A neural network model for hydrogram simulation in hydrographic basins.

The present work aims to provide the means of simulating hydrographs on the basis of previously existing data on rainfall and/or runoff On the basis of previously-recorded rainfall and run-off data, and using a model based on the MATLAB environment, artificial neural networks are used to simulate hydrographs for the hydrographic basin of the Corgo River in northeastern Portugal. A network architecture was conceived and constructed, along with a learning algorithm for the data collected from each event, indicating in each case if they were intended for training or testing, or neither of these purposes. Then, the order parameters, such as rainfall, water flows and number of divisions (the latter in accordance with the normalized axis of ordinates) were introduced Into the model. The model sought, for all events, the maximum rainfalls and flows, then calculated the divisions of the normalized axis, proceeding next to train the network, and finally perform the corresponding test. Having applied this method, we were able to conclude that there was an excellent degree of similarity between training and test results. Not only was there a minimal difference between peak flows in real and simulated hydrographs (as small as 23m(3)/s, 9m(3)/s and 1 m(3)/s) but, at 1 or 2 hours, the differences in the timing of these peaks were also minimal. Thus, the proposed neural network structure, as well as its learning algorithm, has proven to be extremely accurate in simulating, with a high degree of precision, the water levels observed in the hydrographic basin of the Corgo River.