An assessment of neural network models for predicting river flow and water level | |
| Author | Le Viet Son |
| Call Number | AIT Thesis no.WM-02-10 |
| Subject(s) | Neural networks (Computer science) Streamflow--Forecasting Water levels--Forecasting |
| Note | A thesis submitted in partial fulfillment of the requirements for the degree of Master of Engineering |
| Publisher | Asian Institute of Technology |
| Abstract | The present study is an assessment of Artificial Neural Networks (ANNs) for predicting river flow and water level. Two case studies are examined, the first is real-time flood forecasting for the Red River in Hanoi, Vietnam and the second is flow prediction for the Maipo river basin in Chile. ln the first case study, ANN models are developed using the back propagation algorithm to forecast water level at Hanoi station with a lead times of one day and two days. The study also developed three simple models for the same objectives. The comparison of the results of the ANN model with the other three simple models shows that, in general, the ANN model has substantially better prediction accuracy than these models for both one day and two days lead time. From the results obtained for the first case study, it is found that the ANN model appears to be an effective tool for daily real-time flood forecasting. The model performance for two days ahead forecasting significantly deteriorated compared to the one day ahead forecasting. The reason for this lies in the fact that the water level at Hanoi station in flood season changes much during one day. For the management and operation of the flood control system in the Red river basin, one day forecasting is considered long enough. The developed ANN models can thus be used as part of a flood warning decision support system In the second case study the snowmelt - runoff relationship is investigated to generate river flow from meteorological data using a physically based model, namely Snowmelt Runoff Model (SRM). Then the calibrated SRM was updated using the observed flow at previous days, now called the Real-time SRM, to forecast the flow one or two days ahead. Finally, an ANN model was developed for the same objective as the Real-time SRM. The models are tested by considering two sub-basins in the Maipo river basin, namely the Yeso and the Colorado. The SRM gives an acceptable result for both the Yeso and Colorado sub-basin for the Jong-term snowmelt runoff relationship (i.e. predicting te weekly averaged) behaviour over several seasons. For Real-time flow forecasting, a higher accuracy of prediction is required. The real-time SRM and the ANN model give similar results for real-time flow forecasting. The results in the Colorado are better than the results in the Yeso sub-basin. For a practical application of real-time river flow forecasting, only the results for the Colorado are recommended. |
| 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) | Luketina, David Andrew |
| Examination Committee(s) | Tawatchai Tingsanchali ;Babel, Mukand Singh ;Suphat Vongvisessomjai ;Mark, Ole |
| Scholarship Donor(s) | DANIDA |
| Degree | Thesis (M.Eng.) - Asian Institute of Technology, 2003 |