Print this pageA First Attempt to Predict Delta System Dynamic with Artificial Neural Networks
In a river-dominated delta system the sediment transfer occurs through main flow paths (active channels) which changes its directions along the time and deposits the sediment load in different regions of the delta (distribution of sediment volume). In natural systems this dynamic process is complex and requires long term monitoring, leading us to find alternatives tools of analyses. One of them is physical modeling, where it’s possible to reproduce, observe and measure such changes processes in detail. Also, artificial neural network (ANN) whose adjustment to new proposes of study is its main advantage. Based on that, this work aims to develop a simple ANN to predict the dynamic process of river-dominated delta using physical experimental data.
A controlled experiment of river-dominated delta was performed in laboratory resulting more than 18 hours of simulation. The morphological dynamic of the active channels was registered at every 3 min and its position on the system was registered (considering 5 different regions of 36° in a semicircular delta deposited).
Moreover, the deposited was estimated in these five regions as the river flow was considered homogeneous, steady and linearly distributed.
An adaptative and retropropagative ANN was created with 10 entries attributes: quantity of active channels and volume of distributed sediment for each radial region. From the total data, 64% was used to training the ANN and 36% was used to process the data (verification).
At each step of the verification, the prediction error was used to adjust the ANN parameters (adaptative step) improving its predictability. Despite the prediction of the active channels has presented considerable error, the prediction of sediment volume distribution resulted in a reasonable error (20-30%). Also, the same tests were processed not considering the adaptative step and the prediction error was much higher (40-50%).
The results obtained can be considerable good and shows a great potential of use of ANN tool to predict dynamic changes processes in river-dominated delta. Adding more data and parameters, e.g. the flow rate, sediment flux, basin level variation or topography, the ANN can improve its results even more.
Further, this new approach intends to predict the dynamics of river-dominated delta system and understanding the sand sediment transfer from continental margins to deep marine ambient and hydrocarbons reservoir-rocks.
AAPG Search and Discovery Article #90090©2009 AAPG Annual Convention and Exhibition, Denver, Colorado, June 7-10, 2009