Verification of time delayed neural network for flood flow simulation in channels and flood plains | |
| Author | Nandar Kyaw |
| Call Number | AIT Thesis no.WM-02-05 |
| Subject(s) | Flood forecasting--Simulation methods Floodplains |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering |
| Publisher | Asian Institute of Technology |
| Abstract | River flood forecasting is one of the most important components in many flood control systems. Flood forecasting models are becoming very essential tools to predict future flood events early enough in order to take appropriate control action to minimize damage. There are many models used to forecast runoff from rainfall. Many of them require large amount of input data and their computations are complicated. With the development of computer science and modern technology, models are fast becoming attractive alternatives to provide accurate estimates of future water conditions effectively. One interested area in this emerging field of hydrological research is the application of Neural Networks. Conventional neural network models when applied to flood routing in rivers with large time lag or damping effect may not produce proper results on simulation. A Time Delayed Neural Network (TDNN) model was applied in this study and its result was compared with a finite difference model of unsteady flow in open channels and the Artificial Neural Network Model (ANN). The effect of difference in time delay or lag of the model results was analyzed. First, the simulation was carried out based on a hypothetical floodplain named ABCD floodplain by Wang (2002) in this study. The two-dimensional MIKE 21 model is employed to compute the water levels and velocities of the ABCD floodplain, and then the MIKE 21 model result is used as a reference for comparison with the ANN and TDNN models. By using the same boundary conditions, the ANN and TDNN models are calibrated and verified using the computed water levels and velocities from MIKE21 at various stations inside the floodplain. Both of ANN and TDNN models are found to reproduce the results of MIKE 21 model very closely with very satisfactory performance. In the second stage of this study, the TDNN model is applied to a real case, Klong Dam Irrigation Project Area located in southeast of Bangkok. Hydrological input data are the daily flood levels, rainfall, and pumping discharges in year 1995, 1996, and 2001. Water levels during selected flood periods (August to November) at five water gates inside the project area are simulated and predicted by using TDNN model. Very satisfactory results of simulation and forecasting based on past records are obtained in the Klong Dam floodplain area as ANN. Model sensitivity of TDNN is analyzed to detect which parameters are critical to TDNN model performance and what kind of effects they impose to the model. Recommendations for this research work are given based on the results from the two case studies, mainly; the ABCD hypothetical floodplain and the Klong Dam Irrigation Project area. Compared to the hypothetical case study, it is more difficult to simulate a real case due to a more complicated flow in flood plains and in irrigation canals as well as through pumping stations. Since time delay effect of rainfall to flow in floodplain and irrigation canals, a moving average technique is employed to pre-process the input rainfall data so that the optimal model output can be achieved. In addition, more investigation and data analysis are recommended for improving TDNN model accuracy. |
| Year | 2003 |
| Type | Thesis |
| School | School of Engineering and Technology (SET) |
| Department | Department of Civil and Infrastucture Engineering (DCIE) |
| Academic Program/FoS | Water Engineering and Management (WM) |
| Chairperson(s) | Tawatchai Tingsanchali; |
| Examination Committee(s) | Luketina, David Andrew ;Clemente, Roberto ; |
| Scholarship Donor(s) | Asian Institute of Technology |
| Degree | Thesis (M.Eng.) - Asian Institute of Technology, 2003 |