8.2 Rainfall-runoff simulation through SAWANN

Mapping ability of ANN model can be further improved by hybrid waveletfunction.Although back propagation WANN has the characteristic of high convergence speed，it still cannot avoid the premature convergence and may reachlocal minimum.The Simulated Annealing (SA) algorithm is a global optimization approach and has good performance in searching globalminimum.However，weight optimization by using simulated annealing algorithm may result in low convergence speed.Hence，by creating a hybrid of BPgradient searching algorithm and simulated annealing algorithm，both advantages of convergence speed and local minimum avoidance can be achieved.Inthis section，the SAWANN (simulated annealing wavelet back propagationartificial neural network) in this section was tested and compared with WANNmode ......     （共5088字）    [阅读本文]>>