Water level forecasting model using improved artificial neural network architecture
Reliable water level forecasting can help achieve efficient and optimum use of water resources and minimize flooding damage. Currently, artificial neural network (ANN) has been successfully tested in many forecasting studies, including river flow. However, the accuracy and reliability of the river f...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Language: | English |
| Published: |
2012
|
| Subjects: | |
| Online Access: | http://umpir.ump.edu.my/id/eprint/13483/1/MUHAMMAD%20%40%20S%20A%20KHUSHREN%20SULAIMAN.pdf |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Reliable water level forecasting can help achieve efficient and optimum use of water resources and minimize flooding damage. Currently, artificial neural network (ANN) has been successfully tested in many forecasting studies, including river flow. However, the accuracy and reliability of the river flow forecasting in such application requires continuous research. In this context, this study aims at developing a high accuracy and reliable forecasting model using the ANN to predict high water level events. The aim is achieved by introducing four new approaches in the ANN modeling. Firstly, the internal architecture of ANN is enhanced by utilization of optimal steepness coefficient (OSC) in sigmoid function. Secondly, the external architecture of ANN is improved by applying zoning matching approach (ZMA) where data training selected is based on the target water level to be forecasted. Thirdly, model evaluation of forecasting results is improved by engineering approach where allowable offset errors are used to demonstrate the accuracy of forecasting results. Lastly, the correct prediction of up/down of water level is a new evaluation model to evaluate forecasting capability and reliability of the ANN model. A case study has been applied at the Rantau Panjang station at Johor River where hourly water level data dated from 1963 to 2008 have been examined to forecast daily and several hourly intervals lead-times. The result showed that the use of the OSC and ZMA techniques had not only improved the accuracy of ANN models compared to standard approach (SA) but also achieved high accuracy forecasting results. The allowable offset errors and up/down prediction measures have enriched measures in evaluating forecasting model's accuracy and reliability. The impact of the study is that the techniques can be the basis of future development approach in ANN based data forecasting model specially in monitoring flood events. |
|---|