Comparative Study of Some Improved ANN-Models for Hydrologic Time Series Forecast

In this paper, by taking two real hydrologic series for examples, the performances of some improved ANN-models (including BPNN, gda-BPNN, gdm-BPNN, gdx-BPNN, LM-BPNN, WANN and WNN) have been discussed and compared. Analysis results show that: 1) Among all BPNN-models, gdx-BP, gda-BP, gdm-BP and LM-BP are better than standard BPNN, and gda-BP and gdm-BP are the most satisfying; 2) Compared with BPNN-models, WANN and WNN models are much better especially when hydrologic series are very complex and more long forecast periods are needed, since they can take great advantages of multi-resolution analysis ability of WA; and 3) WANN model has many advantages, such as can understand internal structure and characters of series meanwhile, high forecast accuracy, high eligible rate, can overcome the contradiction between high accuracy and long forecast period, etc. Although WANN and WNN models can meet almost all practical needs, some key and difficult problems about them should be solved furthermore.