Application of neural network model for daily flood forecasting of inflow and release of Sirikit reservoir and downstream flood discharges | |
| Author | Bordin Khanthaprathep |
| Call Number | AIT Thesis no.WM-02-12 |
| Subject(s) | Neural networks (Computer science) Flood forecasting--Thailand |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of the Master of Engineering |
| Publisher | Asian Institute of Technology |
| Abstract | At present, real time operation of Sirikit reservoir in Nan River during flood periods frequently causes severe flooding in downstream areas especially in the vicinity of Uttaradit and Pitsanulok provinces. The excessive release of Sirikit reservoir is due to lacking of information of local inflows from downstream tributaries of Nan river. To forecast flood for real time daily operation and release of the Sirikit reservoir, a flood forecasting model is employed for forecasting daily inflow and release of Sirikit reservoir and daily flood inflows along Nan river downstream of the reservoir. Previous inflow and release of the Sirikit reservoir and flood discharges during large flood years are used in training and testing the model. In this study using Artificial Neural Networks (ANNs, WinNN0.97) is used for forecasting one, three and five days ahead daily inflow of Sirikit reservoir and daily flood discharges along Nan river downstream of Sirikit reservoir at stations N12A(Tha Pla), N27 A (Naresuan dam) and N5A(Phitsanulok). In this study use Back Propagation algorithm and sigmoid transfer function for ANN model. Main streamflow gaging stations and rainfall stations and their observed data were chosen as input data into model. The correlation coefficients between main rainfall stations and streamflow gaging station were determined. The input rainfall and streamflow data for ANN (WinNN0.97) were selected based on highest values of auto and cross-correlation coefficients for training. The parameters of model obtained from training are used in model testing. The accuracy of model is measured by percent of good patterns, the minimum values of Root Mean Square Error (RMSE), Efficiency Index (EI) and also comparison of hydrograph between observed and simulated discharge by ANN model. The models with best performance is selected for sensitivity analysis of model parameters. Each of model is trained and tested for 1 day, 3 days and 5 days ahead flood forecast. The one-day ahead flood forecast has 5 data sets such 1-1, 1-2, 1-3, 1-4 and 1-5. The Three days ahead flood forecast has 3 data sets such 3-1, 3-2 and 3-3. The five days ahead flood forecast have 3 data sets· such 5-1, 5-2 and ·5-3. In this study, there is only one hidden layer and numbers of node in the hidden layer are varied from 1 to 1+2 (I = Number of Input Data). The total cases of one day, three and five days ahead flood forecast in this study are 247 cases. The models with best performance are selected for real time flood forecast. The results show that one day ahead flood forecast are of the best accuracy. The performance of three days ahead flood forecast is better than five days ahead flood forecast. When used ANN model (WinNN0.97) in the real time flood forecasting, the results still good and can be improved by applying a correction factor. Time Delayed Neural Network (TDNN) is used to explore the possibility in flood forecasting at all above-mentioned stations. The accuracy of TDNN models are better than ANN models for the best performance of 1, 3 and 5 days ahead flood forecast. |
| 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 S. ;Babel, M. S. |
| Scholarship Donor(s) | ASEAN Foundation Scholarship |
| Degree | Thesis (M.Eng.) - Asian Institute of Technology, 2003 |