Advanced hydraulics simulation using artificial neural networks - Integration with numerical model

Due to these uncertainties (data uncertainty, parameter uncertunty, and model uncertainty) associate with the modeling processes, it turns to more difficult to accomplish a well-validated model through a conventional approach. Particularly, the models of hydraulic relevant systems with hundreds of thousands of nodes have become increasingly common to undertake more realistic simulation. These models create heavy demands on computing time and computer capacity. As a result, a significant delay usually exists between the required real response-time and the actual computational-time. The management real response-time is an essential factor to justify, a timely decision-making. For example, dealing with an input of a rapid change condition of a hydrologic avow such as a flash flood into a simulation system has become a challenge task. In this paper, ANNs is used to integrate with numerical model for two ways. For the neuro-numetical approach, ANNs is used to extract the existing patterns from a well-validated numerical model and to generate an "input-outpit" simulator. Two demonstration examples - hydrodynamic simulation for a tidal wetland and sediment transport simulation, are presented.