the dynamic laws inside the system via processing gray information in order to transform the desultory data into ordered series for establishing model based on differential equations. Self-memory theory on base of physical motion irreversibility, emphasizes relation of system status, which urges upon evolutional rules of system itself, and then differential equations of dynamic system could be set up for the self-memory models. Combination of gray, self-memory could effectively responds ultra data, but with some phase lag. Neural network has advantage of paralleling distributed processing. On account of integrative prediction, three modeling are combined to forecast annual runoff. It is shown that gray self-memory neural network model has higher prediction accuracy and may be fit for annual runoff prediction.